Posted by Enlightenment on 5 oktober 2007 @ 19:34
Many of you people interested in storage will know the little benchmark utility called HDTune
. Its main features are testing a harddrive's read throughput rate and checking the drive's SMART information. Even though the program is unable to do any advanced measuring of performance, it is nonetheless used often by consumers to test their harddrives' performance.
With more people using or exploring the technology known as RAID, on internet forums I often see HDTune-screenshots posted to measure their RAID-array's performance. In this blog, I want to convince you not to use HDTune on your RAID-arrays.
Why HDTune is not suitable to test RAID arrays
HDTune is a utility created for bare hard drives, it was never meant to measure performance of RAID-arrays to any meaningful degree.
In consequence, HDTune does not use the filesystem on a harddisk to do its testing. Instead, HDTune communicates directly with the hardware. This means a lot of optimizations will go to waste, most notably Read Ahead. By lacking support for read ahead, the RAID array is unable to show its full potential.
How a RAID is different than a single hard drive
RAID uses a technique known as parallelization to achieve higher performance than a single drive could. But: a RAID array cannot process a single I/O request any faster than a single drive could, so why is it faster? The answer, already given, lies in the RAID's ability to process multiple I/O requests at the same time. Whereas a single disk would process I/O requests A-B-C in serial order, a RAID array could process them all at once, thereby completing the task much quicker.
However, HDTune does not allow RAID's to do this. HDTune would send a request, wait until it's done, then send the other request. Whereas a filesystem would send 10 requests at the same time, so that they may be processed in the most efficient way. The results are dramatic.
HDTune screenshot on RAID5 array
ATTO-256 screenshot on the same array
Interpretation: Notice the big difference in measured throughput; where HDTune stalls at around 60 Megs, ATTO reaches up to 180 Megs a second.
Lesson learned?
When benchmarking RAID arrays, use a benchmarking application which does its trick via the actual filesystem, such as ATTO-256 and SiSoftware Sandra. Also know that throughput performance isn't the only variable which determines overall performance. Booting your computer or loading an application, for example, will not depend on throughput performance a lot. For really advanced benchmarking, i recommend using Intel's IPeak-suite or the open source IOmeter application. Both are complicated to use, however, so only people really interested in storage will use them.
. Its main features are testing a harddrive's read throughput rate and checking the drive's SMART information. Even though the program is unable to do any advanced measuring of performance, it is nonetheless used often by consumers to test their harddrives' performance.With more people using or exploring the technology known as RAID, on internet forums I often see HDTune-screenshots posted to measure their RAID-array's performance. In this blog, I want to convince you not to use HDTune on your RAID-arrays.
Why HDTune is not suitable to test RAID arrays
HDTune is a utility created for bare hard drives, it was never meant to measure performance of RAID-arrays to any meaningful degree.
In consequence, HDTune does not use the filesystem on a harddisk to do its testing. Instead, HDTune communicates directly with the hardware. This means a lot of optimizations will go to waste, most notably Read Ahead. By lacking support for read ahead, the RAID array is unable to show its full potential.
How a RAID is different than a single hard drive
RAID uses a technique known as parallelization to achieve higher performance than a single drive could. But: a RAID array cannot process a single I/O request any faster than a single drive could, so why is it faster? The answer, already given, lies in the RAID's ability to process multiple I/O requests at the same time. Whereas a single disk would process I/O requests A-B-C in serial order, a RAID array could process them all at once, thereby completing the task much quicker.
However, HDTune does not allow RAID's to do this. HDTune would send a request, wait until it's done, then send the other request. Whereas a filesystem would send 10 requests at the same time, so that they may be processed in the most efficient way. The results are dramatic.
HDTune screenshot on RAID5 array
ATTO-256 screenshot on the same array
Interpretation: Notice the big difference in measured throughput; where HDTune stalls at around 60 Megs, ATTO reaches up to 180 Megs a second.
Lesson learned?
When benchmarking RAID arrays, use a benchmarking application which does its trick via the actual filesystem, such as ATTO-256
and SiSoftware Sandra. Also know that throughput performance isn't the only variable which determines overall performance. Booting your computer or loading an application, for example, will not depend on throughput performance a lot. For really advanced benchmarking, i recommend using Intel's IPeak-suite or the open source IOmeter application. Both are complicated to use, however, so only people really interested in storage will use them.
There are no replies yet
execution time: 90 msec, queries: 15
