1c recommendations for choosing a server. Solutions

Server for 1C

Large enterprises have been using 1C in client-server mode for a long time. And today this technology is actively moving into medium and small businesses. This article is about what a 1C server with a small number of users should be like.

From what number of users is a server needed for 1C?

Free PostgreSQL for 1C appeared a long time ago. And relatively recently such a wonderful position appeared in the 1C price list:

We confidently recommend using 1C in client-server mode starting with 3-5 users. The file option remains for very small databases and if there is no need to work in the database from home, business trips, or other offices, if left without 1C in the event of a failure for a couple of days, it’s not at all scary. We also consider working in RDP to be an outdated technology, which is only suitable when for some reason the platform is old (8.0 or 8.1) or the platform is ancient, it’s tempting to write “old Russian” (7.7). Therefore, everything that is written below applies to the option “a DBMS and a 1C:Enterprise application server are installed on the 1C server, work is carried out in a thin client 8.2.”

Should I buy a branded server or assemble it myself, or order assembly from suppliers?

If you have up to 10 users, you can use a regular “home” computer as a server with some modifications. It is quite possible to buy it “in parts” and assemble it yourself. If you have a good understanding of what thermal paste is, how to snap the ATX power connector without breaking the board in half, where is the cooler connector on it, how the performance of 1C depends on the memory frequency (almost linearly with sufficient resources) and why you shouldn’t plug in hard drives close to each other.

• more than 15 users
• there is no own super system administrator who knows everything “about computers”
• the business brings in enough money to regret it if 1C stops working

Small clarification. By “brand” we mean IBM, HP and similar brands. Any local “integrators” who assemble computers based on the Intel server platform “to order” with the help of yesterday’s students, with the configuration recorded from the client’s words, are not brands. Even if this server is assembled in a rack-mountable enclosure. Even if they put a nice label on the front panel. This is self-assembly, and we have many examples where, let’s say, assemblers make mistakes when selecting components. For example, we saw how in a serious server costing more than 300,000 rubles, the hardware RAID controller conflicted with the motherboard and produced 15% of the declared throughput. We also saw configurations with huge distortions, for example a machine with four Xeons and a single disk array of two disks. When purchasing a 1C server, contact those who understand what it should be like.

What is important is that today a branded server, guaranteed to be operational, proven and reliable, is worthcheaperself-assembly with comparable characteristics. Therefore, you most likely won’t be able to save money by buying a self-assembled mid-level server. Today, self-assembly has the right to life only in the segment of servers based on conventional (“desktop”) components, that is, at the very entry level, which are practically not produced by brands.

operating system

Without starting holy wars, we recommend using Windows Server 2012. This is a reliable, proven platform for the 1C server. Other solutions have a right to life, but, say, if you decide to install Ubuntu Server, you hardly need recommendations. And they are unlikely to help. Linux for 1C is unique every time and it is impossible to give general recommendations.

For entry-level servers, it is quite possible to use desktop Windows, for example, Windows 7/8. If you do not need Active Directory, RDS, and you will not use MS SQL Server as a DBMS. You just need to keep in mind that by default in Windows 7 Professional the number of simultaneous connections via TCP cannot exceed 20. In Windows 8 this limitation has been removed.

Memory

There should be enough memory. If you install more memory than needed, there will be no increase in performance. If you supply less than necessary, the user experience will turn into torture. The calculation is as follows: at least 2GB for the needs of the operating system, from 2GB for the DBMS, from 4GB for the 1C:Enterprise server. Total minimum 8GB. This volume is quite enough for 5-10 users with one database. The screenshot shows a typical picture of memory allocation during leisurely work of several users with one small database:

Notice how the 1C application server (rphost.exe) loves memory. Immediately after opening the database, he needs about a gigabyte. With intense work, for example, re-processing documents for a month, it is quite capable of mastering 6GB with a single active connection. You also need to keep in mind that the blue bar (“Waiting”) is not free memory, but the system cache. Here is the same system in which we began to work more or less intensively:

The 1C workflow has one feature. As a rule, it grows in one direction. The amount of memory captured by rphost.exe gradually grows and grows during the working day. We highly recommend restarting the 1C server agent every night, immediately before uploading backups.

16GB is enough for 20-30 users with two or three databases. It is imperative to configure the maximum allocated volume for the 1C:Enterprise server (this is done in the administration console) and for the DBMS! If this is not done, then, for example, Postgree out of the box will use only 200-300MB of memory. By default it is "smothered". But the 1C application server, on the contrary, can easily “eat” everything, no matter how much you give it.

Disk subsystem

Firstly, even on an entry-level server it is necessary to separate the disk with the system and the disk with the databases. Next, if finances allow, we add a third disk to store a large amount of data (backups, etc.). For Postgree, we must create a RAM disk to temporarily store statistics.

It is quite possible to use a disk for the system that is not the fastest and not the most voluminous. 500 GB is more than enough.

The disk for the database is definitely SSD. For 1C databases, conventional disks and their arrays do not have the slightest advantage. With a typical database size of an average enterprise of 2-3 GB, a volume of 120 GB, as you understand, is quite enough. And the performance, all other things being equal, differs by 10 times or more. Limitations on the number of rewrite cycles are a thing of the past; today SSDs are more reliable than any “mechanical” disk. You definitely need to take an SSD with TRIM technology (record cache when the power is turned off), and you need to carefully look at the declared write speed; there are many different options on the market now.

If database downtime, and especially restoring yesterday’s copy from the archive, is unacceptable, you need to install a RAID controller and two disks in “mirror” mode. We strongly advise against using software RAID controllers. Hardware only.

Archiving can be configured using a DBMS or you can write a batch file that will be launched by the Windows scheduler and download databases from 1C:Enterprise batch mode. It's not very important. It is important that every night the system automatically creates an archive and writes it to a separate server disk. An archiving system that requires manual activation is not an archiving system, but nonsense. After creating database downloads, we highly recommend uploading them to cloud storage. This could be Google Drive, Yandex Disk, Dropbox or your own ftp resource. The main thing is that these downloads are not stored in the same room where the server is located. Why? Because a normal business should calmly endure anything. Fire, burglary, visit from authorities with seizure. God forbid, of course, but anything can happen.

If everything that is possible has already been configured, and your hands are itching to do something else, you can redirect the Postgree log recording (if you have one) to a separate disk. This will give a small but quite noticeable increase in speed during active operations with the database.

Power supply and UPS

Don't skimp on the power supply. Never. You can save on everything else within reasonable limits, and the power supply for the server should be ideal. A server differs from a regular computer primarily in that it is always on. Double the watt reserve and a manufacturer with a name (Thermaltake, Powerman, Enermax), here is our recommendation. A UPS (Uninterruptible Power Supply) is needed for two purposes. Firstly, in the event of serious power surges, it must allow the server’s power supply to survive by taking the blow on itself (or better yet, on its fuse). Secondly, it must correctly shut down the server when there is a power outage. The UPS should not and cannot allow operation without external voltage, this is an illusion. Even 15 minutes. Its task is to give the server operating system a command to shut down. Therefore, an unconnected and unconfigured UPS is a piece of furniture, nothing more.

CPU

When 5 users work, any “desktop” processor of a fairly high class, for example a quad-core Core i7, will be loaded by 5-7% on average. The processor is generally not a bottleneck. It must meet the remaining parameters, nothing more. Therefore, it is better to choose the processor last, from those suitable for the selected system. Digital characteristics (number of cores, cache size, frequency) are not decisive. For example, the latest Core i5 outperforms the previous generation Core i7.

In any case, you should not buy the most powerful and expensive processor available for the selected platform (motherboard) for a 1C server. Rather, the optimal option would be from the middle of the list, sorted by price.

Software

There is no need to install anti-virus software on the server. It's too much. Safety and absence of viruses is ensured by three rules:

• The server should not contain any programs that are not necessary for its operation
• The server should not be used as one of the client computers
• From the outside, from the Internet and local network, no server resources should be accessible, except those absolutely necessary

We do not recommend combining 1C server and file, mail, proxy and web server. These tasks are perfectly solved by specialized devices and services. For example, the Zyxel Keenetic 4G router and others like it are excellent at distributing the Internet on the network, working as an ftp and file server. It is absolutely best to transfer the web server of a small and medium-sized company to a VDS in a data center. Mail there, or even to Google or Yandex mail for your domain.

A few words about virtual servers

Of course, this is a very popular technology for hi-end server solutions. However, for medium-sized enterprises (20-50 users), the benefits of virtualization are far from obvious, and for small companies they provide nothing but headaches. Firstly, server virtualization is not free, and it costs machine resources. Secondly, all the advantages of “on-the-fly performance tuning” and “mobility” are broken down by a bunch of problems with the loss of 1C software licenses in the event of any changes. Thirdly, there is basically nothing to share (there are few resources), and such an enterprise simply does not need several servers. Today, very few enterprises risk renting a virtual server in a data center under 1C, and even in this case it is better not to rent a server, but to connect to the 1C Online service.

How much does a server cost for 1C?

At the time of writing this article, the approximate cost of the server is:

• from 30,000 rubles for 5-10 users
• from 60,000 rubles for 15-20 users
• from 90,000 rubles for 30-50 users

With a larger number of users, it becomes necessary to distribute the DBMS server and application server to different physical machines.

At your request, we, together with our partners, can select and supply both a branded server manufactured by IBM and a prefabricated entry-level server.

