Re: [PATCH] dtrace probes for memory manager

--On 13. November 2009 17:16:15 -0500 Robert Haas <robertmhaas@gmail.com> 
wrote:

> Don't think.  Benchmark.  :-)
>
> (If you can measure it at all, it's too much, at least IMHO.)

I've tried to benchmark this now on my (fairly slow compared to server 
hardware) MacBook and see some negative trend for those memory probes in 
pgbench. Running dozens of rounds with pgbench (scale 150, 10 clients / 
1000 transactions) gives the following numbers (untuned PostgreSQL config):

AVG(tps) with dtrace memory probes: 31.62 tps
AVG(tps) without dtrace memory probes: 38.24 tps

So there seems to be a minor slowdown at least on *my* machine. However, it 
would be fine if someone can prove these numbers..

What do you guys think, what other tests/parameters can be invoked to test 
for an impact ?

-- 
Thanks
Bernd