On 3/11/24 10:09, John Naylor wrote:
> On Tue, Mar 5, 2024 at 9:42 AM David Rowley <dgrowleyml@gmail.com> wrote:
>> performance against the recently optimised aset, generation and slab
>> contexts. The attached graph shows the time it took in seconds to
>> allocate 1GB of memory performing a context reset after 1MB. The
>> function I ran the test on is in the attached
>> pg_allocate_memory_test.patch.txt file.
>>
>> The query I ran was:
>>
>> select chksz,mtype,pg_allocate_memory_test_reset(chksz,
>> 1024*1024,1024*1024*1024, mtype) from (values(8),(16),(32),(64))
>> sizes(chksz),(values('aset'),('generation'),('slab'),('bump'))
>> cxt(mtype) order by mtype,chksz;
>
> I ran the test function, but using 256kB and 3MB for the reset
> frequency, and with 8,16,24,32 byte sizes (patched against a commit
> after the recent hot/cold path separation). Images attached. I also
> get a decent speedup with the bump context, but not quite as dramatic
> as on your machine. It's worth noting that slab is the slowest for me.
> This is an Intel i7-10750H.
That's interesting! Obviously, I can't miss a benchmarking party like
this, so I ran this on my two machines, and I got very similar results
on both - see the attached charts.
It seems that compared to the other memory context types:
(a) bump context is much faster
(b) slab is considerably slower
I wonder if this is due to the microbenchmark being a particularly poor
fit for Slab (but I don't see why would that be), or if this is simply
how Slab works. I vaguely recall it was meant handle this much better
than AllocSet, both in terms of time and memory usage, but we improved
AllocSet since then, so maybe it's no longer true / needed?
regards
--
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