The 1C:Enterprise platform versions 8.2 and 8.3 are considered a standard application for accounting and management tasks of companies. A wide range of application solutions has been developed for public and private enterprises. When implementing their own information infrastructure, every executive or IT manager of a company has a question about what kind of server is needed for 1C. The problem is complicated by the fact that purchasing equipment requires significant financial costs, and not every enterprise can afford to choose top-end configurations.

We have collected recommendations from leading equipment manufacturers (HP, Dell, IBM) and developers of the 1C 8.3 software product so that our clients can profitably purchase the server they need. The optimal network infrastructure can be obtained on the basis of any operating system, but hardware capabilities play a more important role in this.

Server selection criteria

The 1C platform may require significant hardware resources from the server. If the company's budget is unlimited, which is rarely the case, you can without hesitation take the latest generation platforms, fill all the disk drives and RAM slots and demand from the IT specialist the uninterrupted operation of the system. Selecting equipment with limited funds requires a more considered approach. To understand which 1C server will be able to cope with this, it is necessary to carefully analyze the structure of the computing loads. If they are known in advance, it will be much easier to design a ready-made solution.

When choosing a server for “1C” (8.2; ​​8.3), they are guided by the following points:

• the number of operators simultaneously performing data entry and generating reports;
• the ability to allocate separate physical servers for SQL and the 1C application;
• planned volumes of data processing;
• load distribution structure in client-server architecture

Selecting a processor and RAM

Calculating the frequency, the required number of processor cores, and the amount of RAM is the first and most important step. To consider several options, we will choose a server for 1C taking into account the company’s staff.

Small organization (up to 15 employees). With a small number of users, the database volume, as a rule, does not exceed 2 GB, and the 1C program in the form of a file version is installed on client machines. The OS needs in this case amount to 4–6 GB and another 4 GB are allocated for the system file cache. The processor load distribution looks like this:

• 2 cores – for the OS and terminal users;
• 1 core – for the 1C application server;
• 1 core – for SQL database.

Entry-level machines with one quad-core processor can cope with this task. These can be either rack or tower servers. The latter option is preferable because it does not require the allocation of a separate room for a server room.

Medium organization (up to 40 employees). With such a number of users, 1C developers recommend using terminal mode to access the application. The database size can be up to 4 GB. For such a load you need at least two processors with 4–6 cores. The optimal amount of RAM will be 16–64 GB, since a minimum of 700 MB must be allocated for each user. It is believed that the 1C application solution in which the client machine runs requires from 240 to 480 MB, and another 200–220 MB is allocated for office applications.

With such a number of processes, it is recommended to use one medium-level machine with virtualization or two physical servers. One of them will be used for terminal access, and the second for SQL. It is best to implement the 1C application server on the first machine or even allocate a separate single-processor system for this. The required configuration is selected in each specific case based on an analysis of processor time.

Large organization (more than 40 employees). The basic equipment configuration in this case will consist of three physical servers:

• terminal,
• DBMS,
• "1C".

Database volumes with such a number of employees often exceed 4 GB, and it is recommended to allocate at least as much RAM for the system cache. Another 4 GB will be used by the operating system, and 1C applications will require about 8 GB. Thus, you need at least 16 GB of RAM.

For such tasks, dual-processor servers with support for Intel Xeon E5-2600 or higher are selected. If the number of employees does not exceed 50 people, only one machine can be left for terminal access and 1C applications. However, given the company's growth prospects, it is better to provide a separate server for each task. If the number of personnel involved approaches 100 employees, you need to deploy a cluster of two machines for 1C, and leave one for other tasks.

Selecting a disk subsystem

Server performance directly depends on the disk subsystem. When running 1C applications, data reading and writing operations are performed with high intensity. Most complaints about server operation are related to tables being blocked when a large number of users access them simultaneously.

The task of choosing a server for 1C includes monitoring the disk subsystem, allowing you to find the optimal balance of performance and reliability. An extremely important factor affecting performance is its ability to perform a certain number of read/write operations per second (IOPS). If the database is up to 300 MB, and the number of 1C users is up to 6 people, this parameter is 400–600. If the number of server users reaches 100 people, then the IOPS will be 18,000. Streaming transfer speed plays a secondary role.

For each type of hard drive, read/write speed values ​​are set:

• SATA – 100/80;
• SAS – 240/220;
• SSD – 35,000/8,600.

This shows that solid-state drives are best suited for 1C database servers. The main factor limiting their use is their high cost. Therefore, to reduce the budget, SAS drives are also used. To store critical data, including 1C, hard drives are combined into RAID arrays of different levels, and the redundancy built into them should be included in the calculation of server performance.

When designing a solution, system fault tolerance plays an important role. Both hardware and software are used for this. The servers are equipped with hot-swappable power supplies and disk cages, and use a UPS for uninterrupted power supply. Data safety is ensured by backing it up. A log file is created at least once a day to ensure information recovery in case of system failures.

You can find the required server and configure it for 1C on the website. Our specialists will assist in solving this problem. For advice, contact them by phone or contact the manager via chat.

Server for 1C:Enterprise 8 for a small office

This material may be of interest to small organizations or branches with 3-25 users of the 1C:Enterprise 8 system. The author proceeds from the assumption that it will be read not only by IT specialists, but also by managers or accountants of small enterprises, therefore the material is somewhat simplified in technical terms. Its basic principles are quite applicable to systems with a large number of users. For organizations with 25+ 1C:Enterprise 8 users, two other materials will be published later, describing many years of experience in selecting server equipment and building IT infrastructures for medium and large implementations of 1C:Enterprise 8.

Who should be wondering if they need to buy a server?

First, let's describe the conditions when a server running 1C:Enterprise 8 is not necessary.
These are, as a rule, organizations that meet two criteria as closely as possible:
a) a small organization for 1-5 users,
b) To support the company’s business processes, the capabilities of the standard solution from 1C are quite sufficient (i.e., there is no need to make modifications to the standard configuration, at most changing some printed forms, reports and processing).
According to statistics, depending on the region, the number of organizations using standard configurations from 1C can be up to 60-80%.
If the company meets the criteria listed above, then perhaps instead of purchasing a set consisting of:
- “1C:Enterprise 8”,
- Windows Server and client licenses,
- hardware server,
- services for setting up equipment and software,
- support and maintenance of this complex,
It makes sense to consider renting 1C:Enterprise 8 from a service provider in the Software as a Service (SaaS) format.
Service providers of such services offer fully functional standard 1C:Enterprise 8 configurations, presented in the form of a remote Web site. Documents are printed to your local printer; you can also save documents electronically to your local computer. Of course, it is possible to save a local backup copy of your 1C database at any time. Moreover, you can place the backup copy on a computer disk, on a removable USB drive, or somewhere on a remote “cloud” drive (such as DropBox or Yandex.Disk).
You can rent one of the standard “1C:Enterprise 8” configurations from the 1C company itself on the 1cFresh.com resource, as well as from its partners, for example 1C:Accounting Services.
Among the technological conveniences, this means working anywhere there is Internet and from almost any device, including a tablet and even a smartphone, without the need to involve an IT specialist. Among the financial advantages - the organization, instead of accumulating funds, investing in the purchase of a physical server, the Windows Server operating system and the 1C:Enterprise 8 software itself, the organization simply pays for access to 1C:Enterprise 8 as a service, immediately writing off the costs as operating expenses . In fact, it’s like paying for a public service. If you need to increase or decrease the number of employees, this can be done almost instantly, thereby flexibly managing your costs depending on current needs. And hundreds of users will be able to work in such a service simultaneously.

“Let’s take tests,” or what would be desirable to record on current equipment?

If you already have a server running 1C:Enterprise 8, the ideal option would be to run a standard Windows utility - Performance Monitor (Perfmon), and record load data for one working day. This will be a starting point that will then allow you to compare how effective the new server is and whether it is worth the money spent.
Using Perfmon, it is enough to remove the load on the processor (CPU) in the form of % load on cores, random access memory (RAM) in % of use, and physical(not logical) disks: on which 1C and OS Windows are located (this can be one, but usually these are different disks). In terms of disk load, the most important parameters are:
- “Avg. Disk sec/Transfer” (average disk access time) - ideally up to 10 ms (milliseconds), good up to 25 ms, the limit value for comfortable operation is 40 ms;
- “Current Disk Queue Length" ( current disk queue length) - ideally the presence of narrow peaks (in the graph) and the smallest possible queue value, queue length of 70-100 requests or more, a graph in the form of a “slide” indicates insufficient performance of the disk subsystem;
- “Disk Transfers/sec” (disk access/sec) - here it is important to fix the maximum numbers and their duration, as a rule, this is from 80 IOPS to several hundred, or even thousands.
Having this information, it will be easier to formulate requirements for the server hardware subsystems for the equipment supplier, and then compare them with the calculated values.

Equipment calculation methodology

After deciding to purchase your own server for the needs of 1C:Enterprise 8, we will try to take into account the specific needs of this application - to select the optimal equipment for the task within the budget.

Next, we will consider the methodology for calculating the required resources for 1C:Enterprise 8 depending on:
- version used - file or SQL,
- type of access - via local network, via remote desktop “Remote Desktop”, or Web/thin client,
- number of users,
- number and volume of databases,
- other tasks assigned to the server.
The presented approach is relatively universal, and is equally suitable for running 1C:Enterprise 8 on a physical server (with Intel Hyper-Threading enabled), and for running on a virtualized server with Microsoft Hyper-V or VMware vSphere, and for calculating leased resources from cloud service providers.

