[rrd-users] Performance over NFS
linux at thehobsons.co.uk
Sun Dec 6 13:10:22 CET 2009
Unfortunately, it's such a common term that I can't find anything
particularly useful via searches.
I'm collecting data on one box (a gateway router), and graphing on
another box - with the directory holding the rrd files exported by
NFS from the gateway. Unfortunately, while it seemed to work when I
first set it up, it's now giving very poor performance - system
upgrades have changed the setup from NFSv3 over UDP to V4 over TCP
(if I remember the details correctly).
The box doing the graphing is running Debian Etch and rrd version
1.2.15-0.3. /etc/fstab contains :
w.x.y.z:/var/rrd /gate nfs ro,soft 0 0
The gateway doing the data collection is running Debian Lenny and rrd
version 1.3.1-4. /etc/exports contains :
The main rrd file that's giving problems is 22M byte in size and
contains 510 data sources, with eight aggregations (Average and Max
for four time periods/resolutions). Some graphs only use a couple of
data sources, one frequently used one uses nearly 200 (in and out
traffic for just under 100 addresses in a stacked graph), one
(infrequently used for good reason) graphs 508 sources (in and out
traffic for 254 addresses).
The rrdtool processes end up spending significant amounts of time
with ps showing them in state "rpc_wa" or "sync_p" (and I've just
seen on in state "get_re"). It's not too bad if only one person is
generating graphs (and waiting for them), but once or twice I've seen
people being impatient and the box gets bogged down (not to mention
swapping heavily) with a dozen of more rrdtool processes running.
Even just one patient user can get the load average up into the teens.
Can anyone give me some hints or pointers on this ?
Visit http://www.magpiesnestpublishing.co.uk/ for books by acclaimed
author Gladys Hobson. Novels - poetry - short stories - ideal as
Christmas stocking fillers. Some available as e-books.
More information about the rrd-users