Comparison of SCSI, SAS and SATA interfaces. Difference between SAS and SATA sas connectors

In modern computer systems, SATA and SAS interfaces are used to connect the main hard drives. As a rule, the first option suits home workstations, the second - server ones, so the technologies do not compete with each other, meeting different requirements. The significant difference in cost and storage space leaves users wondering how SAS differs from SATA and looking for trade-offs. Let's see if it makes sense.

SAS(Serial Attached SCSI) is a serial storage device interface designed around parallel SCSI to execute the same instruction set. Used primarily in server systems.

SATA(Serial ATA) is a serial data exchange interface based on parallel PATA (IDE). It is used in home, office, multimedia PCs and laptops.

If we talk about HDD, then, despite the differing technical characteristics and connectors, there are no cardinal discrepancies between the devices. Backward one-sided compatibility makes it possible to connect disks to the server board both one by one and via the second interface.

It is worth noting that both connection options are real for SSDs, but the significant difference between SAS and SATA in this case will be in the cost of the drive: the first can be ten times more expensive with a comparable volume. Therefore, today such a decision, if not uncommon, then it is sufficiently balanced, and is intended for fast corporate-level data centers.

Comparison

As we already know, SAS is used in servers, SATA in home systems. In practice, this means that many users simultaneously access the former and solve many problems, while the latter is dealt with by one person. Accordingly, the server load is much higher, so the disks must be sufficiently fault-tolerant and fast. The SCSI protocols (SSP, SMP, STP) implemented in SAS allow more I / O to be processed concurrently.

Directly for the HDD, the access speed is determined primarily by the spindle rotation speed. For desktop systems and laptops, 5400 - 7200 RPM is necessary and sufficient. Accordingly, it is almost impossible to find a SATA drive with 10,000 RPM (unless you look at the WD VelociRaptor series, designed, again, for workstations), and everything above is absolutely unattainable. SAS HDD spins at least 7200 RPM, 10000 RPM can be considered a standard, and 15000 RPM is a sufficient maximum.

Serial SCSI drives are considered to be more reliable and have higher MTBF. In practice, more stability is achieved through the checksum function. SATA drives, on the other hand, suffer from "silent errors" when data is partially written or damaged, which leads to the appearance of bad sectors.

The main advantage of SAS - two duplex ports, allowing one device to be connected via two channels - also works for the fault tolerance of the system. In this case, the exchange of information will be carried out simultaneously in both directions, and reliability is ensured by the Multipath I / O technology (two controllers insure each other and share the load). The queue of marked commands is built up to 256 deep. Most SATA drives have one half-duplex port, and the NCQ queue depth is no more than 32.

The SAS interface assumes the use of cables up to 10 m long. Up to 255 devices can be connected to one port through expanders. SATA is limited to 1m (2m for eSATA), and only supports one point-to-point connection.

Prospects for further development - what the difference between SAS and SATA is also felt quite sharply. The SAS interface reaches 12 Gbps throughput, and manufacturers are announcing support for 24 Gbps data rates. The latest revision of SATA stopped at 6 Gb / s and will not evolve in this regard.

SATA drives have a very attractive price tag in terms of the cost of 1 GB. In systems where the speed of data access is not critical, and the amount of stored information is large, it is advisable to use them.

table

SAS	SATA
For server systems	Mainly for desktop and mobile systems
Uses the SCSI command set	Uses the ATA command set
Minimum spindle speed of HDD 7200 RPM, maximum - 15000 RPM	Minimum 5400 RPM, maximum 7200 RPM
Supports checksum verification technology when writing data	Large percentage of errors and bad sectors
Two duplex ports	One half duplex port
Multipath I / O Supported	Point-to-point connection
Command queue up to 256	Command queue up to 32
Cables up to 10 m can be used	Cable length no more than 1 m
Bus bandwidth up to 12 Gb / s (in the future - 24 Gb / s)	6 Gb / s throughput (SATA III)
The cost of drives is higher, sometimes significantly	Cheaper in terms of price per GB

In this article, we take a look into the future of SCSI and take a look at some of the advantages and disadvantages of SCSI, SAS, and SATA.

In fact, the question is a little more complex than simply replacing SCSI with SATA and SAS. Traditional parallel SCSI is a tried and tested interface that has been around for a long time. Currently, SCSI offers a very fast data transfer rate of 320 Megabytes per second (Mbps) using the state-of-the-art Ultra320 SCSI interface. In addition, SCSI offers a wide range of capabilities, including Command-Tag Queuing (a method of optimizing I / O commands to increase performance). SCSI hard drives are reliable; over a short distance, you can daisy chain up to 15 devices connected to a SCSI link. These features make SCSI a great choice for productive desktops and workstations, up to and including enterprise servers, today.

SAS hard drives use the SCSI command set and are similar in reliability and performance to SCSI drives, but use a serial version of the SCSI interface at 300 MB / s. Although slightly slower than 320MB / s SCSI, the SAS interface is capable of supporting up to 128 devices over longer distances than the Ultra320, and can expand up to 16,000 devices per channel. SAS hard drives offer the same reliability and rotational speeds (10,000-15,000) as SCSI drives.

SATA drives are slightly different. Where SCSI and SAS drives prioritize performance and reliability, SATA drives sacrifice them in favor of dramatic increases in capacity and cost savings. For example, a SATA drive currently has a capacity of 1 terabyte (TB). SATA is used where maximum capacity is needed, such as data backup or archiving. SATA now offers point-to-point connections at speeds up to 300 Mb / s, and easily outperforms the traditional parallel ATA interface at 150 Mb / s.

