I've been looking at hwpmc a little more lately (argh). One of the things which annoys me is that although hwpmc can capture events from existing processes pmcstat doesn't generate summary output for them. If this is still the case (I've not checked -current over my exam period!), the required changes don't look -that- hard. pmc receives events from hwpmc about the exec and mappings for the binary and libraries; I wonder if pmcstat could be taught to "fake" exec/map-in/map-out entries for stats from processes it hasn't seen yet. It'd also be nice if it recorded some kind of timestamp from the file(s) in question so it could notice if the underlying libraries/binaries have changed between gathering and reporting. I sort of need this with Squid so I can profile things after a "burnin period". I can currently fake it by starting squid + hwpmc and removing the first n kernel/user events from the pmcstat data file. This is slightly silly. :) Has anyone else looked into this? adrianReceived on Wed Jun 25 2008 - 19:59:17 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:32 UTC