AVM Storage Pool Selection, Performance?

Hello All,

Been researching in the forums and have read the AVM PDFs and have yet to find a definitive answer on some of these quesitons, so any assistance would be great before I engage EMC support directly. I hope this discussion will help all.

We are about to deploy our NS-480 for the VMWare over NFS use case. Our primary concern is performance. From what I understand, AVM is the way to go, however, our storage pool choice is causing us some consternation.

For instance, lets assume that I have about 80 15K 300GB SAS drives in my backend Clariion. I have options of many pools, but for now lets narrow it down to just clar_r5_performance and clar_r5_economy.

If I create a clar_r5_performance using 4 disk groups of 4+1 RAID5, then it is my understanding that, when I add a filesystem, AVM will look for available space in two of those disk groups, add a slice on each (balanced across SPs if posible), and then stripe them together. As a result, that filesystem gets the performance of 10 spindles. While this should be sufficient, it does have concern for some of our higher performance workloads, or if we need to scale a filesystem beyond those spindles. What is the best way to handle this? Can I move a filesystem to a new storage group? Can I restripe the filesystem over more disk groups?

Is there any way AVM would create a filesystem and stripe across all the disk groups in the storage pool? Say 4 or 8 disk groups in the clar_r5_performance pool. I understand that we incur a greater risk as if any RAID group backing that filesystem suffers a double-drive failure, we lose the data, but we have contingencies for that.

Also, does clar_r5_economy provide a significant IOPS benefit over clar_r5_performance? It's my understanding that each individual filesystem would get 18 spindles instead of 10  so I would get more IOPS there for my more demanding applications. This is of course assuming two filesystems didn't reside on the same spindles and shared  them of course, we plan to use is_greedy to help avoid this. I could also use a RAID10 storage pool at this point too if necessary, but this is discouraged as these applications are likely to require high capacity but low write performance, mostly high read performance is required and the expense of RAID10 is not really necessary when RAID5 is sufficient if I can get the spindle count adequate to support the load.

For these applications, read performance matters more than extreme data protection (as we are replicating and several other methods to protect). If this is the case then, would clar_r5_economy provide better performance than clar_r5_performance or is there some special aspect of AVM that I am missing?

Also, am I worrying too much? Is there some magic beyond write caching, etc. that makes AVM performance scale linearly as spindles are added, or do I just make special exceptions for those virtual machines that require more than 10 spindles of performance (e.g. a RAID10 pool or manual filesystems created on large metavolume or RAID10 groups)?

I just want to make sure I'm not missing something here and fully understand the strengths and limitations of AVM, and avoid problems down the road if possible. I really don't anticipate any individual VM needing more than 18 spindles of IOPS performance, but I would like to get it as optimized as possible in case one does

Thanks, I know that was long and rambling so if you can answer even a subset, I would appreciate it.