1================
2Delay accounting
3================
4
5Tasks encounter delays in execution when they wait
6for some kernel resource to become available e.g. a
7runnable task may wait for a free CPU to run on.
8
9The per-task delay accounting functionality measures
10the delays experienced by a task while
11
12a) waiting for a CPU (while being runnable)
13b) completion of synchronous block I/O initiated by the task
14c) swapping in pages
15d) memory reclaim
16
17and makes these statistics available to userspace through
18the taskstats interface.
19
20Such delays provide feedback for setting a task's cpu priority,
21io priority and rss limit values appropriately. Long delays for
22important tasks could be a trigger for raising its corresponding priority.
23
24The functionality, through its use of the taskstats interface, also provides
25delay statistics aggregated for all tasks (or threads) belonging to a
26thread group (corresponding to a traditional Unix process). This is a commonly
27needed aggregation that is more efficiently done by the kernel.
28
29Userspace utilities, particularly resource management applications, can also
30aggregate delay statistics into arbitrary groups. To enable this, delay
31statistics of a task are available both during its lifetime as well as on its
32exit, ensuring continuous and complete monitoring can be done.
33
34
35Interface
36---------
37
38Delay accounting uses the taskstats interface which is described
39in detail in a separate document in this directory. Taskstats returns a
40generic data structure to userspace corresponding to per-pid and per-tgid
41statistics. The delay accounting functionality populates specific fields of
42this structure. See
43
44     include/linux/taskstats.h
45
46for a description of the fields pertaining to delay accounting.
47It will generally be in the form of counters returning the cumulative
48delay seen for cpu, sync block I/O, swapin, memory reclaim etc.
49
50Taking the difference of two successive readings of a given
51counter (say cpu_delay_total) for a task will give the delay
52experienced by the task waiting for the corresponding resource
53in that interval.
54
55When a task exits, records containing the per-task statistics
56are sent to userspace without requiring a command. If it is the last exiting
57task of a thread group, the per-tgid statistics are also sent. More details
58are given in the taskstats interface description.
59
60The getdelays.c userspace utility in tools/accounting directory allows simple
61commands to be run and the corresponding delay statistics to be displayed. It
62also serves as an example of using the taskstats interface.
63
64Usage
65-----
66
67Compile the kernel with::
68
69	CONFIG_TASK_DELAY_ACCT=y
70	CONFIG_TASKSTATS=y
71
72Delay accounting is enabled by default at boot up.
73To disable, add::
74
75   nodelayacct
76
77to the kernel boot options. The rest of the instructions
78below assume this has not been done.
79
80After the system has booted up, use a utility
81similar to  getdelays.c to access the delays
82seen by a given task or a task group (tgid).
83The utility also allows a given command to be
84executed and the corresponding delays to be
85seen.
86
87General format of the getdelays command::
88
89	getdelays [-t tgid] [-p pid] [-c cmd...]
90
91
92Get delays, since system boot, for pid 10::
93
94	# ./getdelays -p 10
95	(output similar to next case)
96
97Get sum of delays, since system boot, for all pids with tgid 5::
98
99	# ./getdelays -t 5
100
101
102	CPU	count	real total	virtual total	delay total
103		7876	92005750	100000000	24001500
104	IO	count	delay total
105		0	0
106	SWAP	count	delay total
107		0	0
108	RECLAIM	count	delay total
109		0	0
110
111Get delays seen in executing a given simple command::
112
113  # ./getdelays -c ls /
114
115  bin   data1  data3  data5  dev  home  media  opt   root  srv        sys  usr
116  boot  data2  data4  data6  etc  lib   mnt    proc  sbin  subdomain  tmp  var
117
118
119  CPU	count	real total	virtual total	delay total
120	6	4000250		4000000		0
121  IO	count	delay total
122	0	0
123  SWAP	count	delay total
124	0	0
125  RECLAIM	count	delay total
126	0	0
127