Pavan Deolasee wrote:
> The numbers are not that bad, but definitely not as good as we saw
> on some other platforms.
Well, this machine is definitely designed to hold up under high
concurrency. As I understand it, each core is the memory manager
for two 4GB DIMMs, with two channels to them,
On Tue, Nov 22, 2011 at 4:40 AM, Kevin Grittner
wrote:
> Pavan Deolasee wrote:
>
>> It will be a great help if you could spare few minutes to also
>> test the patch to take out the frequently accessed PGPROC members
>> to a different array. We are seeing good improvements on HPUX IA
>> platform a
Pavan Deolasee wrote:
> It will be a great help if you could spare few minutes to also
> test the patch to take out the frequently accessed PGPROC members
> to a different array. We are seeing good improvements on HPUX IA
> platform and the AMD Opteron and it will be interesting to know
> what h
On Mon, Nov 21, 2011 at 11:01 PM, Kevin Grittner
wrote:
> Pavan Deolasee wrote:
>
>> It will be a great help if you could spare few minutes to also
>> test the patch to take out the frequently accessed PGPROC members
>> to a different array. We are seeing good improvements on HPUX IA
>> platform
Pavan Deolasee wrote:
> It will be a great help if you could spare few minutes to also
> test the patch to take out the frequently accessed PGPROC members
> to a different array. We are seeing good improvements on HPUX IA
> platform and the AMD Opteron and it will be interesting to know
> what h
On Mon, Nov 21, 2011 at 10:44 PM, Kevin Grittner
wrote:
> "Kevin Grittner" wrote:
>
>> I can run one more set of tests tonight before I have to give it
>> back to the guy who's putting it into production. It sounds like
>> a set like the above except with synchronous_commit = off might be
>> des
"Kevin Grittner" wrote:
> I can run one more set of tests tonight before I have to give it
> back to the guy who's putting it into production. It sounds like
> a set like the above except with synchronous_commit = off might be
> desirable?
OK, that's what I did. This gave me my best numbers
Robert Haas wrote:
> I was actually thinking it would be interesting to oprofile the
> read-only test; see if we can figure out where those slowdowns are
> coming from.
CPU: Intel Core/i7, speed 2262 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Clock cycles when not halted) with
a unit mas
Robert Haas wrote:
> Hmm. There's obviously something that's different in your
> environment or configuration from what I tested, but I don't know
> what it is. The fact that your scale factor is larger than
> shared_buffers might matter; or Intel vs. AMD. Or maybe you're
> running with synchr
On Saturday, November 19, 2011 12:18:07 AM Kevin Grittner wrote:
> Andres Freund wrote:
> > I think opannotate -a -s produces output with instructions/code
> > intermingled.
>
> Thanks. I'll check out perf later (thanks for the tips!), but for
> now, here's the function which was at the top of m
On Fri, Nov 18, 2011 at 6:46 PM, Kevin Grittner
wrote:
>>> tps = 21946.961196 (including connections establishing)
>>> tps = 22911.873227 (including connections establishing)
>>>
>>> For write transactions, that seems pretty respectable.
>>
>> Very. What do you get without the patch?
>
> [quick r
Robert Haas wrote:
> Hmm. That looks a lot like a profile with no lock contention at
> all. Since I see XLogInsert in there, I assume this must be a
> pgbench write test on unlogged tables? How close am I?
Not unless pgbench on HEAD does that by default. Here are the
relevant statements:
On Fri, Nov 18, 2011 at 2:05 PM, Kevin Grittner
wrote:
> Robert Haas wrote:
>> Any chance you can run oprofile (on either branch, don't really
>> care) against the 32 client test and post the results?
>
> [ oprofile results ]
Hmm. That looks a lot like a profile with no lock contention at all.
Andres Freund wrote:
> I think opannotate -a -s produces output with instructions/code
> intermingled.
Thanks. I'll check out perf later (thanks for the tips!), but for
now, here's the function which was at the top of my oprofile
results, annotated with those options. I'm afraid it's a bit
i
On Friday, November 18, 2011 11:12:02 PM Andres Freund wrote:
> On Friday, November 18, 2011 09:16:01 PM Kevin Grittner wrote:
> > Andres Freund wrote:
> > > When doing line-level profiles I would suggest looking at the
> > > instructions.
> >
> > What's the best way to do that?
>
> I think opan
On Friday, November 18, 2011 09:16:01 PM Kevin Grittner wrote:
> Andres Freund wrote:
> > When doing line-level profiles I would suggest looking at the
> > instructions.
> What's the best way to do that?
I think opannotate -a -s produces output with instructions/code intermingled.