For clarity of calculations, we will take three enough typical example:
A) 5 users"1C:Enterprise 8", file version, users connected via local network (LAN) via "fat" client 1C. In this mode, the server performs only the role of a file server, and the entire computing load falls on the user's workstations.
b) 10 users, database SQL, “1C:Enterprise 8” was launched. Applications server", users work through "fat" client via LAN, 4 GB database, for 2 years. The server acts as a data warehouse, SQL server, and 1C Application Server.
V) 20 users, database SQL, “1C:Enterprise 8” was launched. Applications server", users work in " Remote Desktop"(RDP), one configuration, 9 GB database for 3 years. The server acts as a data warehouse, SQL server, 1C Application Server, and terminal server.

Calculation of processing power requirements (CPU ) : First we need to calculate how many processor “cores” we need. We will carry out the calculation in “logical CPU cores,” where one logical core corresponds to one core on a physical server in the “Task Manager” with Hyper-Threading enabled (i.e., each physical processor core is represented as two logical ones).
- For the needs of the operating system (OS), we reserve 1-2 cores; for the SQL version, one is usually enough; for the file version, two are better.
- If the SQL version is used, then one core for 20-25 1C:Enterprise 8 users for the needs of MS SQL (more users - more cores);
- Also for the SQL version we reserve one core for the needs of “1C:Enterprise 8. Application Server x64” (rphost process) for 15-20 users (more users - more rphost processes, more cores);
- In the version where users work in the “Remote Desktop” mode, we reserve one logical core for 8 terminal users (more users - more cores).
- To work in Web service mode via a browser or “Thin Client” - we also reserve one logical core for 8 remote users. In fact, for configurations on “Managed Forms” the load is shifted to “1C:Enterprise 8. Application Server x64”, and in the case of a Web service it is also added to the Internet Information Server (IIS) Web server, but for our calculations this doesn't matter.

Calculation examples:
a) for 5 users in the file version, connected via a local network via a “thick” client, 2 logical cores are required for the OS (including the role of a file server) - a total of one physical processor core;
b) for 10 users in SQL and through a “thick” client, one core is required for OS, one core for MS SQL Server, one core for “1C:Enterprise 8. Application Server x64” (rphost) - a total of 3 logical cores, or 2 physical;
c) for 20 users in SQL and in “Remote Desktop” mode, you need one core for OS, one core for MS SQL Server, one core for “1C:Enterprise 8. Application Server x64” (rphost), 2.5 cores per servicing terminal user sessions (20:8) - a total of 5.5 logical cores, or 3 physical ones.
A server with the lowest Intel Xeon E3 12xx processor contains 4 physical cores, or 8 logical ones. Thus, even the minimum version of the server processor in terms of cores fully covers the needs of a small organization of 25 1C:Enterprise 8 users. And, for example, a single-processor server based on a more powerful Intel Xeon E5 16xx, containing 8 physical or 16 logical cores, can easily cope with a load of 50-75 1C:Enterprise 8 users in Remote Desktop or Web mode -service/thin client.

At the same time, one of the most important characteristics for a processor is its nominal frequency (not to be confused with Turbo Boost), which we will discuss in more detail below. To put it simply, the comfort of users working with 1C:Enterprise 8 in SQL mode, and especially when working via Remote Desktop (RDP), grows almost linearly with processor frequency.

Calculation of RAM requirements ) : As when calculating the need for processor cores, our task is to carefully take into account the needs of all services.
- For operating system (OS) needs, we reserve 4 GB;
- If you use the SQL version, then at least 20-30% of the volume of database tables (DB) must be placed in SQL RAM Cache. If there are several databases, then 20-30% of the volume of data tables for each database. Or, as a criterion for calculating SQL RAM Cache needs, you can use the volume of data for 1 year. Since small organizations usually have 1C databases that are not too large, it is very often possible to place the entire database in SQL RAM Cache, i.e. 100% of the volume of database tables (this is ideal). The minimum volume is from 2 GB.
- For the needs of “1C:Enterprise 8. x64 Application Server”, the amount of RAM is calculated based on the number of running rphost processes, about 1 GB for each rphost. Typically this is 1-2 GB for 10-25 users.
- When users work in Remote Desktop mode, everything is somewhat more complicated. First of all, it is necessary to clarify the configurations used, or even better, physically see how much each configuration consumes RAM per user in local network mode or in terminal mode. Moreover, you need to look not immediately after the start, but after about 20-30 minutes of intensive work. For example, “Accounting” on average consumes 250-300 MB of RAM for each running session, “Trade Management” on average consumes 300-350 MB of RAM for each running session. Next, we calculate how many users will simultaneously use each configuration, multiply their number by the required amount of RAM for the 1C:Enterprise 8 configuration, and sum by configuration to obtain the total volume. As a rule, for one configuration and 10 users this is 3-4 GB of RAM.
- In Remote Desktop mode, it is also necessary to take into account other possible loads. For example, 1C is often downloaded into MS Excel format for further processing of printed forms or reports. Or users are given access to use the Internet, other MS Office applications and other programs. Accordingly, it is necessary to calculate the RAM resources they consume. For MS Word and MS Excel this is approximately 100 MB each, for MS Outlook about 150 MB, Internet Explorer about 200 MB for each running instance for each user. For other programs, it is optimal to look at their actual RAM consumption on a PC and take it into account in the same way.
- To work in Web service mode via a browser or Thin Client, you can calculate RAM consumption using the same principles as for various 1C:Enterprise 8 configurations in Remote Desktop mode, counting the users of each confirmation, multiplying to the corresponding RAM consumption in “thick client” mode, and summed up. These additional resources will actually go to 1C:Enterprise 8. Application Server x64 and IIS, but for preliminary calculations it is quite suitable.
Calculation examples:
a) for 5 users in the file version, connected via a local network via a thick client, the minimum required is 4 GB of RAM under the OS, and preferably 8 GB of RAM;
b) for 10 users in SQL and through a “thick” client with a 4 GB database for 2 years, 4 GB for OS, 1 GB (this is 25% of the volume) or 2 GB (data for a year) for MS SQL Server are required, and 4 GB is better, so that 100% of the database fits in RAM, 1 GB for “1C:Enterprise 8. Application Server x64” (one rphost thread), a total of 6 GB to 9 GB of RAM;
c) for 20 “Trade Management” users in SQL with a 9 GB database for 3 years and in “Remote Desktop” mode, you need 4 GB for OS, 3 GB for MS SQL Server (or better yet, 9 GB for 100% The database was placed in RAM), 1-2 GB for “1C:Enterprise 8. Application Server x64” (1-2 rphost threads), 6-7 GB for servicing terminal user sessions, a total of 14 GB to 22 GB RAM.
After calculating the required amount of RAM, it is correct to add 20-30% of the reserve for load growth (an increase in the number of users, for example, or a database growth). Fortunately, RAM is inexpensive now, and modern processors support a lot of it - Intel Xeon E3 16xx up to 64 GB RAM, and Intel Xeon E5 16xx up to 1540 GB RAM. There is no such thing as too much RAM in a server ;-).

Disk subsystem :

The disk subsystem consists of two components:
- input/output subsystem in the form of water/output controllers (HBA) and RAID controllers;
- data storage devices, or in our case - SSD and HDD drives.

I/O subsystem ( RAID).
Since we are discussing a server whose task is to reliably store information, it is absolutely necessary to reserve hardware resources for storing data, i.e. disks.
Small businesses typically use RAID1, or “mirror,” where data is written to two disks simultaneously. In this mode, even if one of the disks physically fails, the data is saved.
There are several options at once constructionRAID1 on a small server.

1. For example, a completely software RAID (Soft RAID) can be created using Windows Server. This option does not apply to the system drive that contains the operating system (OS). For a disk with a database, you can try using Windows Storage Spaces technology. In real life it is used extremely rarely, we will not recommend it.

2. You can use hardware and software built on the basis of a chipset from Intel and Intel® Rapid Storage Technology ( IntelRST). Its essence is that all input/output operations at the hardware level are performed by the motherboard chipset, practically without loading CPU resources. But this array is controlled at the software level, using drivers for Windows.
This The most common, and at the moment the most high-performance option to build RAID1 for a not very loaded server with 2 or 4 disks.
True, like any compromise solution, it has some disadvantages.
a) Its operation depends on the drivers loaded into the operating system. And this carries some potential risk that when updating drivers or OS, a situation may arise that the RAID disk will be unavailable. It is extremely unlikely, because... Intel and Microsoft are very friendly and test their software very well, but it is possible. In fairness, it should be noted that over the past 8 years the author has not encountered cases of such failures.
b) Based on the results of experiments in the Entry test laboratory, indirect evidence suggests that the Intel RST driver model uses RAM resources for write caching. This gives a performance boost, but also carries some risks of data loss in the event of an unplanned server power outage. In the previous “reincarnation” of this technology, Intel Matrix RAID, write caching could be explicitly disabled at the command level. In the modern version of Intel RST, the user does not have the opportunity to influence this parameter in any way, or even find out its status. This issue can be easily resolved by installing a relatively “smart” uninterruptible power supply (Smart UPS), which can monitor the state of its batteries and, when they are low, gives a command to turn off the server. In fact, a UPS must be installed on the server in any case, so this is not a problem, the main thing is not to be lazy to make the settings. There is some issue with portability in case of motherboard failure. During the warranty period, this issue will probably be able to be resolved by the equipment supplier, but during the post-warranty period, you may need to search for a similar motherboard.
The cost of such a solution in most cases is already included in the price of the motherboard and, in fact, the user gets it “for free”.

