1 2 3 4 Previous Next 47 Replies Latest reply: Nov 20, 2013 2:24 PM by Miihkali Go to original post RSS
  • 15. Re: useless RAID5 performance with rapid storage-technology
    sMaestr0 Community Member
    Currently Being Moderated

    Interesting results Sagsagman.  Seeing an initial burst of performance is normal especially with cache enabled, but your throughput is still awfully low.  Single disk performance should be ideally around 100MB/sec sustained for sequential reads and wires.  In a three disk RAID 5 I usually expect well above single disk performance, around double makes me happy in software RAID.  Did you use the default Allocation Unit size (4K) when you created the volume in Windows?  I noted previously that I had much better performance from the 128KB stripe size and a 64K Allocation Unit size.  If you're not too far along you may want to try that combination as it has been working very well for me for over a week.  My three disk array (of slow 5900RPM disks) has better sustained write performance (over 200MB/sec) than my Crucial C300 SSD now.  I even see bursts of close to 400MB/sec for small transfers, according to the Windows 7 file transfer dialog. 

     

    I am a little confused about your disk arrangement.  You say you are using "2 RAID 5 volumes."  I assume you meant two partitions on the same three disk RAID 5 array.  If this is the case, I'm guessing your problem lies in the the fact the the OS is installed in the same RAID array as your storage.  You're much better off dedicating a single disk, or different array set for the OS.  Is the allocation unit size the same for the OS partition and Storage partition?  If they're not the same I think there is a greater potential for misalignment which causes performance degradation.  I'm and IT jack of all trades, not a storage specialist so correct me if I'm wrong... 

  • 16. Re: useless RAID5 performance with rapid storage-technology
    Currently Being Moderated

    TBH, this isn't a softraid v.s. hardraid related issue, it's purely bad software.  The Linux kernel is easily able to write to my multi-device raid5 array at around n-1 performance, which is completely expected since an X58/i7 rid can easily handle ~280MB/s throughput, even with the parity calculation overhead, not only does my system get that performance, but it doesn't exceed 10% CPU time on a single core while doing so (in fact, most 'hardware' raid cards are just an embedded computer on a PCIe/PCI-X card running a stripped down and customised Linux kernel in the firmware).

     

    Intel just needs to get their act together and provide a RAID5 implimentation which isn't a complete joke.

  • 17. Re: useless RAID5 performance with rapid storage-technology
    michaelgotberg Community Member
    Currently Being Moderated

    I have a similar configuration. I was only able to create one array in the RAID bios setup or rather I was not able to include a drive in more than one array. But maybe I am missing something. So I created one array consisting of all the available drives.Then I can setup volumes in that array with different RAID policies for every volume.

    In that volume I can create different partitions.

    I guess it is a differene of nomeclature? AFAIK RST is unique it its ability to have more than one raidpolicy per array. I guess you can refer to these "volumes" as separate arrays?

     

    My setup is two RAID5 volumes, one 2TB one with 64kb stripe size with a bootable NTFS system partition. Then a 8TB 128kb stripe data storage volume with a GPT partition on it.

    I cannot understand why this setup should create misalignment? AFAIK there is no option for deciding how these volumes are represented on the actual drives.

     

    I wonder if the performance degradation due to write size is per job or per file written? Ie if you are continously writing small files will it go down aswell or is it only when writing very large files? There is no data on the policy and size of the write cache used by the Intel RST.

     

    I have an theory about the issue with cluster size affecting performance.

    In an operation where for instance a write is performed the entire stripe is always read and then re-written. Maybe that is the source of the bad performance? Ie you are writing a series of small clusters but for every cluster the entire stripe is read, parity calculated then rewritten?

    So for instance if you have 4kb cluster size and a 64kb stripe size you are reading and rewriting the same stripe 64/4=16 times instead of pooling the writes and reading and writing it once?  This would imply a sensationally stupid write strategy for the RST driver. But the performance hit is definately in the ballpark of the numbers we are seeing?

    The solutions then would be to set stripe size=cluster size. This regardless of stripe size.

     

    Anyone that wants to help me test this idea? Is it totally bonkers or what do you all think?

