In article ,
Faeandar wrote:

I disagree with this. Performance is never better than with local
scsi storage. DAS is still king when it comes to raw performance.

Well, you'd think so, but it's not actually true. For one thing,
you can't really fit eighty or a hundred spindles inside that server case,
and if you find an external enclosure that works you'll need to move to
fibre channel anyway and lose the theoretical speed advantage of a direct
scsi 320 connection. (Maybe if they actually *had* scsi 1280 I'd agree
with you :-) And a SAN is (at least usually) a lot more than just an external
storage cabinet and a fibre strand. These things have multiple raid levels,
even custom raid arrangements, gigabytes of battery backed cache and
sophisticated caching algorithms (one reason throughput to the SAN can be
faster than direct I/O to a local disk. You don't actually hit a disk),
multiple hot spares, redundant controller heads, all kinds of stuff that's
pretty hard to engineer into a standalone server. Couple that level
of architecture with an FC fabric and it's hard to argue that any
kind of point storage is technically superior.

Reliability is actually better in most DAS environments because you
have fewer bits in the middle. No shared cache to get corrupted. No
zoning to go bat**** and lip storm. Things like that.


Hmmmm, well, do you think that an FC fabric is inherently less
reliable than a 40 foot scsi cable? I'm not sure what you mean by shared
cache corruption; the cache in something like an EMC just doesn't somehow
"get corrupted" any more than your server memory somehow "gets corrupted".