3. Among IT specialists, it is quite common to want to have a fully hardware RAID in the server. A good example of such a solution is the use of a SAS controller (SAS HBA) in RAID1 mode. For example, LSI HBA 9211 and its successors. To do this, special BIOS firmware is installed in the SAS HBA; for the LSI 9211 it is “IR” firmware. This scheme does not provide any performance benefits. Theoretically, if the motherboard fails, you can quickly connect the disks and controller to another server... but with the same probability as the motherboard, the SAS controller can burn out, so from the author’s point of view, this advantage is somewhat illusory, solving more psychological problems than technological ones.
The cost of LSI HBA 9211 is $250-300, which is noticeably more expensive than the previous version based on Intel RST. This price increase is quite significant for a budget solution. From the author’s point of view, if you really want to make a “hardware” RAID, then it is better to choose a slightly more expensive solution based on the Intel® RAID Controller RS3WC080. This SAS HBA is also built on an LSI chip, but the next generation, LSI SAS 3008, supports the SAS 3.0 standard (12-Gb/s), at a price of about $300.

4. Sometimes, in case of dishonesty or insufficient qualifications of the seller, they try to sell inexpensive RAID controllers of outdated models for 1C servers. For example, Adaptec 6405E. The disadvantage of such controllers is that the performance chip built into them is designed to support a certain number of HDDs, and does not cope well even with the load of two junior server SSD models. For example, modern SSDs can easily produce 80,000 IOPS (requests per second) for reading each, and the processor of a RAID controller, for example, is capable of processing only 60,000 IOPS... Also, when using RAID1 and SSD, there is no need for a write cache on the controller - writing to the RAM cache on the controller and directly to the SSD occurs at almost the same speed as reading. Moreover, modern RAID controllers, even having 1 GB of RAM Cache on board, do not use it when working with SSDs. This doesn't mean the Adaptec 6405E is a bad controller, it's just that the tool is designed for a different use.
The cost of the Adaptec 6405E is about $250.

As a brief conclusion, I propose to look at the testing schedule of four SSDs in RAID10 in three RAID configurations: Adaptec 6405E, LSI 9211, Intel RST (Entry Test Laboratory). It is clearly seen that the most productive option is with Intel RST, the least productive is with Adaptec 6405E.

Storage devices ( SSDAndHDD)

“1C:Enterprise 8” in its work, in addition to the actual location of the Database tables (folder for the file version, DB tables for the SQL version), can use the system drive “C:\”. For example, the Tmp system folder of the Windows operating system for storing temporary files. The SQL version can work with it - in particular, store tempDB there. In the “Remote Desktop” mode, local tmp directories of terminal users on the server can be used, which are also usually located on the “C:\” drive, in local user profiles.
- For the required operating system (OS), we reserve at least 120 GB;
- To store a database in a file version, it is advisable to simply look at the volume of folders in which the databases are stored. Usually this is up to 1-2 GB, or even 10-20 GB.
- If the SQL version is used, then we have three types of data - the database tables themselves (DB), temporary tables (tempDB) and SQL log. By default, tempDB will be located on the system drive “C:\”. For small databases it is usually small, about 100-300 MB. Database tables (DB) - as written above, rarely exceed the volume of 10-20 GB for companies with 5-25 users, but here, first of all, you need to look at the current volume plus growth over the year. SQL log can be very large, up to tens of GB, especially when the “Full SQL log” mode is enabled, but it can also be painlessly “trimmed” at the end of each month and archived (or deleted if not needed).
In total, for the needs of the 1C:Enterprise 8 databases, we reach capacities of 20-60 GB, which is actually smaller than the smallest server drives, either HDD or SSD.
- In the case of users working in Remote Desktop mode, it makes sense to take into account 3-4 GB of RAM in users’ personal folders for storing various download or upload files/data to/from 1C. If the server is also used as file storage, this must be calculated completely separately from the needs for 1C, and preferably on other physical disks.

What else needs to be taken into account for design regarding the selection of disks.
1. It is very desirable to distribute the Operating System (OS) and 1C data (data tables) into different physical devices from the point of view of fault tolerance. But if the budget is small, well, let everything be on one storage, with backup through RAID and daily Backup to external media (or service).
2. Do not forget that the server must reserve disks and build a fault-tolerant RAID1 from them (RAID5 and its analogues are not used for databases), i.e. we need at least two identical disks (if the OS and the database are combined), or twice two identical disks for two RAID1 arrays (if the OS and the Database are separated).

Calculation examples:
a) for 5 users in the file version, connected via a local network via a thick client, the minimum required is 120 GB for the OS, or better yet 240 GB, and 10-20 GB for data. In fact, a disk subsystem consisting of two 240 GB Intel SSD s3510 series drives in RAID1 mode can handle this load.
b) for 10 users in SQL and through a thick client with a 4 GB database, a disk subsystem of two disks like a 240 GB Intel SSD s3510 series in RAID1 mode will also be technically sufficient. But, taking into account the number of users, it already makes sense to consider splitting it into two separate volumes - two disks in RAID1 of 120-240 GB for the OS and two disks in RAID1 of 80 GB for the Database.
c) for 20 users of “Trade Management” in SQL with a 9 GB database, a disk subsystem of two disks like a 240 GB Intel SSD s3500 series in RAID1 mode will also be technically sufficient. However, taking into account the number of users, it is strongly recommended to split the OS and Database into two separate volumes - two disks in RAID1 of 120-240 GB for the OS and two disks in RAID1 of 80-120 GB for the database.

I would like to remind you of the “principle of reasonable sufficiency”.
As a rule, the volume of databases for small companies with 5-25 1C:Enterprise 8 users is also small. And here it is very important not to chase disk space, you simply don’t need it. But it makes sense to choose the most reliable and productive disk, albeit with a smaller capacity. A typical mistake made by buyers is the desire to “buy a bigger disk, but it’s cheap,” which leads to the use of cheaper and more capacious non-server disks, which is unacceptable from a technical point of view.

Bypass bottlenecks

Network interface . It is least likely that when working with 1C:Enterprise 8, the bottleneck may be the network interface, because Since a large amount of data is not transferred, there are no special requirements for 5-25 users. Moreover, most servers have two 1 Gbit/s Ethernet network cards installed at once, but there are nuances here.
Network cards are different. Some are designed to work on PCs and laptops, and in them a significant part of the load falls on the CPU. These include, for example, network interfaces on the Realtek RTL8201N chip. Such chips are also used in servers, but on special ports designed to control the server.
At the same time, there are server network chipsets, such as the Intel® i350-AM2 Dual Port Gigabit Ethernet. With them, most of the processing occurs on the chip itself, without involving CPU resources, which is both faster and more efficient.
Actually, the recommendations are simple:
- do not buy a PC “a la server”, because most likely he has an Ethernet network card for his PC;
- if the server has several Ethernet ports, do not use the port intended for management for the workload.

RAM . Everything is simple here. How to calculate is described above. RAM is inexpensive these days. Therefore, do the math, take a reserve of 20-50%, or more, and close the issue. If a significant increase in load is expected in the next year or two, make sure that the server has free slots for installing additional memory modules.

CPU . After calculating the number of physical cores required, the question of frequency remains.
And this is a very important question.
1. If there are enough cores, then the performance of the client part, the 1C:Enterprise 8 Application Server, and in many situations the SQL server, directly depends on frequency processor. Moreover, it is almost linear. Many reports, for example, are executed exactly the same times faster when the processor frequency is increased by 1.5 times.
Therefore, other things being equal, it makes sense to give preference to higher-frequency processors.
2. The next point is not to fall for the marketing ploy of some sellers who give out the processor frequency in Turbo Boost per processor frequency. Yes, Intel® Turbo Boost Technology 2.0 is quite an interesting thing, and in the realities of 1C, when only one thread is running (processing documents or generating a complex report), it allows you to increase the performance of one core by 15-30%, or even more. But you need to remember that in reality, in a server chip, the frequency will rise for a short period of time, for 30 seconds, sometimes for one minute, and then decrease to the standard one. As a result, the gain is very short-term. And a high-frequency processor always operates at a high frequency, which means faster. That's why it costs more.
An example is in the table below:

Table 1

3. For small businesses with 3-10 users, they sometimes offer to buy not a server based on Intel Xeon E3, but a “powerful PC” based on Core-i7, citing its greater performance and reliability.
But you need to consider the following points.
a) The Intel Xeon E3 and Core-i7 processors are actually twin brothers in terms of hardware, and even cost the same. But the internal firmware responsible for setting priorities is different. Very simplified - the Xeon E3 has priority for input/output operations and other server operations, while the Core-i7, tailored for the gaming segment and processing video streams, gives priority to servicing the video card. The price, I emphasize again, is the same (with the same parameters).
b) There is a big difference between desktop and server motherboards. The desktop is designed to work 8 hours a day, 5 days a week, for two or three years. Accordingly, the components for it are selected to withstand the specified service life. Server motherboards are designed to operate 24 hours a day, 365 days a year for three or five years. And the components there are somewhat different. But the difference in price is again minimal, often at $10-20.
c) In an “advanced” PC, with a high degree of probability the network card will not be a server card.
Does it make sense, instead of a server as a specialized device with given operating parameters, to try to buy some kind of “sophisticated PC” for the role of a server, and it’s not even a fact that it’s cheaper - everyone can now make an informed choice.

Discs. This is probably the most pressing topic at the time of writing.
Unfortunately, there are still both users and sellers who are slightly stuck in the last century. And we are absolutely confident that HDD SAS 15,000 rpm is an example of reliability and performance.
In fact, this has not been the case for a long time.

a) First, let's deal with reliability SSD and HDD.
Most often, the theoretical reliability of disks is assessed by the “Non-recoverable Read Errors per Bits Read” parameter, which can be translated as “The probability of a non-recoverable read error occurring per number of bits read.” It shows how much data needs to be read from the disk in order for there to be a high probability of an unrecoverable error occurring.
Another important parameter showing the probability of disk failure is AFR (annual failure rate), or “Annual Failure Rate”.
The table below shows data for typical drives SATA Desktop HDD 7200 prm, SATA Enterprise HDD 7200 prm (SATA Raid Edition), SAS HDD Enterprise 15,000 prm, SATA SSD Enterprise (data taken from official documents of manufacturers, you can check the numbers using the links).

Parameter	Disc type
	Desktop SATA 7200 rpm		Enterprise SAS 15,000 rpm (10,000 rpm)	Enterprise SATA SSD
Non-recoverable Read Errors per Bits Read
Volume, when reading which is statistically expected to cause an unrecoverable error

As can be clearly seen from the table, according to the “Annual failure rate” parameter, a disk for use in a regular PC is on average two times less reliable than a server disk.
Regarding the likelihood of unrecoverable errors, the theory clearly tells us that the Enterprise SATA SSD, for which the Intel® SSD DC S3510 Series was taken as an example, has a 10 times lower error probability than a SAS HDD Enterprise 15,000 rpm, 100 times lower than SATA Enterprise HDD 7200 rpm, and 1000 times lower than SATA Desktop HDD 7200 rpm.
At the same time, the price of a SATA Desktop HDD 7200 rpm and an Enterprise SATA SSD for a volume sufficient to accommodate both the operating system and 1C databases differs by no means 1000 times, or even 10.
I would especially like to draw your attention to “Volume, when reading which an irrecoverable error is statistically expected.” For SATA Desktop HDD this figure is 12.5 TB. And there are already 8 TB and 10 TB disks... thus, if we install, for example, an 8 TB desktop disk, write it and read it twice, we, in theory, will encounter at least one unrecoverable error!

Brief summary:
- use Enterprise-class SSDs in the server, they are at least no worse, and theoretically more reliable than any HDD.

b) Next we estimate performance SSD and HDD.
From the point of view of databases, which, in essence, is 1C, the most important are only three disk parameters
- Latency, or disk response time, is measured in microseconds (less is better);
- the number of read operations per second (Disk Reads/sec), measured in IOPS (more is better);
- the number of write operations per second (Disk Writes/sec), measured in IOPS (more is better).
Let's put these three parameters into one table for the same drives as in the example about reliability.

Parameter	Disc type
	Desktop SATA 7200 rpm	Enterprise SATA\SAS NL 7200 rpm	Enterprise SAS 15,000 rpm (10,000 rpm)	Enterprise SATA SSD
Latency (disk read/write response time), microseconds
Disk Reads/sec (number of read operations per second), IOPS
Disk Writes/sec (number of write operations per second), IOPS

As can be clearly seen from the table, SSD according to the parameter response time superior to HDD 40-80 times, and by number of I/O operations per second in 100-400 times (!!!).
At the same time, if you approach the choice wisely and buy only the storage capacity that is actually in demand, then the difference in cost of Enterprise SATA SSD and Enterprise SATA \ SAS NL HDD will be very insignificant.
Is it reasonable to use HDD to host databases? From the author’s point of view, only if you bought illiquid stock for a warehouse, you really need to sell it, and further relations with the buyer are of little concern to you. Because both the price/performance ratio and the price/reliability ratio are clearly in favor of the Enterprise SSD.

And now it’s time to return to the moment when real indicators of the load on the disk subsystem were measured (if they were able to be done).
If we talk about the “average temperature in the hospital”, then the approximate peak loads by IOPS can be as follows (number of users, 1C database volume; in the bottom line Disk Transfers/sec)

Knowing the number of users and the volume of 1C databases, it is quite possible to approximately estimate the need for the disk subsystem in IOPS by Disk Transfers/sec (=Disk Reads/sec + Disk Writes/sec), or take your actual measurements and formulate the requirements for your disk subsystem in input/output operations per second (IOPS). And armed with numbers, choose those drives that will satisfy them.

c) And to close the question of which SSDs should be installed, let’s figure it out: what is the difference Enterprise SATA SSD and regular desktop SATA SSD .
1. The performance of data read speed in IOPS for both desktop and server drives will be very similar. And here performance speed data recording will differ significantly, and with increasing disk space utilization on desktop SSDs, it will quickly degrade. Yes, in the specifications for desktop disks you can see very high IOPS per write... with 5-8% disk full of data. And at 100% - they are not even given, and for good reason - these indicators often do not differ from those of the HDD. For server SSDs, writing performance testing is performed exactly when the data is 100% full, and its value is, as a rule, either the average or one of the worst results. Conclusion - do not pay attention to the beautiful and large numbers in the manufacturer's specifications for desktop SSDs, this is marketing. For the server, you need to choose the youngest, but Enterprise SSD model. For example, such as Intel® SSD DC S3510 Series.
2. The next important parameter, which clearly demonstrates the fundamental differences between server and desktop SSDs - guaranteed rewrite resource. Most desktop SSDs normally tolerate overwriting up to 0.1% of their capacity per day (without significant degradation of performance and failure) all the same 7 days a week for 2-3 years, if not completely filled. The server disk is designed to overwrite 0.3% of its capacity every day for 3-5 years, even when completely full.
Points 1 and 2 are explained very simply. Data is written to SSDs in 4 KB cells, but is erased... at least in blocks of 256 cells or more. And before erasing the entire block, they are simply marked as ready to be erased. To replace old data with new data, the old data must be erased. To erase even one 4 KB cell, you must first transfer all the data from a column of 256 cells to another place, erase the entire column, and return the data to its place. This is not a quick operation.
This situation is dealt with by placing a hidden area on the SSD disk, which is not accessible to the user, and is used specifically for replacing cells when there is a need to “collect garbage”, i.e. clear unused blocks. This area is called Over Provisioning, or “Reserve Area”. So, for desktop SSDs it is usually 4-8% of the total capacity of flash memory chips on the disk, and for server ones... reaches 42% of the physical capacity of the chips. If we take an example, provided that the device contains 320 GB of chips, then the capacity of a “desktop” SSD will be 300 GB, and in the case of a server SSD, only 180 GB will be available to the user. In fact, everything is much more complicated; server disks use a lot of technologies to increase their “survivability” and ensure stable performance, but for a general understanding of the difference, the example with the “Reserve Area” is quite indicative.
3. Another important difference between server and desktop SSDs relates to the area of ​​data safety. Any SSD has its own volatile RAM, which is used, among other things. for read and write operations. In desktop SSDs, there may be circumstances where data is written to RAM on the SSD, the file system and SQL server have received write acknowledgment, and the data is not actually in non-volatile memory yet. If at this moment a power failure occurs, then the probability of data loss is very high, and it is very difficult to restore what is lost and what is not.
In the same time in server roomsSSD there is a supercapacitor, the capacity of which is sufficient to write into the non-volatile SSD memory all the data located in the RAM inside the SSD. Thus, the likelihood of data loss during a power failure is greatly reduced.
And again, the difference in price between server and desktop SATA SSDs is not significant at all.
Brief summary - to store important data, such as 1C databases, the server should use exclusively server Enterprise SSDs.

Case, power supply and UPS

The most common are three sizes for small, single-processor servers:
- mounted in a 19” rack (Rack-mount),
- in a regular pedestal case (Desktop),
- modern “cubes”.
If you are going to place the server on the site of a hosting provider or a specialized server room, the optimal Rack-mount format will be 1U or 2U high. And such a server is strictly not recommended to be placed in the same room where people are present due to its high noise level.
The Desktop case size is used when the server will be in the same room with people. Such servers are relatively quiet, not much different from a PC, and even often serve as a workstation for one of the employees.
A good example of a “User Friendly” approach are specialized “cube” cases.

Besides the fact that they look nice, they are also quiet.

The picture shows an example of the design of a “cube” server cooling system, which provides good cooling to all components and at the same time creates a minimum of noise:
- the motherboard is on the bottom,
- warm air cools the board and processor, and rises upward from the processor radiator,
- where it is pulled by a low-speed fan of a large-diameter power supply,
- and is thrown out,
This server design is both beautiful, effective in reducing noise, and quite productive.

We should also touch on the eternal topic of whether excess power is needed or not.
From the author’s point of view, with a server costing over $1000, spending on top of it just to provide redundant power supply (two power supplies in the server), an additional $400 or so may really be necessary, only if you do not have the ability to promptly take the server for repair in 2-4 hours .

If the server is rack-mount, then it is located on the provider’s site or in the server room, where, as a rule, high-quality power supply is provided. And if he is in the office, then in the vast majority of cases, in 2-4 hours it is quite possible to go to a service center and replace the power supply. If you are far from the service center, then, as an option, you can buy a spare power supply and put it in the closet as “spare parts”, and if it breaks down, replace the failed power supply with a spare one within 15-20 minutes.

Speaking about the power supply and the likelihood of its failure, it is necessary to remember about the uninterruptible power supply (UPS) for the server. At a minimum, it should include voltage adjustment (AVR) capabilities, and even better, be Interactive. To reduce the risk of data loss, it is desirable that the UPS has the ability to turn off the server when the battery level is low (and that everything is connected and configured). And its power should ensure at least battery operation for 10-15 minutes, which in most cases is enough to shut down the server correctly.