So what happens with SCSI? It works great. The problem with traditional SCSI is that it is just nearing the end of its useful life. The parallel SCSI interface, which has a speed of 320 Mb / s, will not be able to perform significantly faster on current SCSI cable lengths. In comparison, SATA drives will hit 600 MB / s in the near future, SAS has plans to hit 1200 MB / s. SATA drives can also work with a SAS interface, so these drives can be used simultaneously in some storage systems. The potential for increased scalability and data transfer performance far exceeds that of SCSI. But SCSI is not going to leave the scene anytime soon. We will be seeing SCSI in small and medium servers for a few more years. As the hardware is updated, SCSI will be systematically replaced with SAS / SATA drives for faster and easier connections.

High-performance server drives for mission-critical applications rarely come to the attention of IT publications. No wonder, because we are more focused on the mass customer than on system administrators and server hardware vendors. Meanwhile, testing server HDDs is even more important than testing desktop HDDs, for several reasons. First, due to the higher cost of drives and the higher sensitivity of server tasks to performance. After the massive proliferation of solid-state drives, the differences between desktop drives have ceased to be of great importance, and in a server, replacing an HDD with an SSD is far from always advisable. The following circumstance follows from the first: HDD for a desktop or home NAS can be easily selected according to its basic technical characteristics (volume, spindle speed, plate capacity). In the case of a server HDD, a lot depends on the optimization of the firmware, which manifests itself in a complex load and, accordingly, requires special tests to catch these features. Finally, at large scales, the performance-to-power ratio of the drive comes into play.

Choosing enterprise hard drives has definitely become easier over the past few years. Fiber Channel and SCSI models have ceased to be produced. The drives are divided into two classes: models in the 3.5-inch form factor are limited to a rotational speed of 7200 rpm, have a SAS or SATA interface - you can choose from and are designed for storing "cold" data (nearline storage). Drives with a speed of 10,000-15,000 rpm use the SAS interface and most of them have moved to the 2.5-inch form factor (SFF - Small Form Factor), which allows you to increase the number of spindles per unit in the rack. Only HGST still has 15K drives in the 3.5-inch form factor with Fiber Channel ports.

We are already constantly paying attention to nearline disks in SATA configuration, but the SAS / SCSI test is published for the first time on 3DNews.

⇡ Test participants

The following devices took part in the comparison:

• HGST Ultrastar C10K1800 1.8 TB (HUC101818CS4200);
• HGST Ultrastar C15K600 600 GB (HUC156060CSS200);
• Seagate Savvio 10K.6 900 GB (ST900MP0006);
• Seagate Enterprise Performance 10K HDD v7 1.2 TB (ST1200MM0017);
• Seagate Enterprise Performance 15K HDD v5 600 GB (ST600MP0035);
• Toshiba AL13SEB 900 GB (AL13SEB900);
• Toshiba AL13SXB 600 GB (AL13SXB600N);
• WD VelociRaptor 1TB (WD1000DHTZ).

Unlike desktop and NAS hard drives, SAS drives are not that different from each other. All participants:

a) are available in a 2.5-inch form factor with a thickness of 15 mm;

b) have two SAS ports to increase fault tolerance;

c) prepared for 24/7 operation in a telecommunications rack;

d) allow the user to configure the sector size for recording additional metadata;

e) are characterized by the same reliability indicators (MTBF, number of head parking cycles);

f) Sold with a five-year manufacturer's warranty.

Models of the maximum volume in the corresponding rulers were selected for testing. The products of all companies that produce HDDs are presented, with one exception. We have exhausted all the possibilities to get a WD Xe drive for a test (except for just buying it for a lot of money), and recently this brand disappeared from the corporate website of Western Digital - apparently, it is being discontinued. As a result, of all drives with a spindle rotation speed of 10-15 thousand rpm, WD has only VelociRaptor - in fact, a derivative of WD Xe, but with a SATA interface. We included VelociRaptor as a contributor so that WD was somehow represented in the review. Of course, it cannot be considered a 100% replacement for SAS drives, but a lot of servers run on SATA drives, so the VelociRaptor can also be used. In addition, looking from the other side, any of the SAS drives can be used in a workstation with the appropriate Host Bus Adapter (HBA) instead of the VelociRaptor, which also justifies this drive's participation in today's test.

