Unsolved
This post is more than 5 years old
46 Posts
0
664
VP and preallocation
Is there a benefit to preallocating a portion of storage ahead of time, other than the initial write to a block (subsequent writes there is no difference). I'm thinking more along the lines of sequential I/O's and being able to get better cache hits. My app will grow into ~30TB of space (give or take 10-20% because they guesstimate so well), if I were to preallocate 10-20TB of this space will there be a greater chance of the drives being in a better position for read-ahead performance? I'm guessing that if I preallocate that the DMX could throw the next block into cache since the head is already there and would simply be a read of the next block.
I know thinking about this is a bit counter-intuitive to the supposed benefits of VP (just let VP do it and don't worry about thinking of it); but I'm still wondering since I can't run optimizer against VP data devices and I've had previous issues with seeks per sec and hypers per seek.
I know thinking about this is a bit counter-intuitive to the supposed benefits of VP (just let VP do it and don't worry about thinking of it); but I'm still wondering since I can't run optimizer against VP data devices and I've had previous issues with seeks per sec and hypers per seek.
Scott-sd_69
2 Posts
1
April 16th, 2009 09:00
We use VP to get "wide stripping".. We take an empty dmx and allocate all splits (5/spindle @300gig and 17/spindle @1tb) to VP pools (Tier1 and Tier2)...
Every lun we then create is fully allocated. we do so because performance is king. Also, because we cant be wrong because of our budget lead times - it's one thing to tell a potential customer they can't have new space, its a disaster to tell a customer that you given a thin volume to we're over allocated and no new spindles are coming to cover..
Over provisioning has its place and is terrific for those companies that can write a check when more physical spindles can be purchased..
We've seen zero issues with seeks per sec and hypers/seek.. what we did see is that (as of last year), the # of tdevs/metalun mattered.. we always allocate a minimum of 15 tdevs per meta lun. Remember:No stripping on metaluns!
MikeMac1
292 Posts
0
May 6th, 2009 07:00