  • 18. Re: useless RAID5 performance with rapid storage-technology
    sMaestr0 Community Member
    Currently Being Moderated

    Hey Michael, I was not aware of the ability to RAID the volumes that were created on an array.  Do you know of documents describing this?  This is accomplished in the RAID BIOS?  I think your right that alignment wouldn't be a concern then, but why would you want to implement that?  Why not have a single large array with partitions?  I have a SAN that had a very similar feature and the only use I could see for it was to RAID (most likely mirror) arrays between units or JBODs.  Wouldn't there be a large performance hit just by having a second layer like that?  As for testing your equal cluster/stripe size theory, I wish my system was at work and not home.  I do like your theory though, where is Intel on this?  Do they not moderate their own forums?  I don't have the resources or time to move my data off and back on to my system.  My kids would go nuts without their streaming cartoons! 

  • 19. Re: useless RAID5 performance with rapid storage-technology
    michaelgotberg Community Member
    Currently Being Moderated

    It is my understanding that due to limitations in the chipset you cannot boot to a volume of more than than 2.2TB. This will be rectified in the new z68 chipset.

    This is the only reason for me using two separate RAID5 volumes with one partition on each. You could set up two separate arrays with one volume on each but the reason for not doing that is that you cannot include a specific harddrive in more than one array so then in my case you would have only 3 drives in each array.

    The great thing with Intels implementation is that you can for instance set up a RAID0 volume for maximum performance and then use that as a cachedrive when using Photoshop or Lightroom and then have another volume with a higher level of redundancy to store vital information on.

     

    What I am trying to say is that you should think of these "volumes" as similar to what you would normally call an "array". There is no dual layer or anything.

    You cannot install a partition on what intel calls an "array".

     

    I am in the same situation as you and I have to backup about 4TB of data before trying  this out

     

    EDIT: Seems like Partition Manager can handle changing file allocation size without reformating the drive. It is only $30 so will try that when I have time.

  • 20. Re: useless RAID5 performance with rapid storage-technology
    mihies Community Member
    Currently Being Moderated

    More or less same here. Having 3x1.5 GB Seagate disks in RAID 5 array I am seeing 20-25MB/s write performance when a single disk is capable of ~120MB/s.

     

    About Intel software quality:

    http://communities.intel.com/thread/21144?tstart=0

    This is what I've found for IRST management software - awful quality and if the driver is of the same quality no wonders about the preformance.

  • 21. Re: useless RAID5 performance with rapid storage-technology, same on RAID1?
    Currently Being Moderated

    I've been analyzing a similar issue on my Intel DQ57TML with Xeon X3470, 16 GB RAM and 4x Seagate 1TB hard disks using Windows Server 2008 R2 SP1. In my opinion the installing the Intel Rapid Storage 10.1.0.1008 reduces the Sequencial Read/Write (Block Size = 1024KB) perfomance rapidly. The software I used for measuring speed is CrystalDiskMark 3.0.1 x64 using 1000 MB file size.

     

    The operating system has been freshly installed, all required drivers have been installed. Speed measure was before and after RST has been installed. DATA and Hyper-V are just formatted and do not contain any other data. Disk write cache has been enabled by using Device Manager (default settings).

     

    Some examples:

     

    1. OS Partition (Disk 0, Disk 1, RAID 1, 96 GB size, Default Windows Cluster Size)

    Read: 117,7 MB/s (standard Windows Driver)

    Write: 113,9 MB/s (standard Windows Driver)

    Read: 23,48 MB/s (RST Driver 10.1.0.1008)

    Write: 112,8 MB/s (RST Driver 10.1.0.1008)

     

    2. DATA Partition (Disk 0, Disk 1, RAID 1, 836 GB size, Default Windows Cluster Size)

    Read: 111,0 MB/s (standard Windows Driver)

    Write: 111,4 MB/s (standard Windows Driver)

    Read: 17,86 MB/s (RST Driver 10.1.0.1008)

    Write: 111,0 MB/s (RST Driver 10.1.0.1008)

     

    3. Hyper-V Partition (Disk 2, Disk 3, RAID 1, 932 GB size, 64k Windows Cluster Size)

    Read: 120,2 MB/s (standard Windows Driver)

    Write: 113,3 MB/s (standard Windows Driver)

    Read: 45.37 MB/s (RST Driver 10.1.0.1008)

    Write: 113,4 MB/s (RST Driver 10.1.0.1008)

     

    Any ideas why especially the Read performance has become so dramatically poor while the write performance remains nearly identically?