Manufacturer	HGST	HGST	Seagate	Seagate	Seagate	Toshiba	Toshiba	Western digital
Series	Ultrastar C10K1800	Ultrastar C15K600	Savvio 10K.6	Enterprise Performance 10K HDD v7	Seagate Enterprise Performance 15K HDD v5	AL13SEB	AL13SXB	VelociRaptor
Model number	HUC101818CS4200	HUC156060CSS200	ST900MM0006	ST1200MM0017	ST600MP0035	AL13SEB900	AL13SXB600N	WD1000CHTZ / WD1000DHTZ
Form factor	2.5 inch	2.5 inch	2.5 inch	2.5 inch	2.5 inch	2.5 inch	2.5 inch	3.5 / 2.5 inches
Interface	SAS 12 Gbps	SAS 12 Gbps	SAS 6 Gb / s	SAS 6 Gb / s	SAS 12 Gbps	SAS 6 Gb / s	SAS 6 Gb / s	SATA 6Gb / s
Dual-port	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No
Capacity, GB	1 800	600	900	1 200	600	900	600	1000
Configuration
Spindle rotation speed, rpm	10 520	15 030	10 000	10 000	15 000	10 500	15 000	10 000
Data recording density, GB / plate	450	200	300	300	200	240	ND	334
Number of plates / heads	4/8	3/6	3/6	4/8	3/6	4/8	ND	3/6
Buffer size, MB	128	128	64	64	128	64	64	64
Sector size, bytes	4096-4224	512-528	512-528	512-528	4096-4224	512-528	512-528	512
Performance
Max. sustained sequential read speed, MB / s	247	250	195	195	246	195	228	200
Max. sustained sequential write speed, MB / s	247	250	195	195	246	195	228	200
Burst rate, read / write, MB / s	261	267
Internal data transfer rate, MB / s	1307-2859	1762-3197	1440-2350	1440-2350	ND	ND	ND	ND
Average seek time: read / write, ms	3,7/4,4	2,9/3,1	ND	ND	ND	3,7/4,1	2,7/2,95	ND
Track-to-track seek time: read / write, ms	ND	ND	ND	ND	ND	0,2/22	ND	ND
Full stroke seek time: read / write, ms	7,3/7,8	7,3/7,7	ND	ND	ND	ND	ND	ND
Reliability
MTBF (mean time between failures), h	2 000 000	2 000 000	2 000 000	2 000 000	2 000 000	2 000 000	2 000 000	1 400 000
AFR (annualized failure rate),%	ND	0,44	0,44	0,44	0,44	ND	0,44	ND
Number of head parking cycles	600 000	600 000	ND	ND	ND	ND	600 000	600 000
physical characteristics
Power consumption: idle / read-write, W	5,4/7,6	5,8/7,5	3,9/7,8	4,6/8,1	5,3/8,7	3.9 / ND	5,0/9,0	4,2/5,8
Typical noise level: idle / searching	34/38 dBA	32/38 dBA	30 dBA / ND	31 dBA / ND	32.5 / 33.5 dBA	30 dBA / ND	33 dBA / ND	30/37 dBA
Maximum temperature, ° C: disk enabled / disk disabled	55/70	55/70	60/70	60/70	55/70	55/70	55/70	55/70
Shockproof: Disk On (Read) / Disk Off		30 g (2 ms) - write / 300 g (2 ms)	25 g (2 ms) / 400 g (2 ms)	25 g (2 ms) / 400 g (2 ms)	25 g (2 ms) / 400 g (2 ms)	100 g (1 ms) / 400 g (2 ms)	100 g (1 ms) / 400 g (2 ms)	30 g (2 ms) / 300 g (2 ms)
Overall dimensions: D × H × G, mm	101 × 70 × 15	100 × 70 × 15	101 × 70 × 15	101 × 70 × 15	101 × 70 × 15	101 × 70 × 15	101 × 70 × 15	101 × 70 × 15/147 × 102 × 26
Weight, g	220	219	212	204	230	240	230	230/500
Warranty period, years	5	5	5	5	5	5	5	5
Average retail price, rubles *	161 000	36 000	20 000	26 900	49 600	17 800	24 100	14 000 / 12 600

⇡ Description of test participants

HGST Ultrastar C10K1800 1.8TB (HUC101818CS4200)

This is the most capacious drive in the newest 10K HGST line. The Ultrastar C10K1800 series is remarkable in several ways. The models ending in S420x achieve a capacity of 450 GB per platter thanks to the high recording density with 4K sector formatting (native or 512-byte sector emulation). Therefore, the disk holds up to 1.8 TB, and the sequential read / write speed has reached the level of a 15 thousand rpm HDD class.

The rest of the lineup consists of 512-528 byte disks with less outstanding performance and up to 1.2 TB in size.

All models in the C10K1800 line have a so-called media cache. In several places on the surface of the plates, areas are highlighted that serve as a non-volatile cache. Instead of carrying data to the requested sector, the write head of the disk dumps it to the nearest caching area, and in the idle disk, it is moved to the desired location.

By the way, this is the most expensive disc in the test, just fantastically expensive - an average of 161 thousand rubles in Moscow online stores. And in America, by the way, it is much cheaper - $ 800 at newegg.com.

HGST Ultrastar C10K1800 1.8TB (HUC101818CS4200)

HGST Ultrastar C15K600 600 GB (HUC156060CSS200)

The only line of 2.5-inch discs with a spindle speed of 15 thousand rpm in the HGST range. Ultrastar C15K600 drives simultaneously have the maximum sequential read / write speed and low latency. Physical formatting of platters is performed in 512-528 or 4096-4224 byte sectors (with native access or 512 byte emulation). The most capacious model in the line - 600 GB with 4 KB sectors - participates in testing.

HGST Ultrastar C15K600 600 GB (HUC156060CSS200)

Seagate Savvio 10K.6 900GB (ST900MP0006)

These are pretty old drives - the generation before last compared to the current Enterprise Performance 10K line from Seagate. Therefore, the performance of Savvio 10K.6 is no longer the leading one in this class. Plates are formatted in 512-528 byte sectors. However, these disks are still on sale, have a good volume (up to 900 GB) and are relatively inexpensive.

Seagate Savvio 10K.6 900GB (ST900MP0006)

Seagate Enterprise Performance 10K HDD v7 1.2TB (ST1200MM0017)

This series also managed to formally become outdated by the time the test was released, giving way to Enterprise Performance 10K HDD v8. These drives differ from Savvio 10K.6 only in the increased capacity to 1.2 TB, but this was achieved by increasing the number of platters, not the recording density, so there is no difference with the previous generation in terms of the declared performance. The tested model ST1200MM0017 has built-in encryption.

