|Referencing the link https://community.emc.com/message/549242#549242, it has been told that "Have as many as Data devices (smaller size) in the Thin pool. This will help to stripe the I/O across many spindles which will eventually increase the performance"|
|However,the link https://community.emc.com/docs/DOC-8708 says that "The data device sizes should be as large as possible to minimize the number of devices required to encompass the desired overall pool capacity"|
|Could some one please clarify on this, as we are currently planning to use thin provisioning through out for our newly deployed VMAX array and also throw some brief light on how the striping occurs in a thin provisioned environment focusing mainly on how and what factors infulence the spindle selection for a given host write?|
I think the first one is trying to tell you that you don't want 1 or 2 data devices to cover the thin devs. You want your data devices spread over many disks.
The best practice is to limit the number of TDATs on the backend, but to a point. The minimum number you want is eight splits per physical disk. If you have more than one pool on a disk, you want eight splits per pool. If you must have more to use the entire capacity of the disk, use the smallest number to do so.
Striping is at the thin pool allocation size in a round robin manner across the TDATs. 768k is the VP allocation size.
Thanks for the quick reply !!!!!
However I have a few more queries specific to our configuration.
We have in our environment a VMAX array with
Each of our thin pools is associated with one disk group, with each disk group having 64 disks. The over subscription % set for the pool is 100.
Can you please let me know, how the data written by the host is striped across the 64 disks in the pool?
1. Is the striping of the data across the pool (64 disks) happening at the extent level (768KB):
a. Is this 768 KB striped across all the available spindles (64 disks) in the pool or
b. Is this 768 KB allocated to the host from a single spindle.
2. If the above question is deemed true, once the striping happens across all the 64 spindles, will the next write come back to the very first spindle in the pool, assuming that a round robin method is used for striping of the data?
Striping on the disk is at a VP extent size, which is 768K for VMAX 10,20 and 40K. The 768K is allocated from a TDAT device, which is spread over all the disks in the RAID group. In your case that would be 4 drives.
Yes, the process will go around the pool and keep allocating on all 64 disks.