[collectd] collectd versus rrdcollect-remote rrdtool [network] IO

[Florian Forster]

Hi again,

On Tue, Jan 20, 2009 at 02:03:13PM +0100, Greg wrote:
> I have increased again to:
> net.core.rmem_default = 1048576
> net.core.rmem_max = 16777216

> udp    62832      0 10.0.0.11:25826         0.0.0.0:*
> udp   175168      0 10.0.0.11:25826         0.0.0.0:*

okay, this shows that the default value of 124928 was not sufficient. So
that was one source. If the machine isn't doing much else (doesn't have
a ton of other sockets), it shouldn't be a big problem to increase this
value even further. The 4 MBytes (4194304 bytes) I've mentioned earlier
have worked well for my setup.

> 5938 RRD files
> 4GB RAM
> 8 cores @2.4GHz
> 2x 146GB SAS 10K 2.5" RAID1

Sounds okay, though if you expect your installation to grow, you may
want to invest in some RAM. When all RRD files together are
approximately the amount of RAM you have, that's when stuff tends to
become difficult, so buying another 4gig might be the easy way out ;)

> ext3 with default mount options, no tuning

Uh oh, not good. Try mounting the partition, the RRD files are on, with:
  -o noatime,nodiratime,commit=120
This alone should improve performance considerably.

So how's your system handling the load otherwise? How much time is spent
in IO-wait? (Should be close to 12.5% with 8 CPUs.) Do you have any gaps
in the ``local'' data, i. e. the data that was not sent over the
network?

What I'd be most interested in: Does commit 0b6de87f [0] really improve
performance, or was this a wrong guess?

Regards,
-octo

[0] <http://git.verplant.org/?p=collectd.git;a=commit;h=0b6de87f3d08326ee3200db8e16f3738df027ac0>
-- 
Florian octo Forster
Hacker in training
GnuPG: 0x91523C3D
http://verplant.org/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://mailman.verplant.org/pipermail/collectd/attachments/20090120/edd5a092/attachment.pgp