cpuacct.rst (2008B)
1========================= 2CPU Accounting Controller 3========================= 4 5The CPU accounting controller is used to group tasks using cgroups and 6account the CPU usage of these groups of tasks. 7 8The CPU accounting controller supports multi-hierarchy groups. An accounting 9group accumulates the CPU usage of all of its child groups and the tasks 10directly present in its group. 11 12Accounting groups can be created by first mounting the cgroup filesystem:: 13 14 # mount -t cgroup -ocpuacct none /sys/fs/cgroup 15 16With the above step, the initial or the parent accounting group becomes 17visible at /sys/fs/cgroup. At bootup, this group includes all the tasks in 18the system. /sys/fs/cgroup/tasks lists the tasks in this cgroup. 19/sys/fs/cgroup/cpuacct.usage gives the CPU time (in nanoseconds) obtained 20by this group which is essentially the CPU time obtained by all the tasks 21in the system. 22 23New accounting groups can be created under the parent group /sys/fs/cgroup:: 24 25 # cd /sys/fs/cgroup 26 # mkdir g1 27 # echo $$ > g1/tasks 28 29The above steps create a new group g1 and move the current shell 30process (bash) into it. CPU time consumed by this bash and its children 31can be obtained from g1/cpuacct.usage and the same is accumulated in 32/sys/fs/cgroup/cpuacct.usage also. 33 34cpuacct.stat file lists a few statistics which further divide the 35CPU time obtained by the cgroup into user and system times. Currently 36the following statistics are supported: 37 38user: Time spent by tasks of the cgroup in user mode. 39system: Time spent by tasks of the cgroup in kernel mode. 40 41user and system are in USER_HZ unit. 42 43cpuacct controller uses percpu_counter interface to collect user and 44system times. This has two side effects: 45 46- It is theoretically possible to see wrong values for user and system times. 47 This is because percpu_counter_read() on 32bit systems isn't safe 48 against concurrent writes. 49- It is possible to see slightly outdated values for user and system times 50 due to the batch processing nature of percpu_counter.