Re: Xeon twice the performance of opteron

On Mar 17, 2011, at 9:39 PM, Scott Marlowe wrote:

>
> My experience puts the 23xx series opterons in a same general
> neighborhood as the E5300 and a little behind the E5400 series Xeons.
> OTOH, the newer Magny Cours Opterons stomp both of those into the
> ground.
>
> Do any of those machines have zone.reclaim.mode = 1 ???
>
> i.e.:
>
> sysctl -a|grep zone.reclaim
> vm.zone_reclaim_mode = 0
>
> I had a machine that had just high enough interzone communications
> cost to get it turned on by default and it slowed it right to a crawl
> under pgsql.


It is set to zero on this machine.

I've tried PG compiled on the box itself, same result.

As for power savings, according to cpuinfo all the cores are running
at 2.1ghz

We had another machine which typically runs as a web server running on
an AMD Opteron(tm) Processor 6128
which after diddling the speed governor to performance (thus bumping
cpu speed to 2ghz from 800mhz) query speed increased to 100ms, still
not as fast as the xeon, but close enough.

I think I'm just hitting some wall of the architecture. I tried
getting some oprofile love from it but oprofile seems to not work on
that box. however it worked on the xeon box:
33995     9.6859  postgres                 j2date
21925     6.2469  postgres                 ExecMakeFunctionResultNoSets
20500     5.8409  postgres                 slot_deform_tuple
17623     5.0212  postgres                 BitmapHeapNext
13059     3.7208  postgres                 dt2time
12271     3.4963  postgres                 slot_getattr
11509

aside from j2date (probably coming up due to that Join filter I'd
wager) nothing unexpected.


--
Jeff Trout <jeff@jefftrout.com>
http://www.stuarthamm.net/
http://www.dellsmartexitin.com/