     

    Best regards,

    Thomas

  • 22. Re: useless RAID5 performance with rapid storage-technology, same on RAID1?
    michaelgotberg Community Member
    Currently Being Moderated

    @Thomas, Hmm how have you set up your RAID? You state 4 drives and it seems you use RAID1?! Try converting that to a RAID10 instead.

     

    What model of seagate drive is it? If it is an advanced format one then you need to align your partitions/volumes.

    Regarding read performane it seems that RST cannot benefit from reading from more than one drive (see posts above) however your readperformance is a lot less than even a single drive.

    You use a 1024k stripe size in the raid which is huuuuge, why not a more common 64-128 kb? What is your windows cluster size? Maybe the problems stem from the disparity between stripe and cluster(windows file allocation size)? My as of yet totaly unproven theory mentioned above when it comes to RAID5 .

     

    With your setup I would try out RAD10 with 4 drives using 64kb stripe and file allocation size.

     

    Next week I will posts some numbers when I have changed cluster size in windows.

  • 23. Re: useless RAID5 performance with rapid storage-technology, same on RAID1?
    Currently Being Moderated

    @Michael:

     

    I'm testing the performance of my machine after I had some strange issues with RST 10.1.0.1008 using RAID 10. One was disk performance issues when Hyper-V Virtual Machines with low disk usage were running. The other one was blue screens during the Windows Server Backup onto an USB drive with paging and disk errors although both the RAID and the USB disk are working fine even on normal file copies.

     

    Therefore I decided to buid the machine from scatch testing different RAID levels to see what will be most performing for me.

     

    The 1024k mentionend in my posting is not the stripe size (this is actually fixed on RAID 1) but the size the performance tool is using for his tests.

     

    The Seagate hard disks are ST31000525SV with

    • Disk data cache enabled
    • NCQ yes
    • SATA 3 Gb/s
    • Physical sector size 512 bytes
    • Logical sector size 512 bytes

     

    The data strip size of the RAID 10 is 64 kB, the data strip size of the RAID 5 is 128 kB (recommended settings within the RAID BIOS). Both RAIDs ain't initialized nor verified as initialization will be only required for verfication as well as I understand. In the next step I will check the performance after initialization.

     

    And here part two of the testing results:

     

    1. OS Partition (Disk 0 - 3, RAID 10, 96 GB size, Windows Partition, Default Windows Cluster Size)

    Read: 222,6 MB/s (standard Windows Driver)

    Write: 192,7 MB/s (standard Windows Driver)

    Read: 204,0 MB/s (RST Driver 10.1.0.1008)

    Write: 161,7 MB/s (RST Driver 10.1.0.1008)

    Read: 235,1 MB/s (RST Driver 10.1.0.1008, write Cache in Tool enabled)

    Write: 158,6 MB/s (RST Driver 10.1.0.1008, write Cache in Tool enabled)

    2. DATA Partition (Disk 0 - 3, RAID 10, 928 GB size, Windows Partition, 64k Windows Cluster Size)

    Read: 232,2 MB/s (standard Windows Driver)

    Write: 218,3 MB/s (standard Windows Driver)

    Read: 235,4 MB/s (RST Driver 10.1.0.1008)

    Write: 222,9 MB/s (RST Driver 10.1.0.1008)

    Read: 235,2 MB/s (RST Driver 10.1.0.1008, write Cache in Tool enabled)

    Write: 175,7 MB/s (RST Driver 10.1.0.1008, write Cache in Tool enabled)

    3. Hyper-V Partition (Disk 0 - 3, RAID 5, 932 GB size, 64k Windows Cluster Size)

    Read: 299,1 MB/s (standard Windows Driver)

    Write: 21,3 MB/s (standard Windows Driver)

    Read: 114,4 MB/s (RST Driver 10.1.0.1008)

    Write: 49,3 MB/s (RST Driver 10.1.0.1008)

    Read: 114,9 MB/s (RST Driver 10.1.0.1008, write Cache in Tool enabled)

    Write: 183,2 MB/s (RST Driver 10.1.0.1008, write Cache in Tool enabled)

     

    What's a surprise for me is the effect of enabling Write-back-Cache twice - once in the Device Manager and in the IRST Tool for RAID 5 systems. That's a huge performance increase! Is there a way (registry key) to check if write cache is enabled or not? Are there any other know registry performance tweaks?