We assemble a server “for the task”

Now, armed with knowledge, let's configure three example servers for three small organizations with different workloads.

a) For five users in the file version, connected via a local network via a “thick” client, a server in the “cube” form factor, with a quad-core Intel Xeon E3 12xx processor, 8GB RAM, two SSD Intel s3510 240 GB in RAID1 on board will be sufficient Intel Rapid Raid.

b) For ten users in SQL and through a “thick” client, a server in a “cube” form factor, with a quad-core Intel Xeon E3 12xx processor, 16 GB of RAM, two SSD Intel s3510 240 GB in RAID1 on an on-board Intel Rapid Raid, will be sufficient.

c) For twenty users in SQL and in Remote Desktop mode, it is better to take a server in Desktop or Rack-mount size, with a six-core Intel Xeon E5 166x processor, 32 GB RAM, two SSD Intel s3510 120 GB in RAID1 for hosting databases 1C and two SATA HDD (RAID Edition) 2-4 TB in RAID1 to host OS, Backup and user data, as well as file storage, and again on the on-board Intel Rapid Raid or on the Intel® RAID Controller RS3WC080.

And some absolutely practical advice.
1. There is no need to “save on small things” and use inappropriate tools - a server processor, a server motherboard and a server SSD work better in a server. The difference in price with desktop components is very insignificant, but in terms of functionality it can be disproportionately large.
2. There is little point in planning resources for more than three years. Technologies change so quickly that in a year or two it may be more efficient to replace a disk or server than to initially plan for a resource for five years.
3. The principle of reasonable sufficiency also applies to the choice of server. With a database volume of 10 GB, a specialized server with a capacity of 80 GB is much preferable to a “desktop” one with a capacity of 200 GB.

Of course, everything described above is not dogma. These are empirically derived calculations based on the author’s many years of experience in optimizing hardware for various 1C company platforms.

Today, the financial product 1C has grown from an accounting application program for accounting into a wide-format complex for accounting and support of almost any type of business, claiming to compete with the world “monsters” SAP R/3 and Microsoft Dynamics AX (Axapta).

Russian companies are increasingly organizing their business processes using modern configurations 1C 8.3 “Trade Management”, “Production Management”, “ERP Enterprise Management” and the like. The accounting, marketing, production, and sales departments are being transferred to 1C, and integration with IP telephony and document management systems is being carried out. However, immediately after the intentions “let’s work in 1C”, questions arise - what resources will the central 1C database work on, what hardware will show the optimal result for a reasonable budget? In this situation, it is easier for giant public sector enterprises - a clear command has been given to numerous full-time IT integrators and architects, the mechanisms of large-budget tenders have begun, with the obligatory condition of providing a turnkey concept and further support of the system by certified specialists. But what about companies that want to purchase and install one of the 1C: Enterprise products themselves, spending their budget wisely?

The most basic mistake, if you do not take into account the use of pirated or untested software, is saving on hardware for 1C. Similar trends are especially common in startups and small companies. There is an opinion that it is not necessary to buy expensive server equipment with Intel Xeon processors, there is no need to first calculate the amount of RAM, the load on the CPU and disk subsystem, that there is no need to create redundancy of disk arrays (Raid), to use professional disk controllers with Cache-RAM and etc. Errors in IT architecture calculations for 1C lead to dire consequences, which the company learns about after business processes have stopped. Therefore, it is very important to pay attention to each hardware node of the server platform for 1C.

Examples of typical problems due to incorrect construction of IT architecture for 1C:

• “Slowdown” of the 1C database and interfaces due to excess load on key resources (usually RAM or disk subsystem).
• Errors and crashes of the 1C program due to instability of incorrectly selected equipment.
• Downtime of the company due to the failure of central hardware.
• Partial or complete loss of 1C data due to random failures of hardware or software.

1C server hardware resources

Let's consider below the most key hardware resources, an error in the selection of which can ruin the entire enterprise automation project when you independently create a server for 1C.

Central Processing Unit (CPU)

Number of physical CPU cores. The topic of eternal debate on all kinds of 1C forums is what is more important: CPU frequency or multi-core. The roots of these contradictions go back to 1C 8.0 or even 1C 7.7. Indeed, the executable processes of 1C earlier versions were purely single-core, i.e. no matter how many cores the central processor provides, the 1C 8.0 enterprise server service or the 1C 7.7 thick client always occupied only one “zero” core in the operating system. Today, the picture has changed - the operating system boldly distributes the tasks of one 1C: Enterprise process (rphost) across several CPU cores (see Figure 1).

Figure 1 - CPU load when 1C server processes are running.

But this absolutely does not mean that if you buy a processor with the maximum number of cores, then a 1C server paired with a DBMS (most often by DBMS we mean MS SQL) will show fantastic performance and re-running accounting periods in the 1C program will be a matter of a few minutes. You need to understand the difference between the speed of performing one operation and the process of simultaneous processing of a large amount of information. The number of physical cores just allows us to solve the issue of stability and performance of simultaneous work with many different tasks by the 1C:Enterprise server and the DBMS. Hence the conclusion - the greater the number of 1C users, the more important the required number of cores will play for the comfortable simultaneous work of these same users. The dependence of the number of users on the number of cores for the 1C server is shown in Table 1.

Number of concurrent users on the 1C:Enterprise server	Processor type and model	Number of cores used
Up to 10 users	Custom Intel Core from 3.1Ghz	No more than 2-4
Up to 20 users	Server Intel Xeon from 2.4 Ghz	From 4 to 6
Up to 30 users	Server Intel Xeon from 2.6 Ghz	6 to 8 cores
Up to 50 users	Server Intel Xeon from 2.4 Ghz – 2 pcs.	From 4 per processor

Table 1 - The ratio of the number of users on the 1C server and the recommended number of CPU cores.

CPU frequency. As opposed to the number of cores, the frequency of the central processor affects precisely the processing speed of one piece of a task at one time, which is the most popular criterion for 1C end users. The processor frequency is precisely the parameter that, when increased, will increase the speed of processing requests by the 1C server and the DBMS for an individual user and reduce the time during which the system provides the final result to the end user. In confirmation of this, the well-known specialist Gilev, in one of his articles based on practical tests, made an unambiguous conclusion - “the speed of 1C is much more influenced by the frequency of the central processor than its other parameters, be it the 1C end client or the 1C: Enterprise server.” This is the architecture of the 1C program.

Cache, virtualization and hyper threading. In the past, when multi-core processors were not yet so common, Intel came up with a special central processor technology that simulated multi-cores, the so-called “hyperthreading”. After enabling it, one physical processor (one physical core) is defined by the operating system as two separate processors (two logical cores). We recommend disabling “hyperthreading” for the 1C server. This technology does not bring any acceleration to 1C.

When using virtual machines for a 1C:Enterprise server and a DBMS, you need to take into account that the cores of virtual machines are “weaker” than real physical cores, although they are called the same – “cores”. There are no exact official coefficients, but articles on Microsoft technical portals recommend counting 4-6 processor cores per physical core in a virtual machine.

A cache is an advanced memory used by the processor to reduce the average access time to computer memory. In fact, it is an integral part of the processor, since it is located on the same chip and is part of the functional blocks. Everything is very clear here - the larger the cache size, the larger “pieces” of information the processor can process. Typically, the size of the cache depends on the processor model - the more expensive the model, the usually larger the amount of cache memory. However, we do not believe that the size of the processor cache radically affects the performance of the 1C server and DBMS. Rather, this belongs to the area of ​​“fine tuning”.

Processor type. Everyone knows that hardware is divided into server and user. Is it possible in some cases to use an inexpensive user central processor as an alternative to a professional, but expensive server CPU? It turns out that it is possible. Let's look at a table comparing the main parameters of two versions of Intel central processors (see Table 2).

	Custom Intel® Core™ i7-6700T Processor (8M Cache, up to 3.60 GHz)	Server Intel® Xeon® Processor E5-2680 v2 (25M Cache, 2.80 GHz)
Cache memory	8MB	25 MB
System bus frequency	8 GT/s DMI3	8 GT/s QPI
Command set	64-bit SSE4.1/4.2, AVX 2.0	64-bit AVX 2.0
Number of Cores	4	10
Base processor clock speed	2.8 GHz	2.8 GHz
Max. amount and type of RAM	64 GB non-ECC	768 GB ECC
Estimated cost	354$	1 280$

Table 2 - Comparison of the main parameters of home and server CPUs from Intel.

As we can see, the server processor has much higher values ​​in the number of cores, cache size, support for more RAM and, of course, a higher price. However, a server CPU is practically no different from a user CPU in support of certain processor commands (instructions) and clock speed. From this we can conclude that for small organizations it is quite acceptable to use a custom central processor for the 1C:Enterprise server. The only question is that a custom processor cannot be installed in a server motherboard socket and support server RAM with parity control (ECC), and the use of custom components entails risks for the stability of the entire system as a whole.

