About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog

Documentation / scheduler / sched-stats.txt


Based on kernel version 4.16.1. Page generated on 2018-04-09 11:53 EST.

1	Version 15 of schedstats dropped counters for some sched_yield:
2	yld_exp_empty, yld_act_empty and yld_both_empty. Otherwise, it is
3	identical to version 14.
4	
5	Version 14 of schedstats includes support for sched_domains, which hit the
6	mainline kernel in 2.6.20 although it is identical to the stats from version
7	12 which was in the kernel from 2.6.13-2.6.19 (version 13 never saw a kernel
8	release).  Some counters make more sense to be per-runqueue; other to be
9	per-domain.  Note that domains (and their associated information) will only
10	be pertinent and available on machines utilizing CONFIG_SMP.
11	
12	In version 14 of schedstat, there is at least one level of domain
13	statistics for each cpu listed, and there may well be more than one
14	domain.  Domains have no particular names in this implementation, but
15	the highest numbered one typically arbitrates balancing across all the
16	cpus on the machine, while domain0 is the most tightly focused domain,
17	sometimes balancing only between pairs of cpus.  At this time, there
18	are no architectures which need more than three domain levels. The first
19	field in the domain stats is a bit map indicating which cpus are affected
20	by that domain.
21	
22	These fields are counters, and only increment.  Programs which make use
23	of these will need to start with a baseline observation and then calculate
24	the change in the counters at each subsequent observation.  A perl script
25	which does this for many of the fields is available at
26	
27	    http://eaglet.rain.com/rick/linux/schedstat/
28	
29	Note that any such script will necessarily be version-specific, as the main
30	reason to change versions is changes in the output format.  For those wishing
31	to write their own scripts, the fields are described here.
32	
33	CPU statistics
34	--------------
35	cpu<N> 1 2 3 4 5 6 7 8 9
36	
37	First field is a sched_yield() statistic:
38	     1) # of times sched_yield() was called
39	
40	Next three are schedule() statistics:
41	     2) This field is a legacy array expiration count field used in the O(1)
42		scheduler. We kept it for ABI compatibility, but it is always set to zero.
43	     3) # of times schedule() was called
44	     4) # of times schedule() left the processor idle
45	
46	Next two are try_to_wake_up() statistics:
47	     5) # of times try_to_wake_up() was called
48	     6) # of times try_to_wake_up() was called to wake up the local cpu
49	
50	Next three are statistics describing scheduling latency:
51	     7) sum of all time spent running by tasks on this processor (in jiffies)
52	     8) sum of all time spent waiting to run by tasks on this processor (in
53	        jiffies)
54	     9) # of timeslices run on this cpu
55	
56	
57	Domain statistics
58	-----------------
59	One of these is produced per domain for each cpu described. (Note that if
60	CONFIG_SMP is not defined, *no* domains are utilized and these lines
61	will not appear in the output.)
62	
63	domain<N> <cpumask> 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
64	
65	The first field is a bit mask indicating what cpus this domain operates over.
66	
67	The next 24 are a variety of load_balance() statistics in grouped into types
68	of idleness (idle, busy, and newly idle):
69	
70	     1) # of times in this domain load_balance() was called when the
71	        cpu was idle
72	     2) # of times in this domain load_balance() checked but found
73	        the load did not require balancing when the cpu was idle
74	     3) # of times in this domain load_balance() tried to move one or
75	        more tasks and failed, when the cpu was idle
76	     4) sum of imbalances discovered (if any) with each call to
77	        load_balance() in this domain when the cpu was idle
78	     5) # of times in this domain pull_task() was called when the cpu
79	        was idle
80	     6) # of times in this domain pull_task() was called even though
81	        the target task was cache-hot when idle
82	     7) # of times in this domain load_balance() was called but did
83	        not find a busier queue while the cpu was idle
84	     8) # of times in this domain a busier queue was found while the
85	        cpu was idle but no busier group was found
86	
87	     9) # of times in this domain load_balance() was called when the
88	        cpu was busy
89	    10) # of times in this domain load_balance() checked but found the
90	        load did not require balancing when busy
91	    11) # of times in this domain load_balance() tried to move one or
92	        more tasks and failed, when the cpu was busy
93	    12) sum of imbalances discovered (if any) with each call to
94	        load_balance() in this domain when the cpu was busy
95	    13) # of times in this domain pull_task() was called when busy
96	    14) # of times in this domain pull_task() was called even though the
97	        target task was cache-hot when busy
98	    15) # of times in this domain load_balance() was called but did not
99	        find a busier queue while the cpu was busy
100	    16) # of times in this domain a busier queue was found while the cpu
101	        was busy but no busier group was found
102	
103	    17) # of times in this domain load_balance() was called when the
104	        cpu was just becoming idle
105	    18) # of times in this domain load_balance() checked but found the
106	        load did not require balancing when the cpu was just becoming idle
107	    19) # of times in this domain load_balance() tried to move one or more
108	        tasks and failed, when the cpu was just becoming idle
109	    20) sum of imbalances discovered (if any) with each call to
110	        load_balance() in this domain when the cpu was just becoming idle
111	    21) # of times in this domain pull_task() was called when newly idle
112	    22) # of times in this domain pull_task() was called even though the
113	        target task was cache-hot when just becoming idle
114	    23) # of times in this domain load_balance() was called but did not
115	        find a busier queue while the cpu was just becoming idle
116	    24) # of times in this domain a busier queue was found while the cpu
117	        was just becoming idle but no busier group was found
118	
119	   Next three are active_load_balance() statistics:
120	    25) # of times active_load_balance() was called
121	    26) # of times active_load_balance() tried to move a task and failed
122	    27) # of times active_load_balance() successfully moved a task
123	
124	   Next three are sched_balance_exec() statistics:
125	    28) sbe_cnt is not used
126	    29) sbe_balanced is not used
127	    30) sbe_pushed is not used
128	
129	   Next three are sched_balance_fork() statistics:
130	    31) sbf_cnt is not used
131	    32) sbf_balanced is not used
132	    33) sbf_pushed is not used
133	
134	   Next three are try_to_wake_up() statistics:
135	    34) # of times in this domain try_to_wake_up() awoke a task that
136	        last ran on a different cpu in this domain
137	    35) # of times in this domain try_to_wake_up() moved a task to the
138	        waking cpu because it was cache-cold on its own cpu anyway
139	    36) # of times in this domain try_to_wake_up() started passive balancing
140	
141	/proc/<pid>/schedstat
142	----------------
143	schedstats also adds a new /proc/<pid>/schedstat file to include some of
144	the same information on a per-process level.  There are three fields in
145	this file correlating for that process to:
146	     1) time spent on the cpu
147	     2) time spent waiting on a runqueue
148	     3) # of timeslices run on this cpu
149	
150	A program could be easily written to make use of these extra fields to
151	report on how well a particular process or set of processes is faring
152	under the scheduler's policies.  A simple version of such a program is
153	available at
154	    http://eaglet.rain.com/rick/linux/schedstat/v12/latency.c
Hide Line Numbers


About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog