HA comparison comments

[ft@angels.att.com]

On Wed, 25 Jun 1997, Doug Hughes wrote:
> >I cannot speak for SGI's HA solutions, but we
> >have Origin 2000's with 4 CPU's and 512mb of memory
> >and Sun 4000's with 4 CPU's and 512mb of memory.
> >The Origins are an order of magnitude faster than
> >the Sun machines.
>
> Ah, but are you sure that this isn't a case of xfs  vs. old UFS?

I can say firsthand that xfs has very nice performance characteristics
(although vxfs gives it a run for its money).  I have been impressed by
the performance of xfs "out of the box" (without a lot of work tuning for
cylinder/stripe alignment, allocation unit alignment, etc.).  I think vxfs
is just growing into that; each release has measurably fewer knobs, and
does a pretty good job of aligning itself (we used to have to spend
considerable time in vxtrace).  However, FWIW, we have experienced our
fair share of nasty behavior (data corruption, etc.) under xfs.  Here is
an excerpt from a note I was copied on yesterday:

"    When I ran my regression tests on corona, a simple
(and I mean very simple) transaction rollback (incredibly)
caused one of my files to go from about 80 chars to 4GB.
On corona at least, XFS is grounds for a law suit. Just wanted
to let you know -- if XFS is working for you now, you may
not want to take any SGI `patches'."

Not to lob barbs at SGI, but this has been a topic of concern for some of
my colleagues for some time.

Fred

--
Fred True				"My name is Ozymandias, King of Kings:
AT&T World Class Database Development	 Look on my works, ye Mighty,
ft@angels.att.com			 and despair!"
						-P. B. Shelley