Random access memory (RAM)

Type of RAM. Random access memory (RAM) varies according to its purpose - for multi-user server systems or for personal devices - PCs, laptops, nettops, thin clients, etc. As in the case of the CPU - the main parameters of RAM modules are approximately equivalent - modern RAM for a PC practically does not lag behind the server RAM either in the volume of one strip, or in the clock frequency, or in the type of DDR modules. The differences between server RAM and “home” RAM are in the use cases and purpose of the hardware platform - hence its higher cost:

• Server RAM has parity control ECC (Error Correction Code) - an encoding/decoding technique that allows you to correct errors in information processing directly by the RAM module
• A server motherboard has many more slots for installing RAM modules than a regular PC.
• Server RAM contains registers (buffers) that provide data buffering (partial Registered or full Full Buffered), thereby reducing the load on the memory controller with many simultaneous requests. Buffered FB-DIMMs are not compatible with unbuffered ones.
• Registered memory modules also allow for increased memory scalability - the presence of registers makes it possible to install more modules in one channel.

We can conclude that the use of server RAM modules makes it possible to install large amounts of RAM in one system, and ECC parity control techniques and the use of buffers allow the server operating system to work stably and quickly.

Amount of RAM. One of the key factors for high performance of the 1C server and DBMS is a sufficient amount of RAM. Of course, the actual RAM needs depend on many factors - the type of 1C configuration, the number of 1C:Enterprise server processes, the size of the DBMS database, and so on. However, it is possible to derive an approximate dependence of the amount of RAM on the number of users (see Table 3).

RAM requirements for 1c server and DBMS	Up to 10 users	Up to 20 users	Up to 30 users	Up to 50 users
Server 1c:Enterprise	4-6 GB	6-8 GB	12-14 GB	18-24 GB
MS SQL Server	4-6 GB	8-10 GB	16-18 GB	24-28 GB

Table 3 - Approximate ratio of the number of 1C server users and recommended RAM for 1C:Enterprise server processes and MS SQL server.

Regarding 1C:Enterprise server processes (rphost.exe) - modern 1C platforms do not allow you to manually specify the number of 1C server processes. Instead, the system requires you to set parameters such as the number of infobases and the number of users per rphost.exe process, after which it automatically determines the optimal number of 1C:Enterprise server processes. You can also configure the rphost.exe process to smoothly release RAM if its volume exceeds a predetermined threshold. In this case, the 1C server creates a new process rphost.exe, which gradually takes over 1C tasks, allowing the required 1C process to be offloaded.

You should also pay attention that the amount of RAM allocated to the SQL service is considered sufficient if the SQL data in the cache is at least 90%. This metric is quite convenient, because... You can't just look at the amount of RAM consumed by a SQL server - recent releases of SQL have dynamically consumed RAM - the maximum possible amount of RAM is grabbed and released as RAM is requested by other processes.

RAM frequency. In short, this is the bandwidth of the channels through which data is transmitted to the motherboard, and from there to the processor. It is desirable that this parameter matches or exceeds the permissible frequency of the motherboard, otherwise the RAM transmission channel risks becoming a bottleneck. Within one type of DDR, increasing/decreasing the frequency does not fundamentally affect the performance of the 1C server and belongs more to the area of ​​“fine tuning”.

RAM timings. This is the latency or latency of the RAM. This parameter is characterized by the data delay time during the transition between different modules of the RAM chip. Lower values ​​mean faster performance. However, the impact on the overall performance of the server system, and even more so on the 1C:Enterprise server, is low. Typically, only gamers and overclockers pay attention to these parameters, for whom every extra drop of performance is most valuable.

Disk subsystem and HDD hard drives

Hard drive controllers. The main device for connecting and organizing hard drives in a hardware system is the hard drive controller. It comes in two types:

1. Built-in – the controller module is built into the system, the hard drive cage is connected directly to the motherboard. It is considered a more economical solution.

2. External – is a separate printed circuit board (device) that is connected to the motherboard connector. It is considered a more professional solution due to the fact that it has separate chips for conducting and monitoring operations with HDD hard drives. Recommended for important server systems, such as 1C:Enterprise server and DBMS.

There is also a third type - a device for receiving/transmitting block data via iSCSI, FiberChanel, InfiniBand, SAS channels. However, in this option, the disk subsystem is “removed” to a separate data storage device (DSD), connected to the server via an optical or copper cable. In our article we analyze the requirements for a standalone server for 1C, so we will not consider this type.

Types and levels of RAID arrays. It is a data virtualization technology that combines multiple disks into a logical unit for redundancy and improved performance. Let's look at the most popular levels of the RAID specification:

• RAID 0 (“Striping”) It has no redundancy, and distributes information immediately across all disks included in the array in the form of small blocks (“stripes”). Due to this, performance increases significantly, but reliability suffers. We do not recommend using this array type, despite the performance benefits.
• RAID 1 (“Mirroring”, “mirror”). It has protection against failure of half of the available hardware (in the general case, one of two hard drives), provides an acceptable write speed and gains in read speed due to parallelization of requests. This type of array will fully support a 1C+DBMS server for up to 25-30 users, especially if 15K SAS disks or SSDs are used.
• RAID 10. Mirrored pairs of disks are arranged in a “chain”, so the volume of the resulting volume can exceed the capacity of one hard drive. In our opinion, the most successful type of disk array, because... it combines the reliability of RAID1 and the speed of RAID 0. In combination with SAS 15K disks or SSDs, it can be used for 1C servers from 40-50 users.
• RAID 5. Famous for its efficiency. By sacrificing the capacity of just one disk from the array for redundancy, we gain protection against failure of any of the system’s hard drives. (its variation of RAID 6 requires an extra two hard drives to accommodate checksums, but it saves data even if two drives fail). This type of array is economical, reliable and has quite noticeable read performance. Unfortunately, the bottleneck of this array is the low write speed, which makes it comfortable to use with 1C server configurations of up to 15-20 users. It is also optimal for applied purposes - storing file data, document flow archives, etc.

Types of hard drive interfaces. By type of connection, hard drives are divided:

• HDD Sata Home. The cheapest hard drive option designed for use in home PCs or network media centers. It is strongly not recommended to use such devices in 1c servers due to the low fault tolerance and operational stability - the components of these disks are simply not designed to operate 24/7 and quickly fail.
• HDD Sata Server. This name usually refers to hard drives with a Sata interface and a spindle speed of 7,200 rpm. The prefix “Server” means that such disks have been tested for performance in server systems and are designed for stable operation 24/7. Typically used in 1C servers to store large volumes of information that do not require high processing speed. For example, 1C archive databases, exchange folders, office document download files, etc.
• HDD SAS Server. There are several differences between the SAS interface (a modern analogue of SCSI) and the Sata interface. Here is the average disk response time, and work in a shared disk shelf, and work with the HDD controller at higher information exchange speeds - up to 6 GB/s (compared to Sata 3 GB/s). But the main advantage is the existence of SAS disk models with a spindle speed of 15,000 rpm. It is this design feature that allows SAS disks to perform almost 3 times more I/O operations per second compared to Sata Server HDDs. Such SAS disks have a small capacity and are recommended to be used for main 1c databases with a constantly high workload.
• SSD drives. These drives differ from the previous ones not in the connection interface, but in their design - they are solid-state and have no moving parts, i.e. In essence, they are analogues of “flash drives”. Such technologies allow SSD drives to produce an “exorbitant” number of input/output operations per second (from 10,000 operations on the simplest SSD models). However, this advantage also has a downside - the higher price of SSD drives and the “threshold of their life,” which depends on the limit on the number of writes to SSD blocks. However, every year these discs become more affordable and durable. Since the cost of SSD disks increases many times depending on the volume, it would be most reasonable to use them for small but over-loaded 1c databases that require high access speed, as well as for temporary databases of the TempDB DBMS.

IOPS – number of input/output operations per second. Essentially, IOPS is the number of blocks of information that can be read or written to the media in 1 second of time. That is, in its pure form, this is the key parameter of the speed of information processing by the hard drive, which affects the performance of the 1C server. If we take a standard 4kb block of information for comparison, we can roughly highlight the following IOPS indicators (see Table 4).

HDD	IOPS	Interface
7,200 rpm SATA drives	~75-100 IOPS	SATA 3 Gb/s
10,000 rpm SATA drives	~125-150 IOPS	SATA 3 Gb/s
10,000 rpm SAS drives	~140 IOPS	SAS
15,000 rpm SAS drives	~175-210 IOPS	SAS
SSD drives	From 8,000 IOPS	SAS or SATA

Table 4 - IOPS indicators on various types of hard drives when working with a 4kb data block.

Of course, in its pure form, IOPS is of little use for calculating final calculations and requirements for the disk subsystem of a 1C server. After all, the total performance of the disk subsystem consists of the type of RAID array, disk types and indicators of its interface speed, response time (Latency), random access time, percentage of the number of read and write operations and many other factors. However, this parameter, in our opinion, is a key indicator of the speed of the disk subsystem and at the stages of developing a server architecture, it helps to determine what type of hard drives will be most suitable for certain needs. (see RAID calculator)

Practice test

What is the relationship between the number of 1C users and the number of iops? Our team conducted a practical test (see Table 5) to measure the load on the disk subsystem with a certain number of 1C sessions. Since the 1C system is a programmable environment and each company can have its own set of business processes in 1C, we needed a link to a certain reference configuration for testing. In this capacity, a specialized configuration of the 1C CPU was chosen, developed for testing and debugging. Based on it, our 1C programmers added a number of queries that simulate the normal operation of an ordinary enterprise, with the formation of accounting queries, postings, drawing up reports and posting operational documents.

		System disk		Disk with databases
Iteration	Users	IOPS write	IOPS read	IOPS write	IOPS read
Average values
1	12	9,1	0,1	13,1	1,5
2	20	7,9	0,1	21,8	0,4
3	32	5,2	0,006	36,1	5,2
4	40	7,7	0,013	27,52	1,3
5	52	7,7	0,006	32,04	0,94

Table 5 - Results of a practical test on the load on the disk subsystem.

