You're absolutely right about the importance of driver quality. Running all the benchmarks took a huge amount of time (more than four months of testing, 20 hours a day). I just had to limit the scope of the article for practical reasons.
There is some information about driver support listed in the feature comparisons on page 13 and 15.
There are really nine adapters: 3ware Escalade 8506-8, 3ware Escalade 9500S-8, Areca ARC-1120, Areca ARC-1160, HighPoint RocketRAID 1820A, LSI MegaRAID SATA 150-4, LSI MegaRAID SATA 150-6, Promise FastTrak S150 SX4 and RAIDCore BC4852.
The results of the LSI MegaRAID SATA 150-4 and MegaRAID SATA 150-6 have been combined in the graphs since there is basicly no performance difference between to two in configurations up to four drives.
The IPEAK benchmarks were developed months before AnandTech and Storage Review expressed their bold statements about the lacking performance of RAID 0 configurations. We have developed our tests with the multi tasking usage patterns of power users in mind, but the (IPEAK) tests are by no means engineered to demonstrate the performance opportunities of RAID 0 configurations.
Similar performance improvements between RAID 0 and single drive configurations were already visible in our 2003 benchmark suite, based on different workloads. We have two independent benchmark suites showing results very contrary to those by AnandTech and Storage Review.
The fact is that the article readily admits that desktop workloads show poor average IOps (under 1.5) and modest average IO size (23K). Those numbers prove that there is little opportunity to accelerate performance either with parallel access or random access designs.
You're mixing statistics from different tests. The average queue depth was 1.34 I/Os in the Storage Review Office DriveMark 2002 trace and 3.22 I/Os in the Business Winstone 2004 trace from Tweakers.net. The low queue depth in SR's trace is one of the probable reasons for the disapppointing performance scaling in the RAID 0 benchmarks from Storage Review.
The results of the RankDisk tests show a respectable performance gain of 36 percent in office workloads on a low-end Promise FastTrak S150 TX2plus with two Raptor WD740GD drives. So even though the average queue size was pretty low and most (75%) of the transfer sizes were lower than the stripe size (64K), the Promise controller managed to show a significant performance increase.
Regarding the tests dispelling the myth of poor RAID 5 performance, hardly! Poor RAID 5 performance is no myth.
We have many benchmark results of RAID 5 configurations available in our Benchmark Database. It is true that most SATA RAID adapters have limited scalability and performance improvement in RAID 5. Many controllers won't scale beyond a 30-40 percent performance improvement over single drive configurations. Still, a 30 to 40 percent performance improvement cannot be considered 'poor' performance.
The MegaRAID SCSI 320-2X / Intel SRCU42X is a different beast. Thanks to a fast I/O processor, fast DRAM interface, large cache and high performance RAID software stack this adapters shows truely remarkable performance in RAID 5 configurations. The 320-2X provides some insight in the performance of future (SATA) RAID 5 adapters.
First off, the RAID 5 configuration was trounced by lesser RAID 0 IDE drives
So you expect a 4-drive SCSI RAID 5 configuration to be faster than a 4-drive RAID 0 configuration using faster SATA drives? Get real.
Second, the benchmarks consistently avoided writes, notably small writes, where RAID 5 massively fails, and uses a large writeback cache to further hide write performance and to cause the configuration to shine is small read tests.
Our desktop tests take more than two hours to run on a fast configuration and consists of almost half a million I/O operations with varying characteristics. In many of the tests, read I/O is outweighted by write I/O and average I/O size is below 50K. Modern RAID implementations have no problems with these type of access patterns.
If you are going to sing the praises of RAID 5 for data protection you should probably mention the data integrity disaster that writeback caches introduce.
That's why you can put a battery backup unit on any decent RAID 5 controller.
[quote] There's a big difference between RAID 0 being theoretically capable of superior performance and it being a performance value to a desktop user. This is a subjective matter and he fails to make his case. [/quote]
This case is clearly being made. For many power users, RAID 0 will improve performance just like moving to a faster hard drive will improve performance. The differences will be subtile improvements in responsiveness, not visible to everyone, better performance in multi-tasking scenarios with heavy I/O and better performance in applications limited by disk I/O. On many systems RAID is already embedded on the mainboard so the costs of RAID 0 are minimal.
