Re: Testing Sandforce SSD - Mailing list pgsql-performance

Greg Smith wrote:
> Note that not all of the Sandforce drives include a capacitor; I hope
> you got one that does!  I wasn't aware any of the SF drives with a
> capacitor on them were even shipping yet, all of the ones I'd seen
> were the chipset that doesn't include one still.  Haven't checked in a
> few weeks though.
I think I did, it was expensive enough, though while ordering its very
easy to order the wrong one, all names on the product category page look
the same. (OCZ Vertex 2 Pro)
>> * How to test for power failure?
>
> I've had good results using one of the early programs used to
> investigate this class of problems:
> http://brad.livejournal.com/2116715.html?page=2
A great tool, thanks for the link!

  diskchecker: running 34 sec, 4.10% coverage of 500 MB (1342 writes; 39/s)
  diskchecker: running 35 sec, 4.24% coverage of 500 MB (1390 writes; 39/s)
  diskchecker: running 36 sec, 4.35% coverage of 500 MB (1427 writes; 39/s)
  diskchecker: running 37 sec, 4.47% coverage of 500 MB (1468 writes; 39/s)
didn't get 'ok' from server (11387 316950), msg=[] = Connection reset by
peer at ./diskchecker.pl line 132.

here's where I removed the power and left it off for about a minute.
Then started again then did the verify

yeb@a:~$ ./diskchecker.pl -s client45.eemnes verify test_file
 verifying: 0.00%
Total errors: 0

:-)
this was on ext2

>> * What filesystem to use on the SSD? To minimize writes and maximize
>> chance for seeing errors I'd choose ext2 here.
>
> I don't consider there to be any reason to deploy any part of a
> PostgreSQL database on ext2.  The potential for downtime if the fsck
> doesn't happen automatically far outweighs the minimal performance
> advantage you'll actually see in real applications.
Hmm.. wouldn't that apply for other filesystems as well? I know that JFS
also won't mount if booted unclean, it somehow needs a marker from the
fsck. Don't know for ext3, xfs etc.
> All of the benchmarks showing large gains for ext2 over ext3 I have
> seen been synthetic, not real database performance; the internal ones
> I've run using things like pgbench do not show a significant
> improvement.  (Yes, I'm already working on finding time to publicly
> release those findings)
The reason I'd choose ext2 on the SSD was mainly to decrease the number
of writes, not for performance. Maybe I should ultimately do tests for
both journalled and ext2 filesystems and compare the amount of data per
x pgbench transactions.
> Put it on ext3, toggle on noatime, and move on to testing.  The
> overhead of the metadata writes is the least of the problems when
> doing write-heavy stuff on Linux.
Will surely do and post the results.

thanks,
Yeb Havinga