SAS vs. SATA: Do you need a personal cow server saddle?
Today we decided to talk about hard drives with the SAS interface, and not by chance. A new round of interest in SAS hard drives arose about a year ago with the release of Asus motherboards based on the X58 chipset. Their distinguishing feature was an integrated SAS controller, whereas earlier, to use hard disks of this type, it was necessary to buy a separate controller, which cost no less than the motherboard. And slowly quite expensive SAS drives with a spindle speed of 15,000 revolutions per minute began to move to the top "game" personalki. We had the opportunity to evaluate the first trial configurations with a RAID array of two SAS Seagate Cheetah 15K.6 from major Moscow distributors, and immediately doubted the feasibility of installing these server “screws” in the desktop computer, although the most expensive one.
450 Gb SAS Seagate Cheetah 15K.6 Noise, vibrations and heat from the SAS were immediately noticeable. During the tests, the discs were heated so that it was difficult to hold the hand. But perhaps these disadvantages are justified by performance? Indeed, at one time, SATA drives with a spindle speed of 10,000 rpm firmly occupied their place in the top desktop systems. Well, let's try to answer this question in practice.
In today's test, we will test 450 Gb SAS Seagate Cheetah 15K.6 ST3450856SS hard drives , both single and in RAID0 and compare their performance with the results of a similar array from SATA's Seagate Barracuda 7200.12. This will allow us to find out whether hard drives with a spindle speed of 15,000 revolutions / minute or not give any advantage.
Compare the characteristics
')
On the Barracuda 7200.12 side, there is a larger volume, twice the angular density of data recording, a small power consumption in comparison with the SAS disks tested in this material, as well as twice the buffer volume. Manufacturers are in no hurry to put a large amount of cache memory on server hard drives - this is explained by the fact that storing a large amount of information in the buffer memory reduces the reliability for which these HDDs are designed.
Server Seagate Cheetah, in turn, has a distinct advantage in the form of twice the speed of rotation and significantly reduced due to this access time for reading and writing.
Testing method
Equipment configuration:
The following software was used for testing:
When testing single disks, the NTFS file system was used with a cluster size by default. Under the partition on the disk allocated all the available space. For hard drives combined in RAID0, the cluster size was chosen to be 16KB. Under RAID0 of two Seagate Cheetah, the maximum possible disk space was allocated, that is, RAID0 of two disks was created on a volume equal to the total volume of disks. The size of the array from Seagate Barracuda was equal to the size of a single disk. RAID0 from SAS-based disks was assembled on the integrated Marvell 88SE6320 controller. SATA drives were tested connected to the ICH10R.
All tests were conducted three times, the graphs show the average values ​​of the data obtained
HAB
"IO Delay" characterizes the access time to the hard disk, shows the speed of the disk response (the time during which the controller issues a request and processes it) and the system (the delay in the data transfer channel plus the delay in the driver plus the delay due to the speed of the processor itself) and speed interface. The smaller this parameter, the better.
The test ended with a clear advantage of the server disk, its reaction speed is almost two times higher than that of the competitor in this article. It is also worth noting that the access time to an array of two SAS disks is reduced compared to a single disk, in contrast to the same time for SATA disks. I will not undertake to assert that this is indeed possible and is not an error of the test program, but the trend itself is good news and gives us hope that the Seagate Cheetah will be able to please with good speed.
“Access” is responsible for the average access time, shows how long the head will fly over the plate surface until it finds the necessary data. This parameter greatly affects the response of the system - the smaller it is, the better.
Hopes are justified - the advantage in access time, which affects the subjective sense of hard disk speed more than the speed of reading, reaches 2.5 times. Both drives gain an advantage in access time when merging into RAID0, but you have to consider that with Seagate Barracuda this is due to the fact that the available disk space in the array is equal to half of the maximum possible disk capacity.
“Linear read” shows the speed of reading sequential data at different block sizes. Than this speed is higher, naturally, better.
SAS disk faster with any block size. Its maximum speed exceeds 160 MB / s versus 125 MB / s for Seagate Barracuda, which nevertheless is one of the fastest drives with a rotation speed of 7200 rpm. Also noteworthy is the advantage of the SAS controller: if RAID0 from Barracuda loses to a single when working with data blocks smaller than the cluster size, this is not observed in the array from Cheetah compared with a single disk.
