[Nagiosplug-devel] novel idea for performance optimization

[sean finney]

hi carlo,

On Sun, May 08, 2005 at 10:28:41PM -0500, Carlo Marcelo Arenas Belon wrote:
> my guess is that most of the C plugins are small enough and designed to be
> standalone programs, that this migration wouldn't show on them a big
> improvement and the risk/cost would be too high.

maybe not on a system that isn't feeling the burn of all the
fork/exec calls, but i'll bet money that it will make a notable
difference on a system where hundreds or thousands of devices
are being polled.  i may do a proof of concept just to see
how true this is.

> the perl plugins (big interpreter spawned and code "re-compiled" on every
> excecution) and shell plugins (with several different processes spawned
> inside) which are the ones which are taking more of the CPU anyway are not
> getting any improvement at all.

this is true.  most notably, the snmp based perl plugins are real
hogs.  i think the clear answer is to eventually implement them
in C :)

> as for making a new standard interface based on DSOs (basically an stable
> API which you could dynamic link probes to) without making it mandatory so
> that new probes could start migrating to it on their own, i am all for it.

the nice thing about the 'ld voodoo' i suggest is that no code needs
to be re-written[1] in any of the existing plugins.  anyway, i'll
see about putting some code where my mouth is before wasting anyone
else's time :)



	sean

[1] minus fixes for memory leaks and being thread-safe, i suppose.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://nagios-plugins.org/archive/devel/attachments/20050508/d502afab/attachment.sig>