Re: storage server & scalability
On Thursday 23 November 2006 17:54, Qweb - Yavuz Aydin wrote:
<snip>
> Issue 1:
> It seems obvious that SCSI is the way to go. However, because of the
> simple fact that (S)ATA provides much more capacity (and is far less
> expensive) we would like to know what you think of using SATA drives
> in a storage server.
Well, you do talk about web services - we can therefore assume that the I/O
will not be very intensive (unless you're ebay.com). SATA should be able to
handle this.
SATA disks are generally regarded as less robust than FC or SCSI disks, but
the price difference makes having more disposable/standby drives worth using
SATA for such scenarios.
> And what RAID level would be a good choice?
SATA disks are big and slow. This means that a single drive failure on a
RAID-5 array could leave the array in degraded mode for hours, until a
hot-spare has fully synced. You obviously want to minimise the time your
array is running in degraded mode, since you loose everything in the event of
a 2nd drive failure before array reconstruction completes. So RAID-6 is the
way to go. N+2 is not a big performance penalty, especially for big disk
arrays.
<snip>
> Issue 2:
> Another question which remains unanswered for us is how one would
> scale storage. For example, if we setup a storage server with RAID 5
> or RAID 6, how can we extend the RAID array with more capacity,
> without losing data? Would it be as simple as plugging in another disk
> at the array?
You need LVM2 to do this. Create a RAID array of physical disks and then
create LVM volumes on top of that. When you need to add a (physical) disk you
will have to somehow add it to the RAID array (or create a new RAID array on
the side) and extend the LVM volume to use the new space. Check out
http://evms.sourceforge.net/ for a system that promises to make all that less
painful.
Cheers
-A
Reply to: