vSphere Storage Appliance

 View Only
  • 1.  Block Alignment

    Posted Jul 26, 2010 04:31 PM

    Is the block alignment and issue with all SAN storage? I know I had this issue with Netapp SAN and now have Hitachi SAN.

    MSINFO32 results for Partition Starting Offset

    32,256/4,096=7.875



  • 2.  RE: Block Alignment

    Posted Jul 26, 2010 04:46 PM

    If you setup your LUN type to VMware, and create your VMFS using Virtual Center, they should be aligned properly but your VMs will still be misaligned.

    So you'll have misaligned VMs writing to aligned VMFS. Not sure how big the performance impact would be but the best solution is to just aligned the VM to your storage and make a template out of it.

    Windows 2008 has a starting offset of 1MB so you're fine there.

    We've been aligning on our netapp for 12mo+ now. It's an on going maintenance issue for us since lots of virtual appliances aren't aligned, lots of vendor provided VMs are not aligned, creating secondary disks on win2k3 and linux aren't aligned unless you take the time to align them using diskpart of fdisk.

    Total pain....but most people aren't aware of the issue so it continues to be a problem.



  • 3.  RE: Block Alignment

    Posted Jul 26, 2010 04:54 PM

    IMO it's always best practice to align the partitions for newly created VMs. I usually align them to 1 MB, which is the default for newer Windows OSs. Alignment may be an issue for systems with high disk I/O like DB servers. For systems with "normal" I/O you may not see any real improvement.

    André



  • 4.  RE: Block Alignment

    Posted Jul 26, 2010 05:35 PM

    What would be the best way to capture/monitor statistics on alignment. Is that degradation in performance seen at the OS level on the VM or on the ESX level?



  • 5.  RE: Block Alignment

    Posted Jul 26, 2010 06:09 PM

    You could capture them through the guest OS if you're running perfmon or something but that doesn't seem reasonable for a large deployment.

    You could use Vcenter4.1 though to capture them too. For us, NFS is our standard, until vcenter 4.1, we couldn't get any NFS stats out of vcenter or esx.

    Now we can!!...that's another option for you especially using vmfs.

    Type esxtop, d, look at DAVG, KAVG, GAVG, QAVE. check man esxtop for details.

    DAVG/cmd

    The average device latency (millisecs) per command.

    KAVG/cmd

    The average ESX VMKernel latency (millisecs) per command.

    GAVG/cmd

    The average Guest OS latency (millisecs) per command.

    QAVG/cmd

    The average queue latency (millisecs) per command.

    We capture latency through the Netapp DFM product and a few other sources.

    The DFM product will show latency on a volume, which is caused by some misalignment, and some other sources.

    Netapp "might" have other methods to detect alignments but right now, that's the only one available.

    hope that helps!

    KC