     

    Best regards,

    Thomas

  • 24. Re: useless RAID5 performance with rapid storage-technology, same on RAID1?
    michaelgotberg Community Member
    Currently Being Moderated

    Performance goes up significantly after initializing the RAID5 array (takes about 24h though).

    Your readperformance seems really low though. What cluster size are you using?

     

    EDIT: saw that you were using 64k cluster size.

     

    your RAID5 results seem pretty ok when it comes to writespeed (which is usually very low with RST) but redo the tests after the RAID5 is completely initialized. Why on earth is readspeed much lower with RST? I get over 450MB/s in readspeed with my 6 drives but extremelly low write performance 30-40MB/s (have both of the mentioned writecache settings enabled).

    Seems that having a larger cluster size definately brings writespeed up into the 180MB/s+ range

  • 25. Re: useless RAID5 performance with rapid storage-technology, same on RAID1?
    michaelgotberg Community Member
    Currently Being Moderated

    Ok I now I did some changes to the system as follows:

     

    Volume 1 System boot volume 2TB RAID5 64kb stripe, changed file allocation size from 4kb to 32kb.

    Volume 2 Data storage volume 7TB RAID5 128kb stripe file allocation size was already set to 64 from the start so did not change it.

     

    The results where partly encouraging partly dumbfounding.

     

    Here are the results from Volume1 where write performance went upp across the board.

    C 32k cluster 64K stripe.jpg

    Here are the result from Volume 2

    D 64K cluster 128K stripe.jpgf i

     

    Performance went up a lot for volume 1 however it remained crappy for volume 2 !?

     

    Write cache is enabled both in the RST, and in the system settings on the actual volumes. Of course the drives where defragmented after every change to a setting.

     

    In order to see if the low performance was due to something else I tried the following settings with Volume 2 to see if it affected performance:

     

    1. Reduced the size of the partition on it to about half the size. No impact at all on performance.
    2. Reduced cluster size to 32kb. No impact at all on performance.
    3. Turned off the caches. Resulted in abysmal performance so turned them back on and results where back to where they were before.
    4. Checked the alignment on the partition on the volume2. It is a GPT partition so it has 128MB before it so it was aligned with the stripe size. The NTFS system partition on Volume 1 was aligned too.

     

    Regarding alignment  these are the rules a partition on a RAID should obey for maxmum performance:

     

    1. Partition offset (in kb) / stripe unit size(in kb) = integer
    2. Stripe unit size(in kb) / File allocation size(in kb) = integer.

    To get Partition offset; Start->Run->MSinfo32. in MSinfo32: Components->Storage->Partition Starting Offset. (this is in bytes so divide it with 1024 to get it in kb.

     

    My problem now is why I see such bad performance on one volume and a good one on the other with there beeing no appreciable difference between them?

     

     

    Trying to get hold of Intel support on this one

     

    UPDATE: Talked to Intel tech support and it was like getting water from a rock. They informed me that performance was expected to be this low and repeatedly stated that there were no issues with Intel Rapid Storage. I questioned them how they can put all the claims they have on the IRST site then if it actually not good enough to be useful? We have invested loads of manhours and equpment trying to gettting this to work.

    I hope someone in the more litigous US takes action then.

  • 26. Re: useless RAID5 performance with rapid storage-technology, same on RAID1?
    gavins Community Member
    Currently Being Moderated

    Hi all,

     

    I setup a raid5 on my asus p5qc last weekend and been seeing the slowness as well.

    I ran ATTO against it and got results I wasn't expecting (see below)

     

    test-run 03 + write-cache_small.JPG

     

    I'm running Win XP (64-bit), raid5 on 3x 1 TB disks with 64k stripe size + 32k NTFS block size.

    Hardware is ASUS P5QC which has an on-board ICH10R raid controller.

     

    I've tried running the raid as an MBR, dynamic and GPT disk, but the ATTO results are consistent regardless.

    I've also tried with/without write-cache and still results are consistent.

     

    Any one with any clues as to why write performance crashes at 512 kb writes and above?

