[SATLUG] lvm2 and raid again
brad at shub-internet.org
Mon Jan 21 23:28:08 CST 2008
On 1/21/08, David Kowis wrote:
> Also, if you want a "turbo Hayabusa" then you should go with a hardware
> raid solution. Raid0. You'd need multiple disks to make a RAID 0
> solution work. Two disks at least.
For a Turbo Hayabusa, you'd need a lot more disks than that. You'll
need 4Gb FCSW SAN networking with wire-speed switches, large numbers
of FC connected 15KRPM SASI disks, and a few terabytes of SSD to act
as a high-speed cache. Oh, and a lot more Intel Xeon quad-core CPUs,
some Infiniband network connections, and a few other things.
Of course, in your case, if you can't spend any more money to upgrade
the system, then the choice really comes down to doing RAID-0 across
the two disks for performance (with the higher risk of failure and
potential data loss, even if you do have nightly backups to a
separate disk system), or RAID-1 across the two disks for increased
reliability and also a certain amount of increased read performance
(write performance may be a bit slower, however).
> Lvm2 will actually slow down your I/O throughput. It does however,
> provide a great deal more flexibility. Software raid is slower than
> hardware raid, because it requires software to make it happen.
In my experience, in most applications software RAID is actually a
lot faster than hardware RAID for normal day-to-day operations, at
least up until you get into the EMC & Hitachi Lightning arena, where
they do things like throw gigabytes upon gigabytes of SSD into the
controllers to act as a large and very, very fast cache, and of
course you're connecting with multiple HBAs over 4Gb FCSW.
Where software RAID usually tends to fall down is when the system is
degraded -- a disk has died, or whatever. It suddenly takes a lot
more CPU to do all the calculations and to try to recover or rebuild
the system, at precisely the same time when you're probably already
suffering a heavier-than-normal load elsewhere.
So, running a software RAID-0 as your primary system and then doing
nightly cron-based backups to another disk system, would actually be
a decent way to get a certain amount of redundancy while also getting
maximum performance -- while still keeping to a budget.
When there is a disk failure, you'll lose everything since the last
backup, but if that's okay for you then I guess there's not too much
we can say against that.
Of course, just writing backups doesn't help you at all. You have to
test them to make sure that they're actually okay and that they will
actually restore properly. You don't want to get several months down
the road, have a disk failure, and then discover that all your
backups are corrupt.
Brad Knowles <brad at shub-internet.org>
LinkedIn Profile: <http://tinyurl.com/y8kpxu>
More information about the SATLUG