When you generate plot output with -P (uppercase), you get space-delimited output but you can also change the separator character with –sep if you prefer something else. And also be sure to look at collectl-utils which comes with a web-based plotting utility called colplot. Colmux allows you to multiplex across a number of nodes (tested at over 1K) in a cluster and sort the output by ‘top’ values.

One thing I added a years ago it the ability to write plugins so if you want to monitor an nvidia gpu, for example, you just add –import nvidia.

One other important concept is that of summary vs detail data. For exampl -sd will show total disk i/o while -sD with show i/o for individual disks. Same thing about -sc and -sC for cpus, -sn for network and -sx for interconnects like Infiniband.

Finally, and something I tell to everyone, if you’re really serious about monitoring and also prefer sar or some other tool, do yourself a favor and rather than use the default monitoring interval of 10 minutes samples that sar uses, use 10 second samples like collectl. There is very little overhead – sar is probably more efficient than collectl and collect is very efficient – and 10 minute sampling results in relatively useless numbers for spotting short lived events like spiked in the the data.

-mark