On Dec 14, 2006, at 16:00 , Greg Smith wrote:
On Wed, 13 Dec 2006, Ron wrote:
The slowest results, Michael's, are on the system with what
appears to be the slowest CPU of the bunch; and the ranking of the
rest of the results seem to similarly depend on relative CPU
performance. This is not what one would naively expect when
benching a IO intensive app like a DBMS.
pgbench with 3000 total transactions and fsync off is barely doing
I/O to disk; it's writing a bunch of data to the filesystem cache
and ending the benchmark before the data even makes it to the hard
drive. This is why his results become completely different as soon
as the number of transactions increases. With little or no actual
disk writes, you should expect results to be ranked by CPU speed.
I also second your suggestion that pgbench should be run with -S to
disable updates. As far as I can see, nobody has reported numbers for
this setting, so here goes. I also increased the buffer size, which I
found was needed to avoid hitting the disk for block reads, and
increased the memory settings.
My PostgreSQL config overrides, then, are:
shared_buffers = 1024MB
work_mem = 1MB
maintenance_work_mem = 16MB
fsync = off
Environment: Linux 2.6.15-23-amd64-generic on Ubuntu. Dual-core AMD
Opteron 280 with 4GB of RAM. LSI PCI-X Fusion-MPT SAS.
Running with: pgbench -S -v -n -t 5000 -c 5.
Results as a graph: http://purefiction.net/paste/pgbench.pdf
Stats for CFLAGS="-O0": 18440.181894 19207.882300 19894.432185
19635.625622 19876.858884 20032.597042 19683.597973 20370.166669
19989.157881 20207.343510 19993.745956 20081.353580 20356.416424
20047.810017 20319.834190 19417.807528 19906.788454 20536.039929
19491.308046 20002.144230
Stats for CFLAGS="-O3 -msse2 -mfpmath=sse -funroll-loops -m64 -
march=opteron -pipe": 23830.358351 26162.203569 25569.091264
26762.755665 26590.822550 26864.908197 26608.029665 26796.116921
26323.742015 26692.576261 26878.859132 26106.770425 26328.371664
26755.595130 25488.304946 26635.527959 26377.485023 24817.590708
26480.245737 26223.427801
Alexander.
