Plattform comparison (lies, damn lies and benchmarks) - Mailing list pgsql-performance

From Guido Neitzer
Subject Plattform comparison (lies, damn lies and benchmarks)
Date
Msg-id B821BB33-FC48-4348-8698-55C4AD76C0C0@event-s.net
Whole thread Raw
List pgsql-performance
Hi.

After I had my hands on an Intel MacBook Pro (2 GHz Core Duo, 1GB
RAM), I made some comparisons between the machines I have here at the
company.

For the ease of it and the simple way of reproducing the tests, I
took pgbench for the test.

Konfigurations:

1. PowerMac G5 (G5 Mac OS X) with two 1.8 CPUs (not a dual core),
1.25GB RAM, Mac OS X Tiger 10.4.8, Single S-ATA harddrive, fsync on

2. PowerMac G5 from above but with Yellow Dog Linux 4.1

3. MacBook Pro, 2GHz Core Duo, 1GB RAM, Mac OS X Tiger 10.4.8,
internal harddrive (5k4, 120GB).

PostgreSQL version is 8.2beta3 compiled with same settings on all
plattforms, on Mac OS X Spotlight was turned off, same memory
settings on all plattforms (320MB of shmmax on Mac OS X, 128MB
shared_buffers for PostgreSQL).

Here we go:

Results with 2 concurrent connections:

G5 Mac OS X: 495
G5 YD Linux: 490 - 520
MBP X: 1125

Results with 10 concurrent connections:

G5 Mac OS X: 393
G5 YD Linux: 410 - 450
MBP: 1060

Results with 50 concurrent connections:

G5 Mac OS X: 278
G5 YD Linux: 232
MBP X: 575

Results with 90 concurrent connections:

Mac OS X: 210
YD Linux: 120
MBP X:  378

The tests were taken with:

[cug@localhost ~]$ for n in `seq 0 9`; do pgbench -U postgres -c 10 -
t 100 benchdb; done  | perl -nle '/tps = (\d+)/ or next; $cnt++; $tps
+=$1; END{ $avg = $tps/$cnt; print $avg }'

Yesterday a friend had a chance to test with a 2.16GHz MacBook Pro
Core 2 Duo (Mac OS X, 5k4 160GB internal harddrive):

10 connections: ~1150 tps
50 connections: ~640 tps

To quantify the performance hit from the harddrive we tested also
with fsync off:

10 connections: ~1500 tps
50 connections: ~860 tps

The G5 with fsync off had only 5% more performance, so the harddrive
didn't have such a high impact on the performance there.

Okay, nothing really special so far, but interesting enough. Only
wanted to share the results with you.

cug

pgsql-performance by date:

Previous
From: Guido Neitzer
Date:
Subject: Re: shared_buffers > 284263 on OS X
Next
From: "Merlin Moncure"
Date:
Subject: Re: Massive delete of rows, how to proceed?