On Sat, 29 Jul 2006, Charles Jones -X \(charljon - Cisco Learning Institute at Cisco\) wrote:
Here is an example of a tomcat garbage collection log:
$ tail -5 gc.log 15424.931: [Full GC 48511K->48216K(1022400K), 0.4856710 secs] 15485.431: [Full GC 48540K->48217K(1022400K), 0.4745360 secs] 15545.927: [Full GC 48513K->48218K(1022400K), 0.4845780 secs] 15606.449: [Full GC 48513K->48215K(1022400K), 0.5136750 secs] 15667.010: [Full GC 48511K->48216K(1022400K), 0.5008330 secs]
A word of warning, GC metrics are *not* good time-series data. gc's happen 'when they need to' and not on fixed intervals. rrd interpolates data to fixed time points.
So if your GCs start happening every 10 seconds but maintain 0.5 second durations, the rrdtool graphs won't show anything peculiar.
The number of GCs that occur every 5 minutes would be good time-series data, but the logic to parse that would be more complicated.
I believe there are tools designed specifically for GC analysis that deal with all of these issues.
-- Scott Walters -PacketPusher