Professional Web Applications Themes

Stripeset layout challenge - AIX

There is a question about striping I asked here at an earlier time, but no relevant answers were posted. The question is still open to me, and I would like to post it again in a simpler and maybe more intellegible way. We have 15 disks on three buses, say each has 32 GB size to a total of 480 GB. The DBA would like to have the following filesystems: 160 GB data 80 GB index 80 GB redologs etc. these should be on different disks and should be rather fast; and the remaining 160 GB are needed for dumps ...

  1. #1

    Default Stripeset layout challenge


    There is a question about striping I asked here at an earlier time,
    but no relevant answers were posted. The question is still open to
    me, and I would like to post it again in a simpler and maybe more
    intellegible way.

    We have 15 disks on three buses, say each has 32 GB size to a
    total of 480 GB.

    The DBA would like to have the following filesystems:
    160 GB data
    80 GB index
    80 GB redologs etc.
    these should be on different disks and should be rather fast; and the
    remaining 160 GB are needed for dumps etc. and could be rather slow,
    as these are not to be used in production mode.

    We should use striping with a stripe-width of 3 (or a multiple of 3)
    to optimally utilize the three buses.

    So we use 6 disks for data (160 GB used, 32 GB spare), 3 disks for
    index (80 GB used, 16 GB spare), and another 3 disks for redologs
    (80 GB used, 16 GB spare). The remaining 3 disks make up another
    96GB for the dumps filesystem - this is not enough, and we need to
    add the spare 64 GB from the other stripesets to this filesystem.

    Now here is the problem: with LVM, the stripe-width cannot be
    smaller than the number of disks that make up a VG, and no file-
    system can span disks than do not belong to the same VG.

    So if we have a filesystem that needs to span all 15 disks (to use
    up the remaining spare parts of the disks), then we must put them
    all in the same VG and must use a stripe-width of 15. While this
    does not harm our constraint to use a multiple of 3 for the
    stripe-width, it will make the other filesystems also span all 15
    disks, and that results in data and index and redologs dwelling on
    the same disks and producing a lot of unwanted head-movement!
    >From the technical viewpoint, there should be no problem in running
    a bunch of 15 disks as one VG with a stripe-width of 3:

    BUS 1 BUS 2 BUS 3
    disk 1 disk 6 disk 11
    |---------| |---------| |---------|
    | block 1 | | block 2 | | block 3 |
    | block 4 | | block 5 | | block 6 |
    | ... | | ... | | ... |
    | blk 298 | | blk 299 | | blk 300 |
    |---------| |---------| |---------|
    disk 2 disk 7 disk 12
    |---------| |---------| |---------|
    | blk 301 | | blk 302 | | blk 303 |
    | ... | | ... | | ... |
    | blk 598 | | blk 599 | | blk 600 |
    |---------| |---------| |---------|
    disk 3 disk 8 disk 13
    |---------| |---------| |---------|
    | blk 601 | | blk 602 | | blk 603 |
    ... ... ...

    But LVM cannot do this. (I think the low-level components of LVM
    can do it -if used in an unsupported way-, but I did not inves-
    tigate further.)
    What we would need here, is something sitting below LVM that would
    simply concatenate all 5 disks on each bus and make them appear
    as one hdisk to LVM. Certain controllers can do such, but they
    usually do also striping on their own and do not let us know about
    the actual physical locations of the blocks they present.

    I would like to hear about approaches to solve this problem, on
    AIX or on other OS. (I know only aout Greg Lehey's "vinum" volume
    manager which is distributed with FreeBSD - there the stripe-width
    and -size can be designed individually for each filesystem, and
    arbitrary parts of physical disks can be put into different
    stripesets, and so it is easily possible to solve the problem.)

    Alright, today there is a habit in using heaps of diskspace from
    SAN and not even knowing about the physical location, but I think
    it is still true that the outer cylinders of a physical disk bring
    nearly twice the throughput than the inner ones, and so there is
    still performance to be gained by a deliberately planned disk
    layout. Or am I wrong here?

