Trying to pin a specific process to a disk can be a problem as collectl doesn't track that sort of thing, however sometimes you can infer a lot by correlating various things. So assuming you have a bunch of raw collectl data. It's pretty easy to see how busy a particular disk is with 'collectl -p filename -sD --dskfilt xxx -oT' noting the oT will stick a timestamp on each line. If you can identify some spikes you think might be attributed to the problem process, this is a good thing. Be sure to note the timeframe.
Now if you play back the same data and report process I/O usage with 'collectl -p filename -sZ --from hh:mm-hh:mm --procanalyze', this will report a summary of the processes that were running between those two times in space-separated format, which you can then load into a spreadsheet and from there you can identify which processes were doing a lot of disk i/o. Keep be mind this will not tell you who was talking to what disk, only that they were doing I/O. This may give you the answer you're looking for and maybe it won't.
The other thing is if the spikes are relatively short, the 1 minute default process monitoring interval may be too long to see anything and running collectl with the a 10 second interval like the disk processing interval, may be more helpful.
No promises this will work for you but it's he best I can come up with at the moment.
good luck
-mark
