









|
[Date Prev][Date Next]
[Chronological]
[Thread]
[Top]
[lists] Re: [NMLUG] More RAID Questions
Something I meant to add to my last post. Get a couple of removeable
drive cages for your RAID drives. Label everything clearly. If you
prep someone at the remote site, they can do drive swaps for you with
phone support. Saves tons of hassle. Regardless of needing special
hardware, doing RAID is worth spending a bit on a good hardware
controller. If you get one with an onboard cache, you'll definitely
notice an improvement on a file/print server over doing it in software.
Besides, if you're using just the onboard IDE and it dies, you're down
to replacing the motherboard or adding a controller anyhow.
Yes, RAID 5 is the least overhead for parity but RAID 1 is a better
tradeoff if performance is an issue.
Remember this too...with software RAID, failover is a manual process,
with hardware it's generally just swap a hotpluggable drive cage and
you're done. I guess it all depends on how hands-on you can be and how
much downtime you can tolerate when a problem occurs.
--
Kelly Wilson
Ed Heron wrote:
My 2 cents...
Software RAID eliminates the need for special hardware. I also
perceive that as a good thing.
In either case, hardware or software, RAID re-syncing should be low
priority and not reduce system performance. However, I'd want to test
it if it were important.
RAID 5 (3 drives or more) should be the preferred goal. You would get
both mirroring, to avoid drive failures, and stripping, to increase
device capacity cheaply. However, unless you need a hugmongeous single
partition, RAID 1 really should do you.
I evaluated the RAID 1 idea briefly. I concluded that mirrored servers
would be the true ideal, from a fault-tolerant point of view. As soon
as I resolve configuration issues for all of the software I'll be
running on my new servers, I'll be looking into methods for
accomplishing it. Many of my servers are more than 2 hours away and a
failure would significantly degrade services. RAID 1 only helps the
system survive if the failure is a hard drive.
>
|
|