[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [SAGE] NetApp--spindles vs. performance
Guy> He's specing out a NetApp filer (F810, I believe), and would like
Guy> to know if anyone has any real experience with (or even
Guy> well-educated guesses about) the performance gain of using 36 GB
Guy> disks instead of 72 GB disks. The intended use will be about
Guy> 50/50 for typical PC office-type files & Oracle DBs.
Well, the office files won't be the bottle neck in this case. It all
depends on how fast the Oracle DB needs to be. Remember, the NetApps
have battery backed NVRAM (so writes can be acknowledged before they
get to disk), and the newer NetApps have a fairly large amount of
memory for caching disks reads/writes.
You also want to think about whether you need to have multiple volumes
or not. For the PC data, a smaller group of larger disks will work
just fine. For the oracle data, a larger number of smaller disks will
probably help out a bit, but maybe not. The NetApps have quite a good
performance without much tuning out of the box.
Guy> He needs to balance maximizing performance against minimizing
Guy> cost. Using 36 GB disks costs more (need twice as many disks to
Guy> achieve the 1 TB storage goal), but that might be offset by
Guy> significant performance gains from having twice as many spindles.
Guy> Are there rules of thumb for this sort of thing in general?
Guy> There must be some point of diminishing returns with shrinking
Guy> disk size to increase spindle count (would that I had a system to
Guy> experiment with!). Thanks!
There's also an issue where if you have too many spindles on a single
path (or controller), then you won't get any further performance
increase since the path is saturated. So if he's really worried about
performance, getting two disk shelves with a Fiber Channel controller
for each shelf should be fine.
Then there's the question of how to organize the data. I personally
don't like to use multiple volumes if I can help it, since it makes
growing/shrinking storage availability a pain. qtrees are a great way
to give people a chunk of disk space, but to limit how much they can
use. If they need more, you just change a number in a file and resize
the quota. Plus you get to share snapshot overhead across multiple
groups/sets of data.
But Oracle Databases don't want to be snapshotted except when you
*know* the DB is locked or in hotbackup mode.
So for your case, I'd probably do the following setup:
NetApp 810 (or even look at used F760s with the appropriate licenses)
- pair of FiberChannel controllers
- gigait ethernet
- pair of DS14 disk shelves, one with 72gb disks, the other with 18 or
36gb.
- CIFS and NFS licenses.
I'd setup one volume as the root (vol0) where the PC data would live.
This would be put on the 72gb disks. You would have snapshots turned
on here, with a useful schedule for your needs, along with qtrees to
manage your user's data needs.
Make sure to leave a spare disk.
Then make another volume called 'oracle' out of the smaller disks.
Turn off automatic snapshots, but use them for nightly backups when
scripted.
Make sure to leave a spare disk here as well.
In general, I'd have your friend join the toasters mailing list for a
place to ask more detailed questions about setup and tuning of systems
like this.
John
John Stoffel - Senior Unix Systems Administrator - Lucent Technologies
stoffel@lucent.com - http://www.lucent.com - 978-399-0479