Crontab? Vxvm filesystem caching?

Discussion in 'OT Technology' started by trouphaz, Apr 13, 2009.

  1. trouphaz

    trouphaz New Member

    Joined:
    Sep 22, 2003
    Messages:
    2,666
    Likes Received:
    0
    we have a system that does a ton of read io and is killing my clariion array. It is solaris 8 with vxvm (don't remember which version). Is there a way to cache this data locally to reduce reads to the array? There is about 30gb of data.
     
  2. crontab

    crontab (uid = 0)

    Joined:
    Nov 14, 2000
    Messages:
    23,459
    Likes Received:
    12
    there are vxfs options for caching and what not, but i don't think will help constant read thrashing

    http://seer.entsupport.symantec.com/docs/285316.htm

    mincache and convosync dsync MAY help, but i doubt it.

    i've only had to tune these parameters for db's to bypass the local cache....

    i think vxfs mount with no options is buffered by default....

    anyway, i personally don't think vxfs tuning will help. can you mirror the data on the clarion? if not can you present another 30gb lun to the host and them mirror it via vxvm? that may spread the read load... will add some latency though. few ms at most.
     
  3. trouphaz

    trouphaz New Member

    Joined:
    Sep 22, 2003
    Messages:
    2,666
    Likes Received:
    0
    yeah, i'm more familiar with hpux and by default, disk caching can and will take up to 50% of your physical memory. it is a really stupid default option, but it is cool that you can control how much of your memory is set aside for buffer cache. so, if you have enough memory, you can do almost all IO in cache. from what i've been able to find, i haven't found anything similar in solaris.

    we don't really care about performance of this system. the problem is that it is overloading one of our raid groups as well as constantly blowing out cache with the constant reads. for now we set this particular lun to not use array caching for reads to help performance of the rest of the array, but that raid group will still get hammered.


    the options you mentioned though are for removing buffering entirely to get the performance/reliability of raw devices with the flexibility of filesystems. i want to try to buffer as much of the data locally so we don't have to do reads from the array as much.
     
  4. crontab

    crontab (uid = 0)

    Joined:
    Nov 14, 2000
    Messages:
    23,459
    Likes Received:
    12
    yeah those options are more for turning caching off for unbuffered or direct io.

    found something for decreasing the fs buffer size in mem.

    http://seer.entsupport.symantec.com/docs/196721.htm

    since they show how to calculate the tunables, this can be used to increase the buffer....

    never done this before but seems similar to stuff i've tried for linux...
     
  5. trouphaz

    trouphaz New Member

    Joined:
    Sep 22, 2003
    Messages:
    2,666
    Likes Received:
    0
    awesome, i'll take a look. we currently have little free memory, but we can certainly add some more.
     

Share This Page