    rgds,
    PMc


    Peter Much Guest

  2. #2

    Default Re: Stripeset layout challenge

    Peter Much wrote:
    > There is a question about striping I asked here at an earlier time,
    > but no relevant answers were posted. The question is still open to
    > me, and I would like to post it again in a simpler and maybe more
    > intellegible way.
    >
    > We have 15 disks on three buses, say each has 32 GB size to a
    > total of 480 GB.
    >
    > The DBA would like to have the following filesystems:
    > 160 GB data
    > 80 GB index
    > 80 GB redologs etc.
    > these should be on different disks and should be rather fast; and the
    > remaining 160 GB are needed for dumps etc. and could be rather slow,
    > as these are not to be used in production mode.
    >
    Would using all the disks on one bus as a group really make any
    difference in the performance? This way you could tie the five disks on
    one bus for your data, the five on another for the index and redologs
    and the final five for the dumps.
    > We should use striping with a stripe-width of 3 (or a multiple of 3)
    > to optimally utilize the three buses.
    >
    > So we use 6 disks for data (160 GB used, 32 GB spare), 3 disks for
    > index (80 GB used, 16 GB spare), and another 3 disks for redologs
    > (80 GB used, 16 GB spare). The remaining 3 disks make up another
    > 96GB for the dumps filesystem - this is not enough, and we need to
    > add the spare 64 GB from the other stripesets to this filesystem.
    >
    > Now here is the problem: with LVM, the stripe-width cannot be
    > smaller than the number of disks that make up a VG, and no file-
    > system can span disks than do not belong to the same VG.
    >
    > So if we have a filesystem that needs to span all 15 disks (to use
    > up the remaining spare parts of the disks), then we must put them
    > all in the same VG and must use a stripe-width of 15. While this
    > does not harm our constraint to use a multiple of 3 for the
    > stripe-width, it will make the other filesystems also span all 15
    > disks, and that results in data and index and redologs dwelling on
    > the same disks and producing a lot of unwanted head-movement!
    >
    >>From the technical viewpoint, there should be no problem in running
    > a bunch of 15 disks as one VG with a stripe-width of 3:
    >
    > BUS 1 BUS 2 BUS 3
    > disk 1 disk 6 disk 11
    > |---------| |---------| |---------|
    > | block 1 | | block 2 | | block 3 |
    > | block 4 | | block 5 | | block 6 |
    > | ... | | ... | | ... |
    > | blk 298 | | blk 299 | | blk 300 |
    > |---------| |---------| |---------|
    > disk 2 disk 7 disk 12
    > |---------| |---------| |---------|
    > | blk 301 | | blk 302 | | blk 303 |
    > | ... | | ... | | ... |
    > | blk 598 | | blk 599 | | blk 600 |
    > |---------| |---------| |---------|
    > disk 3 disk 8 disk 13
    > |---------| |---------| |---------|
    > | blk 601 | | blk 602 | | blk 603 |
    > ... ... ...
    >
    > But LVM cannot do this. (I think the low-level components of LVM
    > can do it -if used in an unsupported way-, but I did not inves-
    > tigate further.)
    > What we would need here, is something sitting below LVM that would
    > simply concatenate all 5 disks on each bus and make them appear
    > as one hdisk to LVM. Certain controllers can do such, but they
    > usually do also striping on their own and do not let us know about
    > the actual physical locations of the blocks they present.
    >
    > I would like to hear about approaches to solve this problem, on
    > AIX or on other OS. (I know only aout Greg Lehey's "vinum" volume
    > manager which is distributed with FreeBSD - there the stripe-width
    > and -size can be designed individually for each filesystem, and
    > arbitrary parts of physical disks can be put into different
    > stripesets, and so it is easily possible to solve the problem.)
    >
    > Alright, today there is a habit in using heaps of diskspace from
    > SAN and not even knowing about the physical location, but I think
    > it is still true that the outer cylinders of a physical disk bring
    > nearly twice the throughput than the inner ones, and so there is
    > still performance to be gained by a deliberately planned disk
    > layout. Or am I wrong here?
    >
    > rgds,
    > PMc
    >
    >
    Michael W Ryder Guest

  3. #3

    Default Re: Stripeset layout challenge

    [email]pmccitylink.dinoex.sub.org[/email] (Peter Much) wrote in message news:<HLxopD.81tcitylink.dinoex.sub.org>...
    > There is a question about striping I asked here at an earlier time,
    > but no relevant answers were posted. The question is still open to
    > me, and I would like to post it again in a simpler and maybe more
    > intellegible way.
    >
    > We have 15 disks on three buses, say each has 32 GB size to a
    > total of 480 GB.
    >
    > The DBA would like to have the following filesystems:
    > 160 GB data
    > 80 GB index
    > 80 GB redologs etc.
    > these should be on different disks and should be rather fast; and the
    > remaining 160 GB are needed for dumps etc. and could be rather slow,
    > as these are not to be used in production mode.
    One general remark: do NOT separate index and data onto different
    (sets of) disks.
    Check out:
    [url]http://groups.google.nl/groups?hl=nl&lr=&ie=UTF-8&oe=UTF-8&threadm=guYw8.40277%24Rw2.3101459%40bgtnsc05-news.ops.worldnet.att.net&rnum=1&prev=/groups%3Fhl%3Dnl%26ie%3DUTF-8%26oe%3DUTF-8%26q%3Dkyte%2Bindex%2Braw%2Bseparation%26sa%3DN%2 6tab%3Dwg%26lr%3D[/url]
    Hans de Git Guest

  4. #4

    Default Re: Stripeset layout challenge


    Peter Much wrote:

    > Now here is the problem: with LVM, the stripe-width cannot be
    > smaller than the number of disks that make up a VG,
    Are you telling that if there are 5 disks in a VG, a LV needs to use all 5 ?

    Thats not true; of course you can use only 3 f.e.

    Holger van Koll Guest

Similar Threads

  1. Help plz! Who's up for the challenge?
    By mr.tibbs webforumsuser@macromedia.com in forum Macromedia Flash Actionscript
    Replies: 1
    Last Post: January 17th, 09:18 AM
  2. Challenge #32
    By Grant Dixon in forum Adobe Photoshop Elements
    Replies: 55
    Last Post: September 19th, 01:54 AM
  3. OO Challenge
    By Henry Gilbert in forum Ruby
    Replies: 13
    Last Post: September 15th, 07:38 AM
  4. Challenge #31
    By Grant Dixon in forum Adobe Photoshop Elements
    Replies: 103
    Last Post: September 11th, 10:25 PM
  5. Little challenge for ya
    By Bill Ray in forum Macromedia Fireworks
    Replies: 0
    Last Post: July 11th, 09:43 PM

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139