Everest, HD Tune, HD Tach.
The access time of hard drives and arrays of them was also measured in other popular programs.
It is not surprising that in them a disk with a rotation speed of 15,000 rpm breaks out far ahead - 6 ms versus 15 ms. Interestingly, the opinion of various programs on whether the response from the RAID0 array of SAS-disks is faster than that of a single disk is divided in half. HD Tach in this question is in solidarity with HAB and notes some gain, while the results of HD Tune and Everest testing, on the contrary, show deterioration of the disk access time by 0.6-0.8 ms.
Also in these three programs linear speeds of reading of hard disks were measured.
* On the graph with the results in the HD Tach program, there is no Burst Speed ​​value for the array of SAS disks, since the program consistently produced numbers greater than 3500MB / s, which clearly cannot be true.
Seagate Cheetah shows excellent results for a mechanical disk - even at the end, the linear read speed does not fall below 100 MB / s, and in the case of a RAID0 array, below 200 MB / s. In the case of a single disc, the average reading speed is 30–50% higher than that of one of the fastest representatives of 7,200 rpm discs. In the case of merging disks in RAID0, the situation on the graphs for SAS disks is not so advantageous. A feature of this SAS controller is a very flat linear reading schedule when hard drives are running in a RAID0 array - the differences between the speeds at the beginning and the end of the disk are quite small compared to the corresponding schedule for regular hard drives. Hence, the seemingly paradoxical loss of server disks in linear read rates.
PCMark04
Next, let's compare the work of hard drives in PCMark packages. From PCMark04 only one test is interesting - “File Copying”, which is unique, that is, it is found only in this version of PCMark. This test assesses the speed at which a set of files is copied within a single hard disk partition.
PCMark04
The results of this test (moreover, they are well repeated in various operating systems) indicate its unsuitability for testing SAS disks. Just can not drive with such a speed of rotation and such a short response time so much lose the usual hard drive, even if one of the fastest. The acceleration of this test when using RAID0 from server disks was also not observed.
PCMark05
The following subtests are included in the 2005 test package: “Windows XP Startup”, which reflects the speed of the drive during the operating system load; "Application Loading", demonstrating the performance of the disk system with the sequential opening and closing of six popular applications; "General Usage", which displays the speed of hard drives when running a number of frequently encountered applications; File Write, estimating the speed of file creation; "Virus Scan", which measures the performance of the hard disk during the scan of files in the system for viruses.
PCMark05 - reading
Full fiasco Seagate Cheetah? No, rather, it's a test failure from Futuremark. No matter how I liked the graphics benchmarks of this company, but for testing some hard drives, its creations are not suitable at all. Of the 10 tests, SAS disks won only two: loading the Windows XP operating system and loading applications in the RAID0 array mode. The final result of the 2005 HDD Test Suite package is presented in the following diagram:
PCMark05 - final score
PCMark Vantage
Futuremark's newest overall system performance test includes as many as 8 hard disk performance tests.
In the “Windows Defender” subtest, the hard disk operates under multi-threaded load, one of the threads of which is file scanning. In “Gaming”, the drive behavior under load typical for computer games is emulated. In the Windows Photo Gallery subtest, the work of the drive is estimated when downloading images from the photo gallery. In Windows Vista Startup, the drive behavior is simulated when the Windows Vista operating system boots. In Windows Movie Maker, performance is estimated under a load characteristic of video editing. In the “Windows Media Center” subtest, the hard disk is tested in the mode typical for operation in the “Media Center”. In “Windows Media Player”, the addition of files to “Windows Media Player” is emulated. In "Application Loading" demonstrates the speed of the disk when loading several popular applications.
PCMark Vantage - reading
Judging by the results, little has changed in the PCMark test over the years between 2005 and Vantage. SATA-drives and one by one, and in the array look more confident. You can, of course, try to explain this with a large amount of cache memory or more efficient firmware from the Seagate 7200.12, but this advantage clearly does not reflect the true relationship between these disks. A single Cheeatah wins Barracuda in a gaming subtest, loading the Windows Vista operating system. The RAID0 array from server hard drives, in turn, wins when scanning files with antivirus, adding images to the photo gallery and loading Vista.
PCMark Vantage - Final Score
And I do not want to comment. According to the results of PCMark Vantage RAID0 of two 15,000 rpm hard drives running at the level of a single Seagate Barracuda ...
Iometer-2006.07.27
Iometer is a complex, fully synthetic test that can simulate the operation of a hard disk in various operating modes, for example, as a file server or workstation. Intel IOMeter gives you complete freedom to choose to configure this test application. During testing, IOMeter was configured in accordance with the recommendations of Intel and the methodology developed by the site Storagereview.com.
Intel IOMeter works with so-called "workers" (workers). For single-processor configurations, Intel recommends creating one at a time for such a worker. Each worker tests the “target” or “targets” (target), which are either an unpartitioned physical disk or one or more partitions (partition) on the disk. For each worker, the workers are assigned the so-called work rules, the “access pattern” (access pattern), which is a set of parameters in accordance with which the worker is accessed to his target (target).
The variables that make up the access model include:
Another important variable that is not directly included in the access model - # of Outstanding I / Os - determines the number of simultaneous I / O requests for a given worker and, accordingly, the disk load. If this parameter is set to 1, then we get the measurement of random access time. A value of 4 corresponds to the loading of a completely elementary application. On real applications, this parameter takes the value of 30-50. A parameter value greater than 100 corresponds to a very serious disk load, such as during defragmentation. In accordance with this test will be conducted using the following five values ​​of this parameter.
The article tested three common access models.
File Server - this pattern imitates the operation of a hard disk as a server disk subsystem.
Workstation - this pattern imitates the work of a hard disk when executing 2D / 3D design programs and video editing.
Database - this pattern imitates the operation of a hard disk during active work with databases.
IOMetr test settings
After running the test, Iometer creates a file with a large number of different numbers: the average number of requests made per second, the average operation time, the maximum operation time, the total number of bytes read and written, and the percentage of processor load. In order not to clutter the article, the diagrams will give three numbers for each configuration of the disk subsystem, which are ratings in the File Server, Workstation and Database access models, respectively. The rating is calculated as the arithmetic average of Total I / Os Per Second for all values ​​of the number of simultaneous I / O requests as a percentage of the corresponding hard disk 7200.11 ST31500341AS with a capacity of 1500GB, which will be compared in Iometer test in subsequent reviews.
Iometer File Server
After the disappointing results in PCMark, we will offer the disks to play on the “home” for the server-side Seagate Cheetah field, that is, just in the field of serser tasks :).
File Server Script Results
Let me remind you that 100 corresponds to the speed of the previous generation Seagate Barracuda 7200.11, the new 7200.12, as we see, is a little faster than it. But this increase in speed fades compared with the acceleration obtained from the work of the SAS disk. 130% difference for single disks, Cheetah double advantage over RAID0 from 7200r / min disks. If this speed is not enough for someone, the second SAS allows you to increase it by another 40%. In general, which of these disks to put in the file server, such a problem simply does not exist - the whole question is how many pieces to take.
Iometer workstation
Workstation Script Results
According to the results, it is clear that in the workstation one should not save on the disk subsystem. 15000 rpm look very impressive. drives. Is that the scalability of RAID0 in this case is worse - only 20%, but it is not so important - the usual drives are still far behind. By the way, in this pattern, Seagate 7200.12 was almost equal in speed to its predecessor, in other tests it usually had a bigger advantage.
Iometer database
Database Script Results
Databases also like high-speed hard drives - we again see their advantage more than two times. And really, they like high load SAS-disks, they proved to be excellent in all access models in the Iometer test.
Conclusion
The conclusion is that the use of SAS-drives on desktop systems is unjustified. The high price, strict cooling requirements and inadequate performance of many popular test packages make you think that with all the drawbacks, the user will not receive noticeable (with the exception of certain specific tasks) performance advantages. So cooler is not always better. And the use by some manufacturers of ready-made SAS disk systems in their configurations is more like a marketing ploy to attract wealthy buyers.