Seagate Enterprise Performance 10K HDD 1.2TB (ST1200MM0007)

Seagate Enterprise Performance 15K HDD v5 600 GB (ST600MP0035)

This is the current line of Seagate drives with a spindle speed of 15 thousand rpm. Disks have sector markings of 512-528 or 4096-4224 bytes (natively or with 512-byte emulation). The maximum volume (600 GB) drive with 4 KB sectors has been tested.

Seagate Enterprise Performance 15K HDD 600GB (ST600MP0035)

Toshiba AL13SEB 900 GB (AL13SEB900)

In terms of its main characteristics, it is an analogue of Seagate Savvio 10K.6: 10,000 rpm, volume up to 900 GB, formatting in 512-528 byte sectors. Toshiba does not offer any drives with built-in encryption in this series.

Toshiba AL13SXB 600 GB (AL13SXB600N)

In this series of disks with a spindle speed of 15,000 rpm, models with the names of the type AL13SXB ** 0N are formatted with a sector size of 512-528 bytes. We took the eldest of them for testing. Models named AL13SXB ** E * use 4K sectors and also support 12 Gb / s SAS. There is no built-in encryption in the entire AL13SXB series.

Toshiba 900 GB (AL13SEB900)

WD VelociRaptor 1 TB (WD1000CHTZ / WD1000DHTZ)

Physically, VelociRaptor differs little from its prototype - WD Xe: the same 10,000 rpm and almost the same linear performance. VelociRaptor uses Advanced Format markup (4 KB sectors), and the volume available to the user is higher than that of similar WD Xe (1 TB in the case of the older model).

Since this is a SATA drive, it is not functionally a complete analogue of SAS drives. In particular, you can forget about dual-port connectivity, sector size configuration and built-in encryption. In addition, SAS drives are usually made more reliable, which is noticeable when comparing their declared MTBF value with that of VelociRaptor. And yet, from the performance standpoint, this drive can be regarded as a server 10K for the poor. There are varieties of "lizard" with a heatsink adapter to the 3.5 "form factor (DHTZ), as well as" naked "2.5" versions (CHTZ).

WD VelociRaptor 1TB (WD1000DHTZ)

⇡ Testing methodology

Isolated performance tests

Performed using Iometer 1.1.0. The volume and data transfer rate are indicated in binary units (1 KB = 1024 bytes). Block boundaries are aligned relative to the 4KB markup.

1. Sequential read / write of data in 128 KB blocks with a request queue depth of 256.
2. Random read / write blocks from 512 bytes to 2 MB with a request queue depth of 256.
3. Mixed read / write of 128 KB blocks with a request queue depth of 256. The share of read and write operations varies from 0 to 100% in 10% increments.
4. The dependence of the throughput on the length of the command queue. Blocks of 4 KB are read, the depth of the request queue varies from 1 to 256 with a power-of-two step. A similar test for writing blocks is not performed, since hard drives are not distinguished by this parameter.
5. Settled response time. Random read / write of 512-byte blocks with a request queue depth of 1. The test continues for 10 minutes.
6. Constancy of response time. Random read / write of 4 KB blocks with a request queue depth of 256 is performed. For each test segment of 1 s duration, the average and maximum response times are recorded, on the basis of which: a) average values ​​of both indicators are calculated; b) standard deviation of the mean response time.
7. Multi-threaded read / write. Four threads are created that perform sequential read / write of 64 KB blocks with a request queue depth of 1. The threads have access to non-overlapping 100 GB address spaces, which are located in the disk space close to each other, starting from sector zero. The total throughput of all streams is measured, as well as each of them separately.

Emulated load tests

Performed in Iometer 1.1.0. The volume and data transfer rate are indicated in binary units (1 KB = 1024 bytes). Block boundaries are aligned relative to the 4KB markup. The command queue depth is 256.

Block size	Share of all requests	Read share	Random access share
Database
8 Kbytes	100%	67%	100%
File server
512 bytes	10%	80%	100%
1 KB	5%	80%	100%
2 Kbytes	5%	80%	100%
4 KB	60%	80%	100%
8 Kbytes	2%	80%	100%
16 kB	4%	80%	100%
32 kB	4%	80%	100%
64 kB	10%	80%	100%
Work station
8 Kbytes	100%	80%	80%
Web server
512 bytes	22%	100%	100%
1 KB	15%	100%	100%
2 Kbytes	8%	100%	100%
4 KB	23%	100%	100%
8 Kbytes	15%	100%	100%
16 kB	2%	100%	100%
32 kB	6%	100%	100%
64 kB	7%	100%	100%
128 kB	1%	100%	100%
512 kB	1%	100%	100%

Test bench

The drives were connected to the LSI SAS 9211-8i adapter, for which we would like to express our gratitude to the Russian representative office of LSI.

⇡ Performance, basic tests

Sequential read / write

• Disks with a spindle rotation of 15 thousand rpm reign supreme in the sequential read / write test. However, this group has its own leader - Seagate Enterprise Performance 15K HDD v5.
• Ultrastar C10K1800, due to its high recording density, is not inferior to 15K category drives.
• But the presented 10K models differ little in terms of linear access speed.

Random read

• 15-thousanders in this discipline also dominate their slow-speed counterparts.
• The scatter of indicators within HDD categories with the same spindle rotation speed is small. Only HGST Ultrastar C15K600 can be distinguished as a formal leader in its group and VelociRaptor, which is clearly inferior to its peers.

Arbitrary recording

The results of the random write test turned out to be less predictable than in the previous test, since they are determined not only by the mechanics of the HDD, but also by the nature of the buffer usage.

