[rrd-users] Performance over NFS

Simon Hobson linux at thehobsons.co.uk
Tue Dec 8 13:35:18 CET 2009

Tobias Oetiker wrote:

>  > >also try exporting nfs with the async option ... less of a data
>>  >integrity guarantee, but it should be faster ...
>>  As I read the man page, that'll only affect writes. I've exported the
>>  directory as read-only - and it doesn't appear to have any effect.
>ah it is the graphing which is slow not the updating ?

Yes, sorry if I didn't make that clear. Updates are fine (local 
filesystem), graphing is slow.

Just as a check, I did a simple copy of the file from one machine to 
the other :
Via NFS it took around 90 to 120 seconds, via SCP around 30-35 
seconds. A bit slow for a 22MByte file ... hmm, 648kbyte/s, that 
sounds rather close to our internet connection speed ...

At which point, that clinking noise is the sound of the penny 
dropping ! A while ago, as well as doing some upgrades I installed 
keepalived and setup two routers in an active/spare setup. We do 
traffic shaping on the routers, and the traffic shaping is configured 
to allow (in practical terms) unrestricted rate from the routers IP 
addresses*. And guess what I forgot to do when I changed the IP 
address :-/

Strangely, now I've fixed the traffic shaping, things are back to 
normal (copying the file takes about 4s) - so it wasn't NFS at all.

I'll just wander off and clean this egg off my face then ...

* To shape inbound traffic, we shape egress on the internal 
interface, but of course we need to allow for traffic that originates 
on the firewall itself. By forgetting to change the traffic shaping 
setup, we were including the rrd data in the inbound traffic shaping.

Simon Hobson

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 mailing list