Source OCClub.ru
450 Gb SAS Seagate Cheetah 15K.6 Noise, vibrations and heat from the SAS were immediately noticeable. During the tests, the discs were heated so that it was difficult to hold the hand. But perhaps these disadvantages are justified by performance? Indeed, at one time, SATA drives with a spindle speed of 10,000 rpm firmly occupied their place in the top desktop systems. Well, let's try to answer this question in practice.
In today's test, we will test 450 Gb SAS Seagate Cheetah 15K.6 ST3450856SS hard drives , both single and in RAID0 and compare their performance with the results of a similar array from SATA's Seagate Barracuda 7200.12. This will allow us to find out whether hard drives with a spindle speed of 15,000 revolutions / minute or not give any advantage.
Compare the characteristics
')
On the Barracuda 7200.12 side, there is a larger volume, twice the angular density of data recording, a small power consumption in comparison with the SAS disks tested in this material, as well as twice the buffer volume. Manufacturers are in no hurry to put a large amount of cache memory on server hard drives - this is explained by the fact that storing a large amount of information in the buffer memory reduces the reliability for which these HDDs are designed.
Server Seagate Cheetah, in turn, has a distinct advantage in the form of twice the speed of rotation and significantly reduced due to this access time for reading and writing.
Testing method
Equipment configuration:
- Asus P6T6 Revolution Motherboard
- Intel Core i7 975 processor
- Maxtor DiamondMax <6L300S0> system hard drive
- 3GB DDR3 1800 MHz RAM
- Microsoft Windows Vista Operating System Service Pack 2
The following software was used for testing:
- Everest Ultimate Edition 5.02.1750
- HAB v0.4a
- HD Tach 3.0.4.0
- HD Tune v3.50
- PCMark04
- PCMark05
- PCMark Vantage
- Iometer-2006.07.27
When testing single disks, the NTFS file system was used with a cluster size by default. Under the partition on the disk allocated all the available space. For hard drives combined in RAID0, the cluster size was chosen to be 16KB. Under RAID0 of two Seagate Cheetah, the maximum possible disk space was allocated, that is, RAID0 of two disks was created on a volume equal to the total volume of disks. The size of the array from Seagate Barracuda was equal to the size of a single disk. RAID0 from SAS-based disks was assembled on the integrated Marvell 88SE6320 controller. SATA drives were tested connected to the ICH10R.
All tests were conducted three times, the graphs show the average values ​​of the data obtained
HAB
"IO Delay" characterizes the access time to the hard disk, shows the speed of the disk response (the time during which the controller issues a request and processes it) and the system (the delay in the data transfer channel plus the delay in the driver plus the delay due to the speed of the processor itself) and speed interface. The smaller this parameter, the better.
The test ended with a clear advantage of the server disk, its reaction speed is almost two times higher than that of the competitor in this article. It is also worth noting that the access time to an array of two SAS disks is reduced compared to a single disk, in contrast to the same time for SATA disks. I will not undertake to assert that this is indeed possible and is not an error of the test program, but the trend itself is good news and gives us hope that the Seagate Cheetah will be able to please with good speed.
“Access” is responsible for the average access time, shows how long the head will fly over the plate surface until it finds the necessary data. This parameter greatly affects the response of the system - the smaller it is, the better.
Hopes are justified - the advantage in access time, which affects the subjective sense of hard disk speed more than the speed of reading, reaches 2.5 times. Both drives gain an advantage in access time when merging into RAID0, but you have to consider that with Seagate Barracuda this is due to the fact that the available disk space in the array is equal to half of the maximum possible disk capacity.
“Linear read” shows the speed of reading sequential data at different block sizes. Than this speed is higher, naturally, better.
SAS disk faster with any block size. Its maximum speed exceeds 160 MB / s versus 125 MB / s for Seagate Barracuda, which nevertheless is one of the fastest drives with a rotation speed of 7200 rpm. Also noteworthy is the advantage of the SAS controller: if RAID0 from Barracuda loses to a single when working with data blocks smaller than the cluster size, this is not observed in the array from Cheetah compared with a single disk.
Everest, HD Tune, HD Tach.
The access time of hard drives and arrays of them was also measured in other popular programs.
