UNSOLVED

lenystr

updated

11 years ago

L

lenystr

43 Posts

2

11428

July 27th, 2011 21:00

VMAX Meta Members Best Practice

Is there a best practice for the number of meta members a meta should have? We are creating a 1920GB (8 x 240GB tdevs) striped meta for a VMWare  VMFS. I have always tried to keep my metas base-2 (2,4,8,16) but not sure if this really matters.

  • Quincy561

    1281 Posts

    1904

    0

    Posted July 28th, 2011 02:00

    8 meta members should be enough for performance considerations with VP.  Powers of 2 were a very good idea with thick volumes but is not important for VP volumes. 

  • modaslam

    2 Intern

    217 Posts

    1904

    0

    Posted August 2nd, 2011 02:00

    what are VP voulmes?

  • Quincy561

    1281 Posts

    1903

    0

    Posted August 2nd, 2011 03:00

    And for best performance, the VP (TDEVs or thin) metas should be striped.  Concatenated metas can be expanded easier, but they will have lower overall performance.

  • Quincy561

    1281 Posts

    1904

    0

    Posted August 2nd, 2011 03:00

    VP = Virtual provisioning volumes (thin).  You mentioned TDEVs which are VP volumes.

  • Quincy561

    1281 Posts

    1903

    1

    Posted August 2nd, 2011 10:00

    You will get better overall performance from striped metas.  Yes the backend is already striped, but IO to a single meta on the front-end can still restrict performance, even with 100% cached IO.

    If you won't ever need to expand the meta, use striped metas.  If ease of expansion is more important than performance, use concatenated meta volumes.

  • SAN_AK

    2 Intern

    125 Posts

    1903

    0

    Posted August 2nd, 2011 10:00

    As the tdevs are already striped on back end data devices, is it okay to stripe again for the metas, rather than going for concatenated Metas ?, isn't striping on striping  a issue ?, help me understand.

    Thanks,

  • SKT2

    2 Intern

    1340 Posts

    1903

    0

    Posted September 5th, 2011 14:00

    two levels of stripping accepted, like RAID5 makes the first  , striped metalun makes the second  part ad  creating a file system with striping makes the third one( we avoid the third layer of stripping) ; yes there is a little confustion on scenario like while using Oracle ASM( which again striped internally)

  • Quincy561

    1281 Posts

    1903

    0

    Posted September 5th, 2011 18:00

    Too many levels of striping is a bigger concern for traditional non VP volumes.  For best performance on VP volumes you should use striped metas even though that could be the third level of striping in the Symmetrix.  (RAID =1, VP = 2 META = 3 ), they will have higher performance than a concat meta, but are harder to expand.

  • moe2197

    23 Posts

    1475

    0

    Posted September 21st, 2011 15:00

    Quincy,

    Wondering if you could provide a little more detail regarding your statement "Powers of 2 were a very good idea with thick volumes but is not important for VP volumes."

    I recently read a "VMAX Blueprint" document that indicated that metas should be built based on a TDEV size range and configured using base-2 ... 2,4,8,16,32-way. I also seem to recall John Adams at EMC World saying the same thing ... build metas using base-2 ... but I could be wrong there.

    So what's the final verdict? Are there best practices for meta members/sizes in VP? Can I create a meta with any number of members I wish without any issues?

    Thanks!

  • Quincy561

    1281 Posts

    1031

    0

    Posted September 21st, 2011 18:00

    Yup. :-)

    Metas on thick provisioning have to fit into the backend, this is why the powers of 2 were important.  Since VP metas have nothing to do with the backend layout at all, there is no such need.  A 5 member meta may perform slightly better than a 4 member meta in some cases, and would be fine.  However there is nothing wrong with sticking with powers of 2, since they would also work fine.