UNIX people... question about block sizes

Discussion in 'OT Technology' started by trouphaz, Aug 6, 2008.

  1. trouphaz

    trouphaz New Member

    Joined:
    Sep 22, 2003
    Messages:
    2,666
    Likes Received:
    0
    We use Veritas volumes for our raw devices for Sybase. The DBA just requested that we make our block size match theirs which is 8k. As far as I remember, you only set the block size when creating filesystems. There is no concept of block sizing at the volume level, correct?
     
  2. Peyomp

    Peyomp New Member

    Joined:
    Jan 11, 2002
    Messages:
    14,017
    Likes Received:
    0
    We use Gravitas mass for our steamed crevices for DBase.
     
  3. crontab

    crontab (uid = 0)

    Joined:
    Nov 14, 2000
    Messages:
    23,441
    Likes Received:
    12
    I don't think there is a block size on a raw volume.

    i've only specified block size on the creation of a file system. there are block sizes or chunks on a SAN volumes, but one normally cannot change that.

    is it a stripe in any way on the vxvm volume. are the stripe widths a denomination of the block size they want? but an 8k width is probably too big.

    I would tell them that they are out of their mind.

    i hate dba's who like to dictate storage etiquette. we're not in 1995 anymore.
     
  4. trouphaz

    trouphaz New Member

    Joined:
    Sep 22, 2003
    Messages:
    2,666
    Likes Received:
    0
    i hear you. i wish dbas would just get the fuck out. tell me what your IO requirements are instead of telling me how many spindles you think you need.


    anyway, i told him to knock it off. there is no block sizing for volumes and i'm not even going to mention stripe sizing.
     
  5. crontab

    crontab (uid = 0)

    Joined:
    Nov 14, 2000
    Messages:
    23,441
    Likes Received:
    12
    don't mention a thing. the more you tell them how your storage is architected, the more they want to "tune it". Tune it with methodologies pertaining to old ass single scsi JBOD setups of course.

    When I first started here, the dba manager demanded that latency for all their oracle files systems be sub 2ms all the time.

    "Are your app/db's performing slowly at peak times? All the time? Indexing too slowly?"

    "i don't know, I just look at the graphs you generate and I think latency should be close to 0."

    "Okay, let me see what I can do"

    So I got veritas, sun, and hitachi in a con call to literally tell her she was nuts. I haven't gotten an insane request from her since.
     
  6. Mike99TA

    Mike99TA I don't have anything clever to put here right now

    Joined:
    Oct 3, 2001
    Messages:
    4,553
    Likes Received:
    0
    Location:
    Greenville, SC
    Just to chime in, there is no way to set a block size of a volume, just the filesystem...

    FWIW we use 8k block sizes for our database and archive log volumes here, there is a huge performance increase over the default Veritas 1k block size for DB filesystems, just FYI. Its really simple to do and theres not really a tangible disadvantage to doing it during the creation of the filesystem.
     
  7. trouphaz

    trouphaz New Member

    Joined:
    Sep 22, 2003
    Messages:
    2,666
    Likes Received:
    0
    yeah, he started pushing that one and i put up a bit of a fuss just because the guy is irritating, but i know doing 1 IO vs doing 8 will help a lot. i just like to give DBAs a hard time.

    just an example, i was on vacation two weeks ago and one of the DBAs sent an email asking what storage a particular filesystem was on (for whatever reason, my company has our production systems on about 4 or 5 separate storage arrays from HDS, Storagetek and EMC). i couldn't help myself but respond, "what is your question? are you having a performance issue that you need to resolve? if so, please ask that and don't worry about the particulars about the storage."
     

Share This Page