Re: [SAGE] NetApp--spindles vs. performance

[Doug Hughes <doug@eng.auburn.edu>]

On Thu, 16 Jan 2003, Skottie Miller wrote:

>
> These days, with 10K RPM disks, as long as there are "enough" spindles,
> the potential difference between N 72G disks and N*2 36 GB disks on
> a filer doesn't really affect performance.
>
> the number of FC loops, the number of GigE conenctions, the cache size
> and the CPU  and (of course) your workload, will tell.
>
> as for a rule of thumb, with 72 GB disks I try for raid group sizes of
> 9 spindles.  to get 1 TB usable, you'll need at least 16 spindles,
> which means two ds14 disk shelves. with two 9-disk raid groups and
> a hot spare, (19) disks, there will be room to add spindles to the shelves.
>
> get two Fiberchannel adapters, or one dual-port FC adapter, to provide
> two FC loops.
>
> an 810 with two FC loops and 18 active disks will run out of CPU *long*
> before it runs out of spindle ops.
>
To add a little bit more to the mix:
it really depends.. If you are going for max IOPS, then more spindles
really will help (like a TPC setup). If it's a typical oracle sequential
database, then fewer spindles is fine (or better!). Maximizing
IOPS is all about more spindles because you never fill the controllers.
Then again, I haven't done any IOPS workloads on a NetApp..

Wide thin striping is another good use of technology to take maximum
advantage of performance while minimizing administrative overhead.
(http://www.sun.com/solutions/blueprints/1000/layout.pdf)