4 Tasks encounter delays in execution when they wait
5 for some kernel resource to become available e.g. a
6 runnable task may wait for a free CPU to run on.
8 The per-task delay accounting functionality measures
9 the delays experienced by a task while
11 a) waiting for a CPU (while being runnable)
12 b) completion of synchronous block I/O initiated by the task
16 and makes these statistics available to userspace through
17 the taskstats interface.
19 Such delays provide feedback for setting a task's cpu priority,
20 io priority and rss limit values appropriately. Long delays for
21 important tasks could be a trigger for raising its corresponding priority.
23 The functionality, through its use of the taskstats interface, also provides
24 delay statistics aggregated for all tasks (or threads) belonging to a
25 thread group (corresponding to a traditional Unix process). This is a commonly
26 needed aggregation that is more efficiently done by the kernel.
28 Userspace utilities, particularly resource management applications, can also
29 aggregate delay statistics into arbitrary groups. To enable this, delay
30 statistics of a task are available both during its lifetime as well as on its
31 exit, ensuring continuous and complete monitoring can be done.
37 Delay accounting uses the taskstats interface which is described
38 in detail in a separate document in this directory. Taskstats returns a
39 generic data structure to userspace corresponding to per-pid and per-tgid
40 statistics. The delay accounting functionality populates specific fields of
42 include/linux/taskstats.h
43 for a description of the fields pertaining to delay accounting.
44 It will generally be in the form of counters returning the cumulative
45 delay seen for cpu, sync block I/O, swapin, memory reclaim etc.
47 Taking the difference of two successive readings of a given
48 counter (say cpu_delay_total) for a task will give the delay
49 experienced by the task waiting for the corresponding resource
52 When a task exits, records containing the per-task statistics
53 are sent to userspace without requiring a command. If it is the last exiting
54 task of a thread group, the per-tgid statistics are also sent. More details
55 are given in the taskstats interface description.
57 The getdelays.c userspace utility in this directory allows simple commands to
58 be run and the corresponding delay statistics to be displayed. It also serves
59 as an example of using the taskstats interface.
64 Compile the kernel with
65 CONFIG_TASK_DELAY_ACCT=y
68 Delay accounting is enabled by default at boot up.
71 to the kernel boot options. The rest of the instructions
72 below assume this has not been done.
74 After the system has booted up, use a utility
75 similar to getdelays.c to access the delays
76 seen by a given task or a task group (tgid).
77 The utility also allows a given command to be
78 executed and the corresponding delays to be
81 General format of the getdelays command
83 getdelays [-t tgid] [-p pid] [-c cmd...]
86 Get delays, since system boot, for pid 10
88 (output similar to next case)
90 Get sum of delays, since system boot, for all pids with tgid 5
94 CPU count real total virtual total delay total
95 7876 92005750 100000000 24001500
98 SWAP count delay total
100 RECLAIM count delay total
103 Get delays seen in executing a given simple command
104 # ./getdelays -c ls /
106 bin data1 data3 data5 dev home media opt root srv sys usr
107 boot data2 data4 data6 etc lib mnt proc sbin subdomain tmp var
110 CPU count real total virtual total delay total
114 SWAP count delay total
116 RECLAIM count delay total