M
Mark Seger
I don't know if anyone on this list is interested or not, but since it's
written in perl I thought perhaps someone might be. Collectl,
available at http://collectl.sourceforge.net/ is a performance
monitoring tool I wrote about 4 years ago and have since released as
Open Source this past summer.
It basically replaces the functionality of virtually all the existing
'stat' utilities (eg vmstat, nfsstat, memstat, iostat, etc), sar and
then some. It even includes statistics for Infiniband and Lustre file
systems. Like many tools collectl can run interactively or log data to
a rolling set of files and later play them back, but rather than have to
have a bunch of windows open each running a different tool, collectl
does it all and reports its output in several different but consistent
formats. One format, which I call 'plot format', is actually space
separated data that can be consumed by gnuplot or even read by excel.
Another format can be consumed by rrdtool if you're of that persuasion.
It can even run at subsecond monitoring intervals and if you think
that's not important consider that most linux kernels do NOT correctly
report network statistics if monitored once a second but rather you need
to monitor at 0.9765 seconds (see the website for a detailed explanation).
The best thing is it uses less than 0.1% cpu when collecting samples
every 10 seconds as well as process and slab statistics every 60 seconds.
Anyhow, if you're a linux person take it for a spin.
-mark
written in perl I thought perhaps someone might be. Collectl,
available at http://collectl.sourceforge.net/ is a performance
monitoring tool I wrote about 4 years ago and have since released as
Open Source this past summer.
It basically replaces the functionality of virtually all the existing
'stat' utilities (eg vmstat, nfsstat, memstat, iostat, etc), sar and
then some. It even includes statistics for Infiniband and Lustre file
systems. Like many tools collectl can run interactively or log data to
a rolling set of files and later play them back, but rather than have to
have a bunch of windows open each running a different tool, collectl
does it all and reports its output in several different but consistent
formats. One format, which I call 'plot format', is actually space
separated data that can be consumed by gnuplot or even read by excel.
Another format can be consumed by rrdtool if you're of that persuasion.
It can even run at subsecond monitoring intervals and if you think
that's not important consider that most linux kernels do NOT correctly
report network statistics if monitored once a second but rather you need
to monitor at 0.9765 seconds (see the website for a detailed explanation).
The best thing is it uses less than 0.1% cpu when collecting samples
every 10 seconds as well as process and slab statistics every 60 seconds.
Anyhow, if you're a linux person take it for a spin.
-mark