Note that the purpose of the article is not to advocate RAID 0 or encourage people to use it. Personally I would not recommend RAID 0 to any user unless he or she has no data to care about.
Slashdot Mirror
You're absolutely right about the importance of driver quality. Running all the benchmarks took a huge amount of time (more than four months of testing, 20 hours a day). I just had to limit the scope of the article for practical reasons.
There is some information about driver support listed in the feature comparisons on page 13 and 15.
There are really nine adapters: 3ware Escalade 8506-8, 3ware Escalade 9500S-8, Areca ARC-1120, Areca ARC-1160, HighPoint RocketRAID 1820A, LSI MegaRAID SATA 150-4, LSI MegaRAID SATA 150-6, Promise FastTrak S150 SX4 and RAIDCore BC4852.
The results of the LSI MegaRAID SATA 150-4 and MegaRAID SATA 150-6 have been combined in the graphs since there is basicly no performance difference between to two in configurations up to four drives.
The IPEAK benchmarks were developed months before AnandTech and Storage Review expressed their bold statements about the lacking performance of RAID 0 configurations. We have developed our tests with the multi tasking usage patterns of power users in mind, but the (IPEAK) tests are by no means engineered to demonstrate the performance opportunities of RAID 0 configurations. Similar performance improvements between RAID 0 and single drive configurations were already visible in our 2003 benchmark suite, based on different workloads. We have two independent benchmark suites showing results very contrary to those by AnandTech and Storage Review.
You're mixing statistics from different tests. The average queue depth was 1.34 I/Os in the Storage Review Office DriveMark 2002 trace and 3.22 I/Os in the Business Winstone 2004 trace from Tweakers.net. The low queue depth in SR's trace is one of the probable reasons for the disapppointing performance scaling in the RAID 0 benchmarks from Storage Review.
The results of the RankDisk tests show a respectable performance gain of 36 percent in office workloads on a low-end Promise FastTrak S150 TX2plus with two Raptor WD740GD drives. So even though the average queue size was pretty low and most (75%) of the transfer sizes were lower than the stripe size (64K), the Promise controller managed to show a significant performance increase.
Regarding the tests dispelling the myth of poor RAID 5 performance, hardly! Poor RAID 5 performance is no myth.
We have many benchmark results of RAID 5 configurations available in our Benchmark Database. It is true that most SATA RAID adapters have limited scalability and performance improvement in RAID 5. Many controllers won't scale beyond a 30-40 percent performance improvement over single drive configurations. Still, a 30 to 40 percent performance improvement cannot be considered 'poor' performance.
The MegaRAID SCSI 320-2X / Intel SRCU42X is a different beast. Thanks to a fast I/O processor, fast DRAM interface, large cache and high performance RAID software stack this adapters shows truely remarkable performance in RAID 5 configurations. The 320-2X provides some insight in the performance of future (SATA) RAID 5 adapters.
First off, the RAID 5 configuration was trounced by lesser RAID 0 IDE drives
So you expect a 4-drive SCSI RAID 5 configuration to be faster than a 4-drive RAID 0 configuration using faster SATA drives? Get real.
Second, the benchmarks consistently avoided writes, notably small writes, where RAID 5 massively fails, and uses a large writeback cache to further hide write performance and to cause the configuration to shine is small read tests.
Our desktop tests take more than two hours to run on a fast configuration and consists of almost half a million I/O operations with varying characteristics. In many of the tests, read I/O is outweighted by write I/O and average I/O size is below 50K. Modern RAID implementations have no problems with these type of access patterns.
If you are going to sing the praises of RAID 5 for data protection you should probably mention the data integrity disaster that writeback caches introduce.
That's why you can put a battery backup unit on any decent RAID 5 controller.
[quote] There's a big difference between RAID 0 being theoretically capable of superior performance and it being a performance value to a desktop user. This is a subjective matter and he fails to make his case. [/quote]
This case is clearly being made. For many power users, RAID 0 will improve performance just like moving to a faster hard drive will improve performance. The differences will be subtile improvements in responsiveness, not visible to everyone, better performance in multi-tasking scenarios with heavy I/O and better performance in applications limited by disk I/O. On many systems RAID is already embedded on the mainboard so the costs of RAID 0 are minimal.
Note that the purpose of the article is not to advocate RAID 0 or encourage people to use it. Personally I would not recommend RAID 0 to any user unless he or she has no data to care about.