Everest
HD Tune
HD tach
It is not surprising that in them a disk with a rotation speed of 15,000 rpm breaks out far ahead - 6 ms versus 15 ms. Interestingly, the opinion of various programs on whether the response from the RAID0 array of SAS-disks is faster than that of a single disk is divided in half. HD Tach in this question is in solidarity with HAB and notes some gain, while the results of HD Tune and Everest testing, on the contrary, show deterioration of the disk access time by 0.6-0.8 ms.
Also in these three programs linear speeds of reading of hard disks were measured.
Everest
HD Tune
HD tach
* On the graph with the results in the HD Tach program, there is no Burst Speed ​​value for the array of SAS disks, since the program consistently produced numbers greater than 3500MB / s, which clearly cannot be true.
Seagate Cheetah shows excellent results for a mechanical disk - even at the end, the linear read speed does not fall below 100 MB / s, and in the case of a RAID0 array, below 200 MB / s. In the case of a single disc, the average reading speed is 30–50% higher than that of one of the fastest representatives of 7,200 rpm discs. In the case of merging disks in RAID0, the situation on the graphs for SAS disks is not so advantageous. A feature of this SAS controller is a very flat linear reading schedule when hard drives are running in a RAID0 array - the differences between the speeds at the beginning and the end of the disk are quite small compared to the corresponding schedule for regular hard drives. Hence, the seemingly paradoxical loss of server disks in linear read rates.
PCMark04
Next, let's compare the work of hard drives in PCMark packages. From PCMark04 only one test is interesting - “File Copying”, which is unique, that is, it is found only in this version of PCMark. This test assesses the speed at which a set of files is copied within a single hard disk partition.
PCMark04
The results of this test (moreover, they are well repeated in various operating systems) indicate its unsuitability for testing SAS disks. Just can not drive with such a speed of rotation and such a short response time so much lose the usual hard drive, even if one of the fastest. The acceleration of this test when using RAID0 from server disks was also not observed.
PCMark05
The following subtests are included in the 2005 test package: “Windows XP Startup”, which reflects the speed of the drive during the operating system load; "Application Loading", demonstrating the performance of the disk system with the sequential opening and closing of six popular applications; "General Usage", which displays the speed of hard drives when running a number of frequently encountered applications; File Write, estimating the speed of file creation; "Virus Scan", which measures the performance of the hard disk during the scan of files in the system for viruses.
PCMark05 - reading
Full fiasco Seagate Cheetah? No, rather, it's a test failure from Futuremark. No matter how I liked the graphics benchmarks of this company, but for testing some hard drives, its creations are not suitable at all. Of the 10 tests, SAS disks won only two: loading the Windows XP operating system and loading applications in the RAID0 array mode. The final result of the 2005 HDD Test Suite package is presented in the following diagram:
PCMark05 - final score
PCMark Vantage
Futuremark's newest overall system performance test includes as many as 8 hard disk performance tests.
In the “Windows Defender” subtest, the hard disk operates under multi-threaded load, one of the threads of which is file scanning. In “Gaming”, the drive behavior under load typical for computer games is emulated. In the Windows Photo Gallery subtest, the work of the drive is estimated when downloading images from the photo gallery. In Windows Vista Startup, the drive behavior is simulated when the Windows Vista operating system boots. In Windows Movie Maker, performance is estimated under a load characteristic of video editing. In the “Windows Media Center” subtest, the hard disk is tested in the mode typical for operation in the “Media Center”. In “Windows Media Player”, the addition of files to “Windows Media Player” is emulated. In "Application Loading" demonstrates the speed of the disk when loading several popular applications.
PCMark Vantage - reading
Judging by the results, little has changed in the PCMark test over the years between 2005 and Vantage. SATA-drives and one by one, and in the array look more confident. You can, of course, try to explain this with a large amount of cache memory or more efficient firmware from the Seagate 7200.12, but this advantage clearly does not reflect the true relationship between these disks. A single Cheeatah wins Barracuda in a gaming subtest, loading the Windows Vista operating system. The RAID0 array from server hard drives, in turn, wins when scanning files with antivirus, adding images to the photo gallery and loading Vista.
PCMark Vantage - Final Score
And I do not want to comment. According to the results of PCMark Vantage RAID0 of two 15,000 rpm hard drives running at the level of a single Seagate Barracuda ...