> > I don't thi
Andres Freund wrote:
> When doing line-level profiles I would suggest looking at the
> instructions.
What's the best way to do that?
> I don't think cache line contention is the most likely candidate
> here. Simple cache-misses seem far more likely. In combination
> with pipeline stalls...
On Friday, November 18, 2011 08:36:59 PM Kevin Grittner wrote:
> "Kevin Grittner" wrote:
> > samples %image name symbol name
> > 4954633.6718 postgreshash_search_with_hash_value
>
> When lines like these show up in the annotated version, I'm
> impressed that we're still
Robert Haas wrote:
>> I think so. My take was that it was showing 32 of 64 *threads*
>> active -- the hyperthreading funkiness. Is there something in
>> particular you'd like me to check?
>
> Not really, just don't understand the number.
I'm having trouble resolving the vmstat numbers I got
"Kevin Grittner" wrote:
> samples %image name symbol name
> 4954633.6718 postgreshash_search_with_hash_value
When lines like these show up in the annotated version, I'm
impressed that we're still finding gains as big as we are:
44613 0.3306 :if (segp == NU
"anara...@anarazel.de" wrote:
> Kevin Grittner schrieb:
>>samples %image name symbol name
>>9333944.9651 postgresAllocSetAlloc
>>8484764.5134 postgresbase_yyparse
>>7195153.8274 postgresSearchCatCache
> That profile looks like you ran pgben
Kevin Grittner schrieb:
>Robert Haas wrote:
>
>> Any chance you can run oprofile (on either branch, don't really
>> care) against the 32 client test and post the results?
>
>Besides the other changes we discussed, I boosted scale to 150 and
>ran at READ COMMITTED isolation level (because all
Robert Haas wrote:
> Any chance you can run oprofile (on either branch, don't really
> care) against the 32 client test and post the results?
Besides the other changes we discussed, I boosted scale to 150 and
ran at READ COMMITTED isolation level (because all threads promptly
crashed and burne
Robert Haas wrote:
> Yeah, I'd just drop -S.
Easily done.
> Make sure to use -c N -j N with pgbench, or you'll probably not be
> able to saturate it.
Yeah, that's part of the script I copied from you.
> I've also had good luck with wal_writer_delay=20ms, although if
> you have synchronou
On Fri, Nov 18, 2011 at 12:45 PM, Kevin Grittner
wrote:
> OK. Sorry for misunderstanding that. I haven't gotten around to a
> deep reading of the patch yet. :-( I based this on the test script
> you posted here (with slight modifications for my preferred
> directory structures):
>
> http://arc
Robert Haas wrote:
> Kevin Grittner wrote:
>>> Then again, is this a regular pgbench test or is this
>>> SELECT-only?
>>
>> SELECT-only
>
> Ah, OK. I would not expect flexlocks to help with that; Pavan's
> patch might, though.
OK. Sorry for misunderstanding that. I haven't gotten around t
On Fri, Nov 18, 2011 at 12:03 PM, Kevin Grittner
wrote:
>> Then again, is this a regular pgbench test or is this SELECT-only?
>
> SELECT-only
Ah, OK. I would not expect flexlocks to help with that; Pavan's patch
might, though.
>> Can you by any chance check top or vmstat during the 32-client
>>
"Kevin Grittner" wrote:
> We have a 32-core Intel box (4 x X7560 @ 2.27GHz) with 256 GB
> RAM.
In case anyone cares, this is the same box for which I posted STREAM
test results a while back. The PostgreSQL tests seem to peak on
this 32-core box at 64 clients, while the STREAM test of raw RAM
Robert Haas wrote:
> Then again, is this a regular pgbench test or is this SELECT-only?
SELECT-only
> Can you by any chance check top or vmstat during the 32-client
> test and see what percentage you have of user time/system
> time/idle time?
You didn't say whether you wanted master or fle
On Fri, Nov 18, 2011 at 11:26 AM, Kevin Grittner
wrote:
> Robert Haas wrote:
>> Nate Boley's AMD 6128 box (which has 32 cores) and an HP Integrity
>> server (also with 32 cores).
>
>> [clear improvement with flexlock patch]
>
> Hmm. We have a 32-core Intel box (4 x X7560 @ 2.27GHz) with 256 GB
>
Robert Haas wrote:
> Nate Boley's AMD 6128 box (which has 32 cores) and an HP Integrity
> server (also with 32 cores).
> [clear improvement with flexlock patch]
Hmm. We have a 32-core Intel box (4 x X7560 @ 2.27GHz) with 256 GB
RAM. It's about a week from going into production, at which p
31 matches
Mail list logo