1cpudist(8) System Manager's Manual cpudist(8)
2
3
4
6 cpudist - On- and off-CPU task time as a histogram.
7
9 cpudist [-h] [-O] [-T] [-m] [-P] [-L] [-p PID] [-I] [-e] [interval]
10 [count]
11
13 This measures the time a task spends on the CPU before being desched‐
14 uled, and shows the times as a histogram. Tasks that spend a very short
15 time on the CPU can be indicative of excessive context-switches and
16 poor workload distribution, and possibly point to a shared source of
17 contention that keeps tasks switching in and out as it becomes avail‐
18 able (such as a mutex).
19
20 Similarly, the tool can also measure the time a task spends off-CPU be‐
21 fore it is scheduled again. This can be helpful in identifying long
22 blocking and I/O operations, or alternatively very short descheduling
23 times due to short-lived locks or timers.
24
25 By default CPU idle time are excluded by simply excluding PID 0.
26
27 This tool uses in-kernel eBPF maps for storing timestamps and the his‐
28 togram, for efficiency. Despite this, the overhead of this tool may be‐
29 come significant for some workloads: see the OVERHEAD section.
30
31 Since this uses BPF, only the root user can use this tool.
32
34 CONFIG_BPF and bcc.
35
37 -h Print usage message.
38
39 -O Measure off-CPU time instead of on-CPU time.
40
41 -T Include timestamps on output.
42
43 -m Output histogram in milliseconds.
44
45 -P Print a histogram for each PID (tgid from the kernel's perspec‐
46 tive).
47
48 -L Print a histogram for each TID (pid from the kernel's perspec‐
49 tive).
50
51 -p PID Only show this PID (filtered in kernel for efficiency).
52
53 -I Include CPU idle time (by default these are excluded).
54
55 -e Show extension summary (average/total/count).
56
57 interval
58 Output interval, in seconds.
59
60 count Number of outputs.
61
63 Summarize task on-CPU time as a histogram:
64 # cpudist
65
66 Summarize task off-CPU time as a histogram:
67 # cpudist -O
68
69 Print 1 second summaries, 10 times:
70 # cpudist 1 10
71
72 Print 1 second summaries, using milliseconds as units for the his‐
73 togram, and include timestamps on output:
74 # cpudist -mT 1
75
76 Trace PID 185 only, 1 second summaries:
77 # cpudist -p 185 1
78
79 Include CPU idle time:
80 # cpudist -I
81
82 Also show extension summary:
83 # cpudist -e
84
86 usecs Microsecond range
87
88 msecs Millisecond range
89
90 count How many times a task event fell into this range
91
92 distribution
93 An ASCII bar chart to visualize the distribution (count column)
94
96 This traces scheduler tracepoints, which can become very frequent.
97 While eBPF has very low overhead, and this tool uses in-kernel maps for
98 efficiency, the frequency of scheduler events for some workloads may be
99 high enough that the overhead of this tool becomes significant. Measure
100 in a lab environment to quantify the overhead before use.
101
103 This is from bcc.
104
105 https://github.com/iovisor/bcc
106
107 Also look in the bcc distribution for a companion _example.txt file
108 containing example usage, output, and commentary for this tool.
109
111 Linux
112
114 Unstable - in development.
115
117 Sasha Goldshtein, Rocky Xing
118
120 pidstat(1), runqlat(8)
121
122
123
124USER COMMANDS 2016-06-28 cpudist(8)