by paul
In the server monitoring we can monitor the load and the cpu usage. But when the the load or CPU usage are high, it is not always clear what the culprit is in case of high disk I/O. To be able to quickly see that the disk I/O is the problem I would like to see the different categories of CPU load Linux has (user, system, nice, iowait etc) stacked in the CPU graph. This way when I/O is the problem, the iowait percentage is high.
For even better visibility of disk I/O load problems it would be great if the I/O usage in percentage per disk was logged as well. E.g. the %util per device that 'iostat -x 2 2' returns. Now only the raw mb/s of disk I/O is logged, but on non SSD disks this value alone is not a good indicator if a disk is overloaded.
I've currently made a plugin which returns the user, system, nice, iowait and steal percentages which the top command returns. But I hope these values can easilty be added to the standard CPU monitoring which for Linux is in scripts/cpu.sh, since that script uses the top command to get the idle pertentage.
Since the number of disks are different per server, a plugin for %util per disk is not an option at the moment, since one plugin should always return the same attributes.
I've included the cpu stats plugin. Feel free to use it if it is usufull for you. Just make sure iostat (usually in the sysstat package) is installed.
Attachments
cpustats.sh.txt
Size: 0.38 KB
Like (6)
Reply
Replies (0)