McGuire_Ed@GRINNELL.MAILNET (07/08/86)
***** Is there a wizard in the house? ***** I have written a DCL procedure which predicts the process capacity of page/swap files by projecting from their current usage and the number of processes on the system. (I assume that the page file space should be kept 1/2 free and swap file space 1/4 free.) I subtract the usage of system processes from page/swap file capacity, for a more accurate prediction, projected solely from the real workload of the systems. And now my question: Is it possible to determine the page file usage for a process, short of scanning the page tables for swapped pages? Right now, I make the excessively liberal estimate that system processes don't map any page-file pages. Of course, I could also err on the side of conservatism (and I don't want to do THAT :-)), and estimate that they are using working-set-size minus process-page-count pages, but that makes even more inappropriate assumptions. I am computing swap file usage for the processes by raising working- set-size to the next multiple of MPW_WRTCLUSTER. Any suggestions for page/swapfile management would be MOST appreciated: we're now running a 4-node VAXcluster with an aggregate of 80MB--these babies are getting LARGE! Not to mention the dump files... [My procedure reports the number of processes on the system broken down by MODE; MAXPROCESSCNT and Login Interactive Limit; free and total page/swapfile pages; projected capacity. If anyone wants a copy, send me a note. I'll distribute it individually, or post it if I get more than 10 requests.]
kjs@TUFTS.CSNET.UUCP (07/08/86)
I'd like to take a look at it. Would be best to send it to my bitnet address which is kjs@tufts, but you could also send it to me on this machine, kjs%tufts@csnet-relay. Thanks so much. Is there some documentation? I guess that I'll find out. Kevin Sullivan Unix Systems Programmer Tufts University.