On Tue, 2002-09-10 at 21:44, Jim Wildman wrote:
> It's interesting that the big disk frames (EMC, Hitachi, IBM Shark)
> are (in my experience) configured thusly..
EMC is big on the whole cache thing. Their performance is already so
piss poor (and incredibly overpriced, >$1000/GiB is usual) that they're
forced to throw large RAM caches at it to speed things up and be
competitive on performance.
I have little love for EMC. I have my reasons.
> Each frame is populated with disk 'modules', each of which is a pair
> of SCSI drives, internally mirrored. These present to the frame as
> one drive (ie 2 36G drives look like a single 36G drive). The
> modules are then raided with each other, then presented to the hosts in
> 8G or 14G LUN's or chunks.
The "frame" idea with mirrors is a very EMC mentality thing. For the
longest time, EMC did only mirroring and no RAID whatsoever. Only over
the past couple of years have they really begun to do so, but they're
not real big on the idea. Why let a customer get by on RAID5 when you
can sell them the hardware for a 5 way mirror?
More modern drive arrays present volumes off of RAID5 stripe sets as a
individual SCSI LUN (unique SCSI device instance), and hide stripes from
other machines on the same FCAL loop or FC switch infrastructure using
LUN mapping techniques.
> And the frames will have 4-8G of read cache and 4-8G of write cache.
> Now figure out what the optimal disk layout is....
Right.
> The OS will then usually group them back together into volume
> groups, etc.
If presented with "virtual physical devices", a OS really doesn't need
software volume groups at all. There are, of course, some times when you
might want to have two volumes mirrored in software - but for the most
part this would ideally be handled in the SAN configuration. That's just
part of the reason why you coughed up the big dollars for a SAN in the
first place, right? ;)
I'm really hoping iSCSI explodes. There are currently 3 different iSCSI
initiators (SCSI clients) for Linux at the moment, but no-one has really
set off to write an iSCSI target (SCSI device) server yet. As soon as
this happens, I foresee a great depression in the current high-priced
SAN market (hopefully with the demise of EMC).
You know, an iSCSI target server project really would hit the enterprise
much as SAMBA has. Does anyone know of ANY OpenSource efforts to attempt
this yet?
> Did I get that right?
Yeah, you were spot-on.
- Ian C. Blenke <icblenke@nks.net> <ian@blenke.com>
http://ian.blenke.com
This archive was generated by hypermail 2.1.3 : Fri Aug 01 2014 - 19:15:55 EDT