[prev in list] [next in list] [prev in thread] [next in thread] 

List:       ceph-devel
Subject:    A different Ceph benchmark - pgbench
From:       Mark Kirkwood <mark.kirkwood () catalyst ! net ! nz>
Date:       2012-08-31 5:53:58
Message-ID: 50405176.6050907 () catalyst ! net ! nz
[Download RAW message or body]

I think I had previously mentioned that I am interested in using RBD for 
data backing a database. With the recent notable improvements in IO 
performance for smaller block sizes that came with 0.50 and 0.51 
versions I decided to do some benchmarking with Postgresql.

So here it is:

System : 4x (hyperthread) i3 3.3 Ghz 8G, 2x 128G Crucial M4
Ceph 0.51, 1x osd on 1 M4, 20G journal on the other, ext4 filesystems
Postgresql 9.2 with minimal tuning (shared_buffers = 1G, 
checkpoint_segments = 128)
1x RBD for postgres data, 1x RBD for postgres transaction log, ext4 
filesystems

Scale 300 (approx 5G)

connections | tx per second
------------+--------------
   4          |  287
   8          |  422
   16         |  591
   32         |  779
   64         |  866


Now this is quite an acceptable result (previous attempts had maxed out 
at 4 tx per sec). While it does not compare favourably with running the 
same setup directly on the attached drives (4000 tx per sec with 64 
connections), the difference is a 'mere' factor of 4 instead of orders 
of magnitude, so I am greatly encouraged!

By default Postgreql does IO in 8K blocks using library functions read 
and write - so this *was* a completely pessimal performance test - so it 
is really nice to see how things are improving - keep up the good work!

Regards

Mark

--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
[prev in list] [next in list] [prev in thread] [next in thread]