Iometer-2006.07.27
Iometer is a complex, fully synthetic test that can simulate the operation of a hard disk in various operating modes, for example, as a file server or workstation. Intel IOMeter gives you complete freedom to choose to configure this test application. During testing, IOMeter was configured in accordance with the recommendations of Intel and the methodology developed by the site Storagereview.com.
Intel IOMeter works with so-called "workers" (workers). For single-processor configurations, Intel recommends creating one at a time for such a worker. Each worker tests the “target” or “targets” (target), which are either an unpartitioned physical disk or one or more partitions (partition) on the disk. For each worker, the workers are assigned the so-called work rules, the “access pattern” (access pattern), which is a set of parameters in accordance with which the worker is accessed to his target (target).
The variables that make up the access model include:
- Transfer Request Size is the minimum data block that a test application can access.
- Percent Random / Sequential Distribution - the percentage of random queries. The remaining requests, respectively, are sequential.
- Percent Read / Write Distribution - the percentage of read / write requests.
Another important variable that is not directly included in the access model - # of Outstanding I / Os - determines the number of simultaneous I / O requests for a given worker and, accordingly, the disk load. If this parameter is set to 1, then we get the measurement of random access time. A value of 4 corresponds to the loading of a completely elementary application. On real applications, this parameter takes the value of 30-50. A parameter value greater than 100 corresponds to a very serious disk load, such as during defragmentation. In accordance with this test will be conducted using the following five values ​​of this parameter.
- Linear - 1 Outstanding I / O
- Very Light - 4 Outstanding I / Os
- Light - 16 Outstanding I / Os
- Moderate - 64 Outstanding I / Os
- Heavy - 256 Outstanding I / Os
The article tested three common access models.
File Server - this pattern imitates the operation of a hard disk as a server disk subsystem.
Workstation - this pattern imitates the work of a hard disk when executing 2D / 3D design programs and video editing.
Database - this pattern imitates the operation of a hard disk during active work with databases.
IOMetr test settings
After running the test, Iometer creates a file with a large number of different numbers: the average number of requests made per second, the average operation time, the maximum operation time, the total number of bytes read and written, and the percentage of processor load. In order not to clutter the article, the diagrams will give three numbers for each configuration of the disk subsystem, which are ratings in the File Server, Workstation and Database access models, respectively. The rating is calculated as the arithmetic average of Total I / Os Per Second for all values ​​of the number of simultaneous I / O requests as a percentage of the corresponding hard disk 7200.11 ST31500341AS with a capacity of 1500GB, which will be compared in Iometer test in subsequent reviews.
Iometer File Server
After the disappointing results in PCMark, we will offer the disks to play on the “home” for the server-side Seagate Cheetah field, that is, just in the field of serser tasks :).
File Server Script Results
Let me remind you that 100 corresponds to the speed of the previous generation Seagate Barracuda 7200.11, the new 7200.12, as we see, is a little faster than it. But this increase in speed fades compared with the acceleration obtained from the work of the SAS disk. 130% difference for single disks, Cheetah double advantage over RAID0 from 7200r / min disks. If this speed is not enough for someone, the second SAS allows you to increase it by another 40%. In general, which of these disks to put in the file server, such a problem simply does not exist - the whole question is how many pieces to take.
Iometer workstation
Workstation Script Results
According to the results, it is clear that in the workstation one should not save on the disk subsystem. 15000 rpm look very impressive. drives. Is that the scalability of RAID0 in this case is worse - only 20%, but it is not so important - the usual drives are still far behind. By the way, in this pattern, Seagate 7200.12 was almost equal in speed to its predecessor, in other tests it usually had a bigger advantage.
Iometer database
Database Script Results
Databases also like high-speed hard drives - we again see their advantage more than two times. And really, they like high load SAS-disks, they proved to be excellent in all access models in the Iometer test.
Conclusion
The conclusion is that the use of SAS-drives on desktop systems is unjustified. The high price, strict cooling requirements and inadequate performance of many popular test packages make you think that with all the drawbacks, the user will not receive noticeable (with the exception of certain specific tasks) performance advantages. So cooler is not always better. And the use by some manufacturers of ready-made SAS disk systems in their configurations is more like a marketing ploy to attract wealthy buyers.
Source OCClub.ru
Source: https://habr.com/ru/post/73009/
All Articles