In normal customer environment, You do day to day activity like allocation storage to different Operating System. If you do not follow best practice then you will see the host facing performance issue. Because in SAN environment, there are so many things to be consider before we present any SAN storage to any HOST. Now, I am discussing what is the best practice for DMX/SYMM? First understand Customer requirement like what application they are running, what is protection level like DMX support RAID 1/0.RAID 5(3+1) , RAID 5(7+1) etc, Are they using different disk alignment. Once you understand the Customer environment then you need to plan about disk configuration at back-end means on DMX.
In Summary decision need to be taken balance off all:

1)The number of physical disk slices
2)The number of meta volume members
3)Channel capacity
4)Host administration required
5)Performance required
6)Suitability for future expansion

As a simple general guide use the following:
(Ideal) Create 18570 volumes, all volumes striped @1MB in host stripe set
(Good) Create 18570 volumes, create metas with 4-8 members
(OK) Create 18570 volumes, create metas with 16 members
(Warn) Create 18570 volumes, create metas with 17+ members


Smaller split sizes can be used for small datasets, and combined into meta volumes for extra-performance with high-load but small-capacity applications.
Larger split sizes are more suitable for larger datasets.A good tip is to have the volumes for an application spread over a minimum 8 physical disks where possible, with only 1 volume per physical.Note most hosts can only see max 512 host volumes (1 metavolume = 1 host volume, 1 non-meta volume = 1 host volume) per channel.

1 Comments:

  1. Anonymous Says:
  2. Very interesting post; still, some questions arise:
    - it's not too expensive to keep only one volume per physical device? That means a very large overallocated space in the array.
    - do you recommend the usage of disk groups in order to separate disk devices for different workloads? or maybe is better to strip all workloads to all physical disks and let optimizer do it's (re)active job?
    - why EMC does not provide readily available performance papers for DMX as they provide for Clariion? Not all companies that are currently buying EMC DMX are willing to pay for employees education.

Post a Comment