• HGST Ultrastar C15K600 demonstrated colossal performance, completely unattainable for competing devices.
• The two remaining 15K drives also have a big advantage over HDDs with a lower spindle speed.
• The 10-thousanders themselves make up a homogeneous group, with the exception of Ultrastar C10K1800. It goes far beyond its class and is second only to the C15K600 from the same manufacturer. Here it is, the vaunted media cache in action!

Settled response time

• Even though the load continues for 10 minutes, it may not completely fill the buffer on some disks, so the results for writing data do not reflect what this test is aimed at - the latency of the drive mechanics.
• On the other hand, when reading with a queue length of one command, the buffer is not a helper. As a result, the rivals lined up in accordance with the spindle speed (the higher it is, the shorter the response time). No significant difference was found between devices of the same category.

⇡ Performance, advanced analysis

Mixed read / write

• The 15K drives are still up to the mark, with the exception of the Ultrastar C15K600, which sagged especially under mixed loads.
• Ultrastar C10K1800 once again stood out among its counterparts. Among other 10K models, we would like to point out Toshiba AL13SEB. They are all roughly the same at 100% read or write, but AL13SEB retains the best performance under mixed workload.

Bandwidth versus command queue length

• All drives are able to benefit from long command queues and peak throughput at 64 commands. Only VelociRaptor is content with a queue of 32 teams.

Multithreaded read

• Most of the test participants distribute resources evenly among the four threads. Which, however, leads to low aggregate productivity.
• Toshiba AL13SEB and WD VelociRaptor, on the other hand, sacrifice one of the streams during multi-threaded reading, thereby increasing the data transfer rate in the rest and the overall throughput.

Multithreaded recording

• When writing to four streams, none of the disks cheats: the performance is evenly distributed among all streams.
• As you can see, not so much depends on the disc mechanics in this test. 15K models from Seagate and Toshiba took the first places, but Ultrastar 15K600 is an obvious outsider.

Consistent response time

• When reading data, all drives are characterized by a significant difference between the average and maximum response times. Only the VelociRaptor stands out for its better average to maximum response time.
• When recording, the peak response times are smoothed by the buffer and differ little from the average.

• Test participants differ the most in the range of write access times. Ultrastar C10K1800 has the most consistent performance. The Toshiba AL13SEB900, on the other hand, has a dramatically increased access time standard deviation.

Among 10K server drives, the drives do not differ so much from each other, but formally - the best results are achieved by Seagate Savvio 10K.6. VelociRaptor, on the other hand, always lags behind.

Most 10K models are similar to each other in basic aspects, but it is worth highlighting the HGST Ultrastar C10K1800 (HUC101818CS4200), which is inferior to more resourceful colleagues of the 15K class only in random read speed and at the same time has a record volume of 1.8 TB. However, these advantages did not affect the results of tests with emulated applications.

Seagate Savvio 10K.6 900GB (ST900MP0006) and Seagate Enterprise Performance 10K HDD v7 1.2TB (ST1200MM0007) deliver consistently high performance without surprises. Toshiba AL13SEB900 performed a little worse than other 10K models.

WD VelociRaptor 1TB (WD1000DHTZ) can be considered as a high-performance HDD "for the poor" if the SAS protocol is not required in the terms of reference. In terms of its characteristics, this is a typical 10K class disk, only in comparison with true server drives, the random read speed leaves much to be desired, which is also evident in the "emulators".

What is SAS, the Background It's time to acknowledge the obvious fact that the SCSI standard, even in the most modern implementations like Ultra320 SCSI, has exhausted its capabilities. At the very least, further scaling of its performance, if theoretically possible, will be very expensive. The situation with this highly respected standard looks especially depressing against the background of the rapid development of all computer technology and the architecture and topology of data storage systems in particular.

Two key factors that are pushing manufacturers to improve the interfaces of hard drives are the growing performance of storage systems with a large number of processed transactions and the speed of data retrieval from large libraries. Of course, "a holy place is never empty," and the emergence of interfaces like optical FCAL or serial SATA to some extent eliminated bottlenecks and added variety to the list of storage architectures. However, users accustomed to the capabilities of SCSI are still fans of this standard. Moreover, a lot of money has been invested in its development.

These are the preconditions for the emergence of a new industrial standard called Serial-Attached SCSI, or simply SAS.


For the sake of fairness, it should be noted that the new standard did not appear suddenly and immediately: the official announcement of SAS technology, which took place on January 28, 2004, was preceded by a serious work of a development team from different companies and industrial groups - the SCSI Trade Association (STA) and the International Committee for Information Technology Standards (INCITS), sponsored by the American National Standards Institute (ANSI). The new standard was first discussed in December 2001, when the board of directors of the SCSI Trade Association (STA) voted to define the Serial Attached SCSI specifications. Further, on May 2, 2002, the development of the standard was transferred to the T10 committee of INCITS (InterNational Committee for Information Technology Standards) created specifically to support, develop and promote SAS, and the first draft SAS specifications were published in mid-2003.

So, the most important thing to rely on when trying to formulate a definition of the SAS standard: Serial-Attached SCSI is a logical and natural sequential extension of the parallel SCSI interface technology used to connect peripherals to computers.
From this, for a start, we will push off.

SAS purpose

To determine the purpose of the SAS standard and its place among modern peripheral interfaces, we turn to the wording set forth in the Serial Attached SCSI FAQ on the T10 website.