The test results show that the lion's share of the load on the disk subsystem occurs when 1C writes to the database of the DBMS server and to the system disk of the operating system (on which the 1C:Enterprise cache server files are located by default).

At the same time, we carried out practical measurements of already working 1C UPP 8.2 databases during the test period - 5 working days. They show that on average a 1C + DBMS server consumes twice as many iops “for writing” as “for reading”. This difference between synthetic tests and monitoring statistics of a real 1C server is due to both periodic sampling of information data from the database during the working day, and regular reading of the database during DBMS backup or replication.

Other components of the hard drive that are worth paying attention to.

• Physical size (form factor). Today, almost all known drives for personal computers and servers are 3.5 or 2.5 inches in size. Please note that 2.5-inch drives are not produced in large volumes.
• Random access time- the time during which the hard drive is guaranteed to perform a read-write operation on a certain section of the magnetic disk. As a rule, server drives have better results. This is a fairly important parameter when building a disk array for a 1C DBMS server.
• Spindle speed- the number of revolutions of the hard drive spindle per minute. Everything here is simple and clear - the access time and average data transfer speed of the hard drive depend on the rotation speed of the spindle with magnetic plates.
• Hard disk buffer capacity- a buffer is a temporary memory designed to smooth out differences in the speed of reading/writing a hard drive and transferring data over the interface.
• Reliability- is defined as the mean time between failures (MTBF). As a rule, reliability directly depends on the manufacturer, price and environment in which the hard drive is used. We consider reliability an important parameter of a hard drive that affects the quality of operation of a 1C server.

The right choice: home or server hardware

The reduction in price of hardware components and the active growth of the potential capacity of “home computers” lead to another disastrous misconception - small businesses actively use workstations as a platform for collaboration with 1C databases. At the same time, not realizing that in addition to the core frequency parameters, memory size and the ability to use budget SSD drives in a regular PC, there are more systemic, deeper and more important requirements for the operation of hardware in a commercial structure (see Table 6).

To solve the issue of organizing a 1C server, we offer rental of 1C cloud servers in Tier III class data centers. The economic feasibility of choosing to rent a server can be found in the article.

Options	Server	Personal Computer
Adequacy of computing power	V	V
Guaranteed system availability 24/7	V	X
Reliability and stability of key hardware components	V	X
Remote Power and Console Management (IPMI) capability	V	X
Budget cost of the hardware platform	X	V

Table 6 - Comparison of home and server hardware according to the criteria required for high-quality operation of the 1C server.

Fault-tolerant operation of 1C

Of course, one of the important requirements for the 1C server part is the stability of its operation and resistance to failures. Microsoft and 1C itself have put a lot of effort in this direction, creating technologies for clustering their services at a fairly serious level (see Table 7).

Fault tolerance of SQL servers

Based on the concept of a single common data warehouse. Built-in SQL Server clustering technology combines two SQL servers into one cluster with a single virtual IP address and a single database. Thus, if the main SQL fails, queries are automatically transferred to the backup one. The second option is the recently introduced AlwaysOn - a technology for automatic regular replication of DBMS databases between the main and backup SQL servers. At the same time, the duplicate SQL server is physically located on another storage, which increases risk resistance

Fault tolerance of 1C:Enterprise server service

1C Enterprise servers are combined into an active-active software fault-tolerant cluster with automatic failover and saving of current sessions.

Table 7 - Fault tolerance of SQL and 1C servers.

However, each technology has both pros and cons. In addition to the key advantages, you need to know some features of 1C and SQL clustering () so as not to end up with a deterioration in the performance of the service:

• SQL clustering uses a virtual IP. This means that interaction between the 1C:Enterprise server and MS SQL will always occur via the network interface, even if both services are on the same operating system. Which, accordingly, will lead to slower operation of 1C in comparison with the classic version of the architecture recommended by 1C itself - the use of Shared Memory. In principle, this obstacle can be “bypassed” using, for example, MS SQL Log Shipping technology. However, in this case, switching to a backup SQL server will no longer be automatic and this option cannot be considered a full-fledged cluster.
• A SQL cluster requires large budget expenditures. If we are talking about classic clustering of the MS SQL service, a single database storage is required, connected to the main and backup SQL servers. Typically, this role is played by expensive storage systems, which increases the budget by an order of magnitude. If we are talking about the newfangled AlwaysOn, then a single database storage is not required; the technology works with local disks of the main and backup servers over the network. But you need a version of SQL Server Enterprise, the license for which costs 4 times more than a regular SQL Server StandarD.
• Number of licenses. Despite the fact that the second SQL server does not process data and is in reserve, licenses will need to be purchased for both servers - both the main and the backup. Particularly painful for the budget are SQL Server Enterprise licenses for implementing a distributed cluster of AlwaysOn high availability groups.

• There is no need to use cheap custom hardware for such an important service as the accounting system of the entire enterprise. The price in this case directly determines the quality, stability and durability of such a platform.
• When choosing a server platform, we recommend paying attention to the presence of two power supplies, a remote IPMI card and the manufacturer’s brand. Of course, everyone chooses a solution based on their budget; top brands are sometimes too expensive and not entirely appropriate, but you shouldn’t skimp on the manufacturer, this can lead to uncontrollable force majeure when working with 1C. Personally, we use Supermicro server platforms in combination with Intel server CPUs.
• There is an opinion, confirmed by practice, that 1C performance depends more on the higher frequency of the CPU than on the number of cores provided.
• There is no need to save on the amount of RAM allocated for the 1C server and SQL service. RAM is currently a fairly cheap resource, and its shortage (even by 10-15 percent) will lead to a significant drop in the performance of the 1C system, because a slower swap system will turn on. Plus, swap will put additional load on the disk subsystem, which will worsen the situation even further.
• The EFSOL company offers comprehensive services for selecting a 1C server, which includes: 1C server design, purchase, configuration and maintenance.
• An alternative to creating your own 1C server is to rent a 1C server. Cloud technologies make it possible to obtain a reliable, fault-tolerant service for comfortable work in 1C at low monthly costs.

System integration. Consulting

In any organization where the number of 1C 8.3 (or 8.2) users is 10 or more, for large volumes of data it is recommended to use the client-server option. This option is based on the use of a third-party DBMS, for example, MS SQL server. Naturally, it is difficult to imagine a client-server mode without a separate server. But each company is unique, each has its own needs, and therefore the choice of server must be approached responsibly. In this article we will try to answer the question of how to choose a 1C server - both software and hardware. Choice is a very important point in the development of a company's information system.

Without software, any computer is useless. High-quality software is especially important in server equipment. It must meet the latest safety and reliability parameters. The 1C client application is multi-platform and available on almost all operating systems, including mobile systems. The server application supports two platforms - Linux and Windows.

There are five options for the DBMS with which the 1C platform works:

Get 267 video lessons on 1C for free:

• built-in DBMS of 1C 8.3 itself, the so-called file mode. The simplest version of work cannot boast of high security. Works on Windows and Linux OS. The database size limit is about 6-10 gigabytes;
• MS SQL Server- the best DBMS for 1C available on the market. According to many experts, SQL Server is generally the best software product from Microsoft. Requires Windows OS to operate;
• IBM DB2 Universal Database is a fairly reliable and secure DBMS management system. Its peculiarity lies in some nuances of information processing and the operation of system methods (for example, sensitivity to the case of string data). The quality of work is significantly influenced by the skills and knowledge of the administrator. Supports Windows, Mac OS X, Linux;
• Oracle Database- a versioned DBMS, which in some cases provides increased performance. Supports Windows, Mac OS X, Linux;
• PostgreSQL- also versioned. The most important advantage is the free distribution of the program. The speed of work is greatly influenced by the qualifications of the administrator. Recommended for a small number of users. Works on Windows, Mac OS X, Linux.

Choosing hardware for 1C

Unlike software, choosing hardware is not as easy. Let's consider the choice of server components for different numbers of users. The number of users is an abstract concept; average numbers for document flow are taken. When selecting equipment, be sure to take into account the volume of paperwork.

Up to 10 users

• CPU: Intel Core i3 or Intel Xeon E3-12xx.
• RAM: 4 gigabytes, which includes 2 GB for the operating system and 2 gigabytes for the DBMS cache.
• Disk subsystem
• Network interfaces

Server from 10 to 40

• CPU: analogue of Intel Xeon E3-12xx or AMD Opteron 4xxx.
• RAM: usually 8-12 gigabytes are enough.
• Disk subsystem: Ideally, a combination of SSD + HDD is desirable. But if this is not possible, you can make do with an HDD.
• Network interfaces: Usually all server applications are installed on one machine.

from 40 to 70

• CPU
• RAM: 16 gigabytes, or better 32.
• Disk subsystem: A traditional array of HDD SAS 15K rpm is sufficient.
• Network interfaces: If the servers are on different machines, use a network with a bandwidth of 10 Gb.

from 70 to 120

With so many users, it makes sense to distribute server applications onto separate server machines.

• CPU: Intel Xeon E5-26xx or AMD Opteron 62xx.
• RAM: from 32 gigabytes.
• Disk subsystem: RAID 10 of reliable server SSDs with a mandatory hardware RAID controller.
• Network interfaces: It is advisable to connect a chain of servers into a network with a bandwidth of 10 Gb. It is recommended to move the index files to a separate SSD, the TempDB temporary table to 1-2 (RAID 1).

from 120 users