Re: SSA Adapter Performance

From: Kurt Bremser (kurt.bremser_at_allianz.at)
Date: 03/17/04

  • Next message: Kurt Bremser: "Re: rte files" 
    Date: 17 Mar 2004 03:18:27 -0800

    As said before, if you have considerable write I/O load, you better
    switch to RAID 1. RAID 5 means that, for every single write, the
    subsystem has to write the data, read in all the corresponding data
    from the data disks in the same array, compute the checksum, and write
    that to the parity disk. The more space-efficient your RAID-5 is, the
    more time you lose when writing.
    Also, consider spreading your load by striping:
    Set up several arrays on your SSA loops (spread disks over both loops
    of the adapter), put these into your vg's, and then use a convenient
    stripe size (Does your application read data in pages? If so, use that
    as a starting point in your considerations) so that the load is spread
    over at least 4 pv's.
    Fast disks will give you a maximum of 30 MB/s (ca), so if you want to
    saturate your SSA loop, you have to make 8 disks or more work
    concurrently!

    KB

    jsmith@fluidmaster.com (Jim) wrote in message news:<82e9a4d8.0403161617.714f4869@posting.google.com>...
    > Hello,
    >
    > We've just installed a new SSA adaptor (IBM SSA 160 SerialRAID
    > Adapter) with 128mb DRAM and 32MB of fast-write cache. We were hoping
    > to see a performance increase but haven't noticed a significant
    > improvement yet. This adapter is supporting a 60gig RAID array that
    > contains our busiest LVs. Many of our busiest files are very large
    > 800mb+. The I/O on this array is very high with %IOWAIT consistently
    > over 50% and at many times 90+%. I've tuned all of the disk device
    > attributes like queue_depth, write_queue_mod, and max_coalesce which
    > have helped a little but not much. I'm thinking the vmtune parameters
    > numclust and/or maxrandwrt might be the next place to look/adjust.
    > Has anyone had success tuning these parameters?
    >
    > Thanks in advance.
    >
    > Jim

  • Next message: Kurt Bremser: "Re: rte files"

    Relevant Pages

    • Re: Linux SuSe, slowest browsing ever seen, IT SUCKS
      ... > 1, RAID wouldn't not work, had to IDE my disks. ... Also the pages won't load like they ...
      (alt.os.linux.suse)
    • Extending Exch 5.5 hard disk on RAID 5
      ... Drive E:\ consists of 3x18gb disks. ... replacement to the existing 18gb drives and also put RAID ... Load the backed up dbase. ...
      (microsoft.public.exchange.admin)
    • Re: Extending Exch 5.5 hard disk on RAID 5
      ... I would put a step in there to test the backp by doing a test restore of ... Drive E:\ consists of 3x18gb disks. ... > replacement to the existing 18gb drives and also put RAID ... > Load the backed up dbase. ...
      (microsoft.public.exchange.admin)
    • Re: difference between striping using mdadm and LVM
      ... way!), is "thou shalt make a plan for restoring from backups, test ... another bonus of software raid - if your hardware raid card dies, ... additional backups on - I store my other backups on Iomega REV disks. ... is part of the live array, and both data and parity blocks are ...
      (comp.os.linux.misc)
    • Re: DC RAID Configuration
      ... If the machine has 6 spindles then OS/Logs go on a hardware mirror, if 10 then they go on a 4 disk hardware based 0+1 RAID. ... Their DIT is only in the teens in size so the disks themselves could be small, you just need many to get the high IOPS that make AD really run smoothly. ... 30 DCs with 2500 users total, unless the environment had a lot of very heavy use AD apps I would likely be ok for even a single mirror for those DCs. ...
      (microsoft.public.win2000.active_directory)