Take the 2-minute tour ×
Super User is a question and answer site for computer enthusiasts and power users. It's 100% free, no registration required.

I need some help in interpretting the following speed tests on two Volumes on my home PC.

The test:

  • CrystalDiskMark 2.2
  • Windows 7 Disk Cache was disabled
  • Both volumes were 0% fragmented before the test

Drive C:

  • A Windows 7 Software RAID1
  • Dual Samsung HD501LJ drives (500GB, 16MB Cache, 7200RPM)

    Sequential Read:     69.580 MB/s  
    Sequential Write:    10.434 MB/s  
    Random Read 512KB:   36.241 MB/s  
    Random Write 512KB:  11.191 MB/s  
    Random Read 4KB:      0.499 MB/s  
    Random Write 4KB:     0.283 MB/s  
    Test Size:              100 MB
    

Drive D:

  • An integrated Intel Matrix RAID 10
  • Four Seagate ST3500320AS drives (500GB, 32MB Cache, 7200RPM)

    Sequential Read:    120.388 MB/s
    Sequential Write:    59.587 MB/s
    Random Read 512KB:   35.677 MB/s
    Random Write 512KB:  27.713 MB/s
    Random Read 4KB:      0.634 MB/s
    Random Write 4KB:     0.299 MB/s
    Test Size:              100 MB
    

Differences

    Sequential Read:      + 73%
    Sequential Write:    + 471%
    Random Read 512KB:     - 2%
    Random Write 512KB:  + 148%
    Random Read 4KB:      + 27% 
    Random Write 4KB:      + 6%

Would you agree with these statements?

  • "Sequential Read" improved probably because of: 4 disks to read from, not 2 AND better cache AND integrated RAID
  • "Sequential Write" improved probably because of integrated RAID
  • "Random Read 512KB" loss is nominal... and probably can't be easily explained in this scenario.
  • "Random Write 512KB" improved probably because of integrated RAID
  • "Random Read 4KB" improvement is a cross between Sequential Read (ideal) and Random Read 512KB (not ideal)
  • "Random Write 4KB" improvement is a cross between Sequential Write (ideal) and Random Write 512KB (not ideal)

EDIT: As suggested by others, I disabled Win7 Disk Cache and then used a different tool (CrystalDiskMark 2.2). Now the results make more sense to me... thank you.

share|improve this question

migrated from serverfault.com Oct 23 '09 at 6:17

This question came from our site for professional system and network administrators.

    
500gig drives are pretty close to the point where you do not want to use anything other than RAID10 (a stripe of mirrors) in order to be able to recover from drive failure. If you were using 1 terabyte dirves, I would say you definitely should not consider anything other than RAID10. On a home PC speed should be secondary to data security. –  Michael Dillon Oct 22 '09 at 19:49
    
You wouldn't use Mirrors with TB drives? –  Jason Oct 22 '09 at 20:03
    
As you have not specified either context or how these arrays are constructed no meaningful answer can be given. The various RAID different levels have pros and cons depending on how the array is used. e.g. Database vs simple file storage. There will also be a difference between using a proper hardware RAID controller or using Windows software RAID. –  John Gardeniers Oct 22 '09 at 21:31
    
@John... thanks for your comment, but I am not asking which is better. I'm asking if somebody can help me understand the numbers that were generated by the CrystalDiskMark tool. I fully understand that how the configurations are used will help me make my decision. –  Jason Oct 22 '09 at 21:34

1 Answer 1

up vote 1 down vote accepted

Your comparison is terrible, as you first use two drives with 16MB cache then drives with 32MB, and first you use software RAID, after which you use the onboard raid controller.

The test seems to be hitting some cache, as it is theoretically impossible to read at 1100mB/s from a RAID-10, even if you had SA-SCSI drives.

When using a proper hardware RAID controller, RAID-10 will undoubtedly offer a great performance improvement, at least 3 times as much read performance as opposed to a single drive setup.

RAID-1 will not bring noticeable speed increases, it will only add redundancy.

Depending on what you're testing for, in most enterprise environments sequential speeds mean nothing. It's all about the IOPS (I/O actions per second) and random read/write performance.

Edit: You may want to look into other benchmarking software, such as Crystalmark.

share|improve this answer
1  
Are you suggesting that the 16MB vs 32MB cache difference accounts for most of the differences in the speed tests? –  Jason Oct 22 '09 at 17:52
2  
I believe that he's suggesting that you're comparing oranges to bricks. You have too many variables and not enough information. Set them up as identically as you can, then change ONE thing. Also make sure that your test is MUCH larger than the available cache on the drives (and/or raid controller.) You don't say how much data you're moving around, you're just saying the rate. That tells us nothing about the actual speed of each setup. Did you disable the Win7 disk cache? –  Anonymous Oct 22 '09 at 18:08
    
I didn't set up the hardware for this test... I just happened to be testing my existing hardware and wanted to see if I could understand the results. Unfortunately, I haven't been able to yet. The drives are not identical, but they are pretty close. The differences should be between software or integrated RAID and RAID1 or RAID10. I'm just looking to understand the results... –  Jason Oct 22 '09 at 18:24
    
I have learned though (thank you @gekkz) that my numbers for the SEQ Read are impossible, so I need to do something to get the test to be correct. I used an off the shelf tool, with little settings, to do the test... I will go back and see what I can find. –  Jason Oct 22 '09 at 18:25
1  
"RAID-1 will not bring noticeable speed increases, it will only add redundancy." - Actually, depending on the controller, you can gain some read speed as it will read from both drives simulatneously. –  Maciej Swic Nov 25 '12 at 22:39

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.