Explain the storage benchmark tests

Berryracer

Platinum Member
Oct 4, 2006
2,779
1
81
Let's take for example, CrystalDiskMark

there are a few tests like:


  • Seq
  • 512K
  • 4K
  • 4K QD32
I don't know the difference all I know is that what a user would experience in a real world usage and what determines the OS snappiness are the 4K results..

can you please explain each value in a bit more detail and under what scenario would that test come into play in real life?

Because according to this.... it seems like the RAID 0 is slower in the 4K benchmarks than a single drive which confuses the heck out of me...I thought it would be double the speed

zv3r0o.png


k47ds1.png
 
Last edited:

Cerb

Elite Member
Aug 26, 2000
17,484
33
86
Sequential: large amounts of data in order (multi-megabyte).

512KB: blocks of 512KB. So, write 512KB here, another 512KB there, etc., then for reading them back, read them back in a predictable pseudo-random order.

4KB: as above, but with the smallest logical size that matters.

512KB would be representative of largish business documents, small pictures (by today's standards), compressed scanned 1-page documents, etc..

4KB is commonly written, but less commonly read, and would be representative of configuration files, icons, bits of logs, lock files, etc..

RAID 0 works in stripes, a contiguous block of data that each drive gets, and usually 64KB (with cheap disks and RAM, these days, if you need to adjust stripe size, you really need faster drives). While a given read does not necessarily need to read a whole stripe, nor write a whole stripe, it still has to manage and update a whole stripe, logically, so there's some overhead in safely handling that.

On top of that, at QD=1, each read or write needs to complete before the next one is issued, so the latency of the drive and AHCI/RAID/SATA limit how fast it can be done. Most drives likely a lie a bit, when it comes to writing, as well, which would help explain the difference (some claim not to, but most ignore the subject, and consumer drives have a long history of lying about data actually being committed, to improve speed).

On of that, since you're never going to be limited to a QD of 1 in practice in a RAID 0 of SSDs, the QD=1 bench is really only useful as a check to make sure things are working like expected, not good as a performance comparison. But, by the same token, since you'll basically never have a QD of 32 per drive for more than a brief instant, outside of synthetic benchmarks, the QD=64 bench, while useful, isn't really telling you how your drive will perform (PCMark and Iometer are among the best all-around tools for that, IMO).
 
Feb 25, 2011
16,992
1,621
126
The reason RAID0 is sometimes slower for random access than a plain drive is that the RAID controller is having to a do a lot more math, keeping track of which sectors/data/blocks are on which disk, that it used to be able to offload to the HDD.

You wouldn't usually notice with old-fashioned spinning HDDs, because they're only capable of a couple hundred random operations per second anyway. But there are SSDs that can toss off 20k IOPS like nothing, so a RAID controller than can only deal with 15k IOPS is suddenly a real bottleneck.
 

Berryracer

Platinum Member
Oct 4, 2006
2,779
1
81
The reason RAID0 is sometimes slower for random access than a plain drive is that the RAID controller is having to a do a lot more math, keeping track of which sectors/data/blocks are on which disk, that it used to be able to offload to the HDD.

You wouldn't usually notice with old-fashioned spinning HDDs, because they're only capable of a couple hundred random operations per second anyway. But there are SSDs that can toss off 20k IOPS like nothing, so a RAID controller than can only deal with 15k IOPS is suddenly a real bottleneck.

so from the above benchmarks I posted, and according to your and the above post, it sounds to me that RAID 0 is useless for the average consumer like me who only surfs the net and occasionally watches a movie
 

Cerb

Elite Member
Aug 26, 2000
17,484
33
86
so from the above benchmarks I posted, and according to your and the above post, it sounds to me that RAID 0 is useless for the average consumer like me who only surfs the net and occasionally watches a movie
RAID 0 is pretty much useless if you don't move large amounts of large files on a regular enough bass that you are waiting for them to finish at ~200+MBps.
 
Feb 25, 2011
16,992
1,621
126
so from the above benchmarks I posted, and according to your and the above post, it sounds to me that RAID 0 is useless for the average consumer like me who only surfs the net and occasionally watches a movie

Pretty much.

A sufficiently capable RAID controller wouldn't bottleneck the drives.

Old-fashioned hard drives are usually slow enough that RAID-ing a couple of them together gets you some benefits even for a consumer workload.

But in either case it's an additional expense for not a lot of benefit.

Well... bragging rights.