Serial Attached SCSI is a logical evolution of modern interfaces and is designed for use in industrial data collection and storage centers. The SAS standard builds on the electrical and physical characteristics of the Serial ATA interface to provide scalability, performance, reliability, and data manageability in servers and storage subsystems. The architectural similarity with SATA does not prevent SAS from possessing the most demanded features of SCSI, at the same time getting rid of its disadvantages: large connectors, short length of connecting cables, limited performance and addressing.

In a broad sense, SAS is a kind of full-duplex SATA with support for two ports, high addressing capabilities, enhanced reliability, performance and logical compatibility with SCSI. Serial ATA, on the other hand, can be thought of as a simplified subset of Serial Attached SCSI for simple systems without critical reliability and performance requirements. This does not mean that Serial Attached SCSI devices cannot be used in conventional workstations and desktop PCs, only a suitable host adapter is required.

In fact, Serial Attached SCSI is SCSI, but not with the usual parallel, but with a point-to-point serial architecture, with direct connection of the controller to the drives. SAS supports up to 128 drives of various types and sizes, connected together with thinner and longer (than in the case of SCSI) cables. While SCSI pushes data through its wires at a rate of about 20 MB / s, and half-duplex SATA of the first generation - 1.5 GB / s in one direction per unit of time, full-duplex SAS signaling serial interface with hot-plug support in the current implementation provides data exchange at speeds up to 3.0 Gb / s per port.

The key difference between SAS and SCSI is the ability to connect SAS drives to two different ports simultaneously, each representing a different SAS domain. You can imagine how significant this impacts on storage reliability and system resiliency. In addition, the "switch" nature of the SAS architecture allows, in theory, to connect thousands of drives "casually" (up to 16384 drives without performance degradation!), Which makes the scalability of such systems theoretically unlimited. The main differences between SCSI and SAS technologies are shown in the table below.

SAS Connector and Cable Specifications

One of the key features of the SAS interface during its development was the possibility of a significant increase in the speed of data exchange. The next-generation SAS specifications currently under development involve data transfer rates up to 6.0 GB / s with full compatibility with the first generation of SAS devices. The next generation has not yet been seriously considered, but there is talk of the possibility of achieving data exchange rates up to 12 GB / s.

When developing connectors for SAS devices, a promising increase in the speed of data exchange was laid, and at the same time, the experience of miniaturization, seen in the SATA specifications, was taken into account. The specificity of the connector lies in the placement of the second data port, since each of the ports of the SAS device is located in different domains and serves to organize independent paths from one SAS device to another to ensure trouble-free operation. If one of the drives in the chain fails, this in no way affects the operation of other devices. Thus, the SAS peripheral connector design was born, in fact, having an architectural similarity to 68-pin connectors for drives with a classic parallel SCSI or SCA-2 interface, but at the same time, by analogy with SATA, which supports hot-plugging. "and reliable contact.

SAS cabling is much more compact than parallel ATA and SCSI cabling, resulting in less confusion and better airflow around the components inside the chassis. Typical lengths of SAS interface cables for applications such as workstations do not exceed 1 m, the maximum length of such a cable can be up to 8 m.In theory, this is comparable to the cable length for the SCSI interface, since some modern devices allow a connection between the host controller and SCSI -peripherals at a distance of more than 8 m. However, in case of need, the distance between SAS-devices can be significantly increased due to the so-called SAS-expanders - a kind of "pipeline pumping stations".

It is interesting to note that when developing the SAS specifications, the working group immediately took into account the need to determine the parameters of connectors and cables not only for internal, but also for external connections, similar to modern SCSI options like "server-JBOD system". For the SATA interface, the adoption of such specifications was postponed, and as a result, the development of External SATA is still not complete.

As for external SAS connections, the basis was taken from Infiniband's proposal, where external connectors and cabling are designed for 4 devices and at the same time provide the performance of the first generation of external SAS connections at the level of 1.2 GB / s in each direction, that is up to 2400 MB / s in full duplex mode! Agree, more than impressive for the front-end.

SAS system topology

The use of point-to-point configurations allows obtaining high throughput, however, the reverse side of the coin is the organization of a specific topology, where the interaction of initiating (host) devices and peripherals implies support for more than two devices "in a bundle". During the development of the SAS standard, the specification immediately laid down the existence of inexpensive expanders that allow you to create systems with more than one initiating host, with support for more than one peripheral device.

Another important goal set by the developers of the new standard is to get away from the limitation of the classic SCSI, which implies no more than 16 devices in one chain. As a result, each SAS system, with the appropriate number of expanders, is capable of addressing up to 16256 devices in a single SAS domain. It is worth noting the flexibility of the configuration of SAS expanders: their specifications imply the creation of heterogeneous systems, where both SAS and SATA devices can coexist as peripheral drives. Agree, it is very convenient, especially when creating budget storage systems or devices with future-oriented scaling.

Illustration for the SAS Domain Organization Principle
maximum capacity

Pay attention to the illustration above: the dark green module in the center represents the fanout expander. Such a "switching" expander can be present in one SAS domain in a single quantity and combine up to 128 SAS devices. However, SAS devices should not be understood exclusively as hard disks, since here we mean any possible combination of so-called "edge expanders" (light green modules), initiating devices and the actual drives. Peripheral expanders, in turn, can also support up to 128 SAS devices, however, no more than one additional expander can be connected to them. Initiators (hosts) are marked with blue modules on the diagram, and SAS or SATA drives are marked with brown cylinders.

SAS Protocols

The creation of a new topology and new interfaces led to the creation of a completely new definition of how to address all possible ports in the SAS domain. With parallel SCSI, of course, everything is easier, since the addressing of all devices in the domain is predefined at the hardware level.