    Any suggestions on what I can do to improve performance? (difference stripe size, etc)

     

    Thanks!

    Gavin

  • 27. Re: useless RAID5 performance with rapid storage-technology
    Currently Being Moderated

    Raid 5 problems over here aswell. I've got a H67 motherboard, 2300t, 4GB ram. OS:Win7 64bit. I' ve got a raid 5 array consisting of 3 Spinpoint F4 2Tb disks, with 2 volumes. The first is 78GB for windows, the second 3.6TB for data, with 128K offset/32K cluster size. Here is an ATTO run.

    [IMG]http://i55.tinypic.com/erzb5t.jpg[/IMG]

     

    Overal I'm pretty happy with the results ATTO comes up with, but when I try to copy a large file from a 1.5TB Spinpoint F3 disk I'm getting a speed of around 10MB/s.... Why does the benchmark give such good results, where a transfer is this slow?

  • 28. Re: useless RAID5 performance with rapid storage-technology
    alexeig Community Member
    Currently Being Moderated

    sagsagman wrote:

     

    I used Windows' Resource Monitor (perfmon) to monitor the system during such transfers. CPU usage is always very low (less than 10%, with perfmon itself using 6-7%). The "Disk activity" tab revelase something interesting: the "disk response time" column shows that response time for the file being written starts at 3sec, then rises steadily and can get as high as 13sec. I feel such a figure, for a local HDD system is an indication of a serious problem.

     

    A somewhat similar problem: R/W rates are 160/120MB/s on my 3-drive RAID5 however latencies are just horrific: 10 to 50 seconds.  You try to open a folder and just sit there waiting, then after 10-40 seconds it "wakes up" and quickly shows the contents.   A similar story with I/O tests: takes 10-40 seconds to start; once it does, it shows good numbers.

     

    I've rebuilt the array once and changed the "data strip size" from 128KB to 64KB - no change.  Nothing obvious in windows error logs.

     

    Here is my setup: RAID5: 3x WD2001FASS, 64KB data strip size, IRST 10.1.0.1008.  HP Z800 12GB RAM, Win7-Pro-64-bit.  I/O test: AJA System Test 1.0.

     

    P.S. I realize the first step is to try a set of different disks, just in case it's a drive problem - haven't had a chance yet.

  • 29. Re: useless RAID5 performance with rapid storage-technology
    sagsagman Community Member
    Currently Being Moderated

    Alex Gerulaitis wrote:

     

    sagsagman wrote:

     

    I used Windows' Resource Monitor (perfmon) to monitor the system during such transfers. CPU usage is always very low (less than 10%, with perfmon itself using 6-7%). The "Disk activity" tab revelase something interesting: the "disk response time" column shows that response time for the file being written starts at 3sec, then rises steadily and can get as high as 13sec. I feel such a figure, for a local HDD system is an indication of a serious problem.

     

    A somewhat similar problem: R/W rates are 160/120MB/s on my 3-drive RAID5 however latencies are just horrific: 10 to 50 seconds.  You try to open a folder and just sit there waiting, then after 10-40 seconds it "wakes up" and quickly shows the contents.   A similar story with I/O tests: takes 10-40 seconds to start; once it does, it shows good numbers.

     

    I've rebuilt the array once and changed the "data strip size" from 128KB to 64KB - no change.  Nothing obvious in windows error logs.

     

    Here is my setup: RAID5: 3x WD2001FASS, 64KB data strip size, IRST 10.1.0.1008.  HP Z800 12GB RAM, Win7-Pro-64-bit.  I/O test: AJA System Test 1.0.

     

    P.S. I realize the first step is to try a set of different disks, just in case it's a drive problem - haven't had a chance yet.

    The behaviour you're describing is quite similar to the one I'm experiencing, and there are some similarities in the setups we have. One major difference is that your HDDs are better than mine. While I do feel sorry for you for facing these problems, I'm also somewhat relieved to learn that these problems do not stem from my choice of HDDs.

     

    To tell you the truth, I've decided, even prior your message, to stop wasting time with Intel's sorry excuse for a RAID and ditch it. By now I'm very determined. I'm looking these days for a good deal on a RAID controller that will also support AF disks. I'd suggest you to do the same, it appears that Intel's RAID is, quite simply, a piece of junk.

More Like This

  • Retrieving data ...