Greetings, all. I've been running my copy stgpools on remote virtual volumes for a skewtch more than a year. I'm not utterly displeased with my reclamation behavior, but it's not what I'd call efficient. Though, it may be optimal. In any case, I wanted to compare notes.
I've got about a dozen servers at my primary site. To a good approximation, they each have one, non-collocated copy stgpool. Of course, just about any time I reclaim some offsite volumes I end up mounting almost all of the primary (collocated) tapes. So I prefer to reclaim with a rather low target, less frequently, rather than a higher target, more frequently. I aim for 50%. Also, being somewhat tape-drive constrained, I can really only afford to have one tape drive occupied doing reclamation work at a time. So, my procedure is roughly: 1) Look at a series of histograms of pct_reclaimed of my offsite stgpools. (I've tossed in some links to examples of the histograms I use.) 2) Pick "the next target" for reclamation. TRGTHRESH=50 Here's an example of one of my servers' copy pool when I decided it was next up. http://nersp.osg.ufl.edu/~asr/tsm-examples/glmail02-2006-10-29-13-28.png 3) RECLAIM STGPOOL [target] thresh=TRGTHRESH. Wait. Could take a week. This example took three days. 4) At least daily, check and see if the reclamation process is still running. 5) If it's stopped, regenerate the histogram. That will permit the easy distinguishing of "I'm finished" from "I got derailed by a restore or some error". Here's an example of "finished." http://nersp.osg.ufl.edu/~asr/tsm-examples/glmail02-2006-10-31-21-35.png 6) If finished, GOTO 1. 7) Since target isn't finished, see if I can locate a pctreclaim higher than 50 which I can set, which will finish the job. Often, there is a big gap: as data gets recopied, volumes get higher pctreclaim values, so the distribution climbs towards the top of the scale. That means I can often set TRGTHRESH=70, 80, 90, etc. for successive runs. Here's an example of 'derailed', where I'd set a new threshold of 59%. http://nersp.osg.ufl.edu/~asr/tsm-examples/glmail02-2006-10-31-05-46.png and here's one where I'd set a new threshold of 85% http://nersp.osg.ufl.edu/~asr/tsm-examples/glmail02-2006-10-31-16-57.png 8) GOTO 3. So how do you-all do this? What am I missing? - Allen S. Rout