As a result, the working group for the development of the SAS protocol decided to select globally unique 64-bit names - WWN (WorldWide Name) for all types of SAS devices - as identifiers. Again, nothing new under the Moon, this is the kind of addressing that has long been used when naming Fiber Channel devices.

Thus, at the moment of power-up, all devices combined into a single SAS space exchange their WWNs with each other, and only after that the set of SAS devices becomes a "meaningful" SAS system. Adding a new device to the SAS system (in this case, adding means just "hot plugging") or removing it from the system results in a notification that notifies all initiators of the event and allows the system to be adjusted to the new configuration. Expanders, in turn, are responsible for "issuing" WWNs to all SATA devices in the system, both in the case of turning it on and in the event of a "hot" connection of a new device. Upon completion of the system initialization process, SATA devices interact using SATA protocols; for SAS devices, the SAS protocol is used, described in other SCSI standards such as SPI (SCSI Parallel Interface).

Further, everything is simpler: the exchange of commands, data, statuses and other information between SAS devices is performed by packets, the specifications of which are very similar to the characteristics of packets for the exchange of information when working with parallel SCSI or Fiber Channel devices. The format of SAS data packets, called "frames", is especially similar to the Fiber Channel specifications: each of them consists of command descriptor blocks (CDBs) and other SCSI constructs defined by other SCSI standards such as the SCSI Primary Command Set or SCSI Block Command. Here's another benefit of the SAS standard: the use of a SCSI-like protocol and architecture allows you to combine SAS constructs with other storage and data processing systems with Infiniband, iSCSI or Fiber Channel architecture, which, in fact, are also SCSI objects.

The SAS protocol contains four traditional layers: the phy layer, the link layer, the port layer, and the transport layer. The aggregation of the four layers in each SAS port means that the programs and drivers used to work with the parallel SCSI ports can be used equally well to serve the SAS ports with only minor modifications.

SAS architecture

Application layers, including drivers and applications themselves, create specific tasks for the transport layer, which, in turn, encapsulates commands, data, statuses, etc. in SAS frames and delegates their transfer to the port layer. Of course, the transport layer is also responsible for receiving SAS frames from the port layer, disassembling the received frames, and transferring the content to the application layer.

The SAS port layer is responsible for exchanging data packets with the link layer in order of establishing connections, as well as for choosing the physical layer through which the packets will be transmitted simultaneously to several devices. The SAS physical layer means the corresponding hardware environment - transceivers and encoding modules that connect to the SAS physical interface and send signals over wired circuits.

By the way, let me remind you that at the physical level, connections in the case of a SAS serial interface are full-duplex differential pairs of circuits, which can also be combined to increase performance (well, just like PCI Express) into "wide" ports. Accordingly, each device can have more than one port, and each of them can be configured as "narrow" or "wide". Host and expander interfaces can be composed of multiple ports, with the address of each host available to each peripheral, and the bandwidth being added together. The organization of multiple paths for data passage due to the presence of "wide" ports implies parallel execution of commands and a corresponding reduction in the loss of time waiting for a queue.

Conclusion

The presented material is only a brief introduction to the principles of building the architecture of the SAS interface and the implementation features of this standard. A more detailed examination of the interface specifications will most likely require the release of a whole series of articles on this topic. It is possible that this is exactly how it will be, fortunately, the start of mass implementation of the interface is just around the corner, and the number of applied questions on the implementation of SAS systems will only grow over time.

The main definition of SAS, which, in my opinion, should not be forgotten - the new Serial Attached SCSI interface was designed for the needs of a wide range of enterprise-grade storage systems, however, it is still a "close-action" interface and by no means is intended to replace any network interfaces, there is no need to "buy" a similar implementation of the "point-to-point" architecture.

For all its "sharpening" for work in large and almost infinitely scalable storage systems, the Serial Attached SCSI interface implies full compatibility with relatively inexpensive Serial ATA drives, which allows you to design affordable systems even for small businesses. At the same time, support for 2-port Serial Attached SCSI drives allows for performance levels that have never been dreamed of in today's SCSI-based systems.

For those who are ready to plunge into the study of the features of Serial Attached SCSI on their own, we conclude with a list of sites where educational and standard-setting documents are located.

Adaptec website resources
Maxtor website resources
Seagate website resources

T10:

Serial Attached SCSI -
SCSI Architecture Model - 3 (SAM-3)
SCSI Primary Commands - 3 (SPC-3)
SCSI Block Commands - 2 (SBC-2)
SCSI Stream Commands - 2 (SSC-2)
SCSI Enclosure Services - 2 (SES-2)

SAS Connector Specifications:

SFF 8482 (internal backplane / drive)
SFF 8470 (external 4-wide)
SFF 8223, 8224, 8225 (2.5 ", 3.5", 5.25 "form factors)
SFF 8484 (internal 4-wide)

Serial ATA Specifications:

Serial ATA II: Extensions to Serial ATA 1.0
Serial ATA II: Port Multiplier
Serial ATA II: Port Selector
Serial ATA II: Cables and Connectors Volume 1

Additional resources:

International Committee for Information Technology Standards
T11 (Fiber Channel standards)
SCSI Trade Association
SNIA (Storage Networking Industry Association)

Why SAS?

Serial Attached SCSI is not just a serial implementation of the SCSI protocol. It does much more than just porting SCSI functions like TCQ (Tagged Command Queuing) through a new connector. If we wanted the most simplicity, then we would use the Serial ATA (SATA) interface, which is a simple point-to-point connection between a host and an end device such as a hard drive.

