Re: spinlocks on HP-UX

On Tue, Oct 18, 2011 at 12:11 AM, Tatsuo Ishii <ishii@postgresql.org> wrote:
>>> That would be great.  What I've been using as a test case is pgbench
>>> -S -c $NUM_CPU_CORES -j $NUM_CPU_CORES with scale factor 100 and
>>> shared_buffers=8GB.
>>>
>>> I think what you'd want to compare is the performance of unpatched
>>> master, vs. the performance with this line added to s_lock.h for your
>>> architecture:
>>>
>>> #define TAS_SPIN(lock)  (*(lock) ? 1 : TAS(lock))
>>>
>>> We've now added that line for ia64 (the line is present in two
>>> different places in the file, one for GCC and the other for HP's
>>> compiler).  So the question is whether we need it for any other
>>> architectures.
>>
>> Ok. Let me talk to IBM guys...
>
> With help from IBM Japan Ltd. we did some tests on a larger IBM
> machine than Tom Lane has used for his
> test(http://archives.postgresql.org/message-id/8292.1314641721@sss.pgh.pa.us).
> In his case it was IBM 8406-71Y, which has 8 physical cores and
> 4SMT(32 threadings). Ours is IBM Power 750 Express, which has 32
> physical cores and 4SMT(128 threadings), 256GB RAM.
>
> The test method was same as the one in the article above. The
> differences are OS(RHEL 6.1), gcc version (4.4.5) and shared buffer
> size(8GB).
>
> We tested 3 methods to enhance spin lock contention:
>
> 1) Add "hint" parameter to lwarx op which is usable POWER6 or later
>   architecure.
>
> 2) Add non-locked test in TAS()
>
> 3) #1 + #2
>
> We saw small performance enhancement with #1, larger one with #2 and
> even better with #1+#2.

Hmm, so you added the non-locked test in TAS()?  Did you try adding it
just to TAS_SPIN()?  On Itanium, I found that it was slightly better
to do it only in TAS_SPIN() - i.e. in the contended case.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company