I'm looking for performance tuning tips for NetApps. We're looking to upgrade from F760's with 9GB 7200rpm drives to F840's with either 18GB or 36GB drives. Due to the performance characterics of the system, more storage is not always better, but if I can work out a reduction in the number of spindles required to serve the same data, we'll save space and power in the datacenter.
Degraded mode performance is also critical: we need to be able to reconstruct a failed disk in a reasonable amount of time, with minimal performance impact.
My thinking is that performance and space requirements drive raid group size up, and degraded mode performance requirements drive raid group size down. I'm looking to find the balance.
The relevant bottlenecks are: 1) CPU 2) Bus bandwidth to disk 3) Individual disk ops/sec bandwidth
The questions I need to answer are:
1) How much CPU is required to do a particular number of ops/second? 2) How much data can I cram through the buses to the disk? 3) How many spindles do I need? 4) What should my raid size be? 5) How much of a performance hit will I take in degraded mode?
Has anyone out there made an effort to solve the problem on this level? I'm working on answering 3) 4) and 5) above, and assume that I can derive results from benchmarks and specs to answer 1) and 2). To answer the spindle question, I need to know how many sequential block reads/sec, random block reads/sec, and sequential block writes/sec I can do with each disk.
Any pointers you have would be helpful.
Thanks,
-Alf