But SAS is based on an object model that defines a "SAS domain" - a data delivery system that can include optional expander and SAS endpoints such as hard drives and host bus adapters (HBAs). from SATA, SAS devices can have multiple ports, each of which can use multiple physical connections to provide faster (wider) SAS connections.In addition, multiple initiators can access any given target, and the cable length can be up to eight meters ( for the first generation SAS) versus one meter for SATA, which understandably offers many opportunities for high-performance or redundant storage solutions, and SAS also supports the SATA Tunneling Protocol (STP), which allows SATA devices to be connected to the SAS controller.

The second generation SAS standard increases the connection speed from 3 Gbps to 6 Gbps. This speed gain is very important for complex environments where high performance is required due to high-speed storage. The new version of SAS also aims to reduce the complexity of cabling and the number of connections per Gbps of bandwidth, increasing the possible cable lengths and improving the performance of expanders (zoning and auto-discovery). Below we will talk about these changes in detail.

Up to 6Gbps SAS Speed

To bring the benefits of SAS to a wider audience, the SCSI Trade Association (SCSI TA) presented a tutorial on SAS technology at the Storage Networking World Conference earlier this year in Orlando, Florida. The so-called SAS Plugfest, which demonstrated 6 Gbps SAS performance, compatibility and features, took place even earlier in November 2008. LSI and Seagate were the first in the market to introduce hardware that is compatible with 6Gb / s SAS, but other vendors should catch up soon too. In this article, we will take a look at the current state of SAS technology and some new devices.

SAS Functions and Basics

SAS Fundamentals

Unlike SATA, SAS operates in full duplex, providing full bandwidth in both directions. As mentioned earlier, SAS connections are always established over physical connections using unique device addresses. In contrast, SATA can only address port numbers.

Each SAS address can contain multiple physical layer interfaces (PHYs), allowing for wider connections via InfiniBand (SFF-8470) or mini-SAS cables (SFF-8087 and -8088). Typically four SAS interfaces, with one PHY each, are combined into one wide SAS interface, which is already connected to the SAS device. Communication can also be done through expanders, which act more like switches than SAS devices.

Features such as zoning now allow administrators to bind specific SAS devices to initiators. This is where the increased SAS 6 Gbps throughput comes in handy, as the four-lane connection will now have twice the speed. Finally, SAS devices can even have multiple SAS addresses. Since SAS drives can use two ports, with one PHY each, the drive can have two SAS addresses.

Connections and interfaces

Click on the picture to enlarge.

SAS connections are addressed over SAS ports using Serial SCSI Protocol (SSP), but lower layer communications from PHY to PHY are done using one or more physical connections for bandwidth reasons. SAS uses 8/10 bit coding to convert 8 bits of data into 10-character transmissions for timing recovery, DC balance and error detection. This results in an effective bandwidth of 300 MB / s for 3Gb / s transfer mode and 600 MB / s for 6 Gb / s connections. Fiber Channel, Gigabit Ethernet, FireWire and others use a similar coding scheme.

SAS and SATA power and data interfaces are very similar to each other. But if SAS has data and power interfaces combined into one physical interface (SFF-8482 on the device side), then SATA requires two separate cables. The gap between the power and data pins (see illustration above) is closed in the case of SAS, which prevents the SAS device from being connected to the SATA controller.

On the other hand, SATA devices can work just fine on SAS infrastructure thanks to STP, or in "native" mode if no expanders are used. STP adds additional latency through the expanders because they need to establish a connection, which is slower than direct SATA connections. However, the delays are still very small.

Domains, expanders

SAS domains can be thought of as tree structures, much like complex Ethernet networks. SAS Expanders can work with a large number of SAS devices, but they use the circuit switched principle rather than the more common packet switching. Some expanders contain SAS devices, others do not.

SAS 1.1 recognizes edge expander, which allows the SAS initiator to associate with up to 128 additional SAS addresses. Only two Edge Expanders can be used in a SAS 1.1 domain. However, one fanout expander can connect up to 128 edge expanders, dramatically increasing the infrastructure capabilities of your SAS solution.

Click on the picture to enlarge.

Compared to SATA, the SAS interface can seem complicated: different initiators access target devices through expanders, which implies laying the appropriate routes. SAS 2.0 simplifies and improves routing.

Remember that SAS disallows loops or multiple paths. All connections must be point-to-point and exclusive, but the connection architecture itself scales well.

New SAS 2.0 Features: Expanders, Performance

	SAS 1.0 / 1.1
Function	Retains legacy SCSI support SATA compatible	Compatible with 3Gbps Improved speed and signal flow Zone management Improved scalability
Storage functions	RAID 6 Small form factor HPC High-capacity SAS drives Ultra320 SCSI Replacement Choice: SATA or SAS Blade servers	RAS (data protection) Security (FDE) Cluster support Support for larger topologies SSD Virtualization External storage 4K sector size
Data rate and cable bandwidth	4 x 3Gb / s (1.2 GB / s)	4 x 6 Gbps (2.4 Gbps)
Cable type	Copper	Copper
Length of cable	8 m	10 m

Expander zones and automatic configuration

Edge expanders and fanout expanders have practically remained in history. This is often attributed to updates to SAS 2.0, but the real reason lies in the SAS zones introduced in 2.0, which remove the separation between edge and expansion expanders. Of course, zones are usually implemented specifically for each manufacturer, and not as a single industry standard.

In fact, several zones can now be located on the same information delivery infrastructure. This means that the targets (drives) in the storage can be accessed by different initiators through the same SAS expander. Domain segmentation is done through zones, access is done in an exclusive manner.