freebsd-nq/cddl/contrib/dtracetoolkit/Examples/anonpgpid_example.txt
2012-05-12 21:25:48 +00:00

74 lines
2.4 KiB
Plaintext

The following is a demonstration of the anonpgpid.d script,
Here we run it on a system that is implementing memory caps using the
resource capping daemon, "rcapd",
# anonpgpid.d
Tracing... Hit Ctrl-C to end.
^C
PID CMD D BYTES
6215 bash R 8192
6215 bash W 126976
5809 rcapd R 245760
6222 memleak.pl R 974848
6222 memleak.pl W 3055616
The "memleak.pl" process consumes memory, and we can see above that it has
encountered both reads and writes to the physical swap device - it is being
paged out. A bash shell was also effected (which was in the same project that
rcapd was monitoring).
The following is an ordinary system that is very low on memory,
# anonpgpid.d
Tracing... Hit Ctrl-C to end.
^C
PID CMD D BYTES
18885 sendmail R 4096
18600 automountd R 4096
1 init R 4096
2456 inetd R 8192
18546 nscd R 8192
2400 bash R 12288
217 utmpd R 28672
221 ttymon R 32768
210 sac R 36864
18777 snmpd R 49152
18440 init R 49152
89 nscd R 61440
318 syslogd R 73728
487 snmpd R 81920
2453 inetd R 102400
165 in.routed R 131072
294 automountd R 135168
215 inetd R 135168
187 rpcbind R 204800
86 kcfd R 290816
7 svc.startd R 1015808
9 svc.configd R 1478656
2 pageout W 23453696
The "pageout" process is responsible for writing all the anonymous memory
pages to the physical swap device, and we can see from the above that it
has written 23 Mb. When processes access anonymous memory that has been
swapped out, a major fault occurs and the memory is paged back in; in this
case we can trace the process that was effected, and from the above we can
see that several processes have been effected by the memory pressure.
The most is "svc.configd", which needed to page back in 1.4 Mb of anonymous
memory.
Sometimes anonpgpid.d doesn't help too much. Here we only have pageouts
to the physical swap device and no pageins,
# anonpgpid.d
^C
PID CMD D BYTES
2 pageout W 61083648
Only pageout is identified.