154 lines
7.0 KiB
Plaintext
154 lines
7.0 KiB
Plaintext
Version 10 of schedstats includes support for sched_domains, which
|
|
hit the mainline kernel in 2.6.7. Some counters make more sense to be
|
|
per-runqueue; other to be per-domain. Note that domains (and their associated
|
|
information) will only be pertinent and available on machines utilizing
|
|
CONFIG_SMP.
|
|
|
|
In version 10 of schedstat, there is at least one level of domain
|
|
statistics for each cpu listed, and there may well be more than one
|
|
domain. Domains have no particular names in this implementation, but
|
|
the highest numbered one typically arbitrates balancing across all the
|
|
cpus on the machine, while domain0 is the most tightly focused domain,
|
|
sometimes balancing only between pairs of cpus. At this time, there
|
|
are no architectures which need more than three domain levels. The first
|
|
field in the domain stats is a bit map indicating which cpus are affected
|
|
by that domain.
|
|
|
|
These fields are counters, and only increment. Programs which make use
|
|
of these will need to start with a baseline observation and then calculate
|
|
the change in the counters at each subsequent observation. A perl script
|
|
which does this for many of the fields is available at
|
|
|
|
http://eaglet.rain.com/rick/linux/schedstat/
|
|
|
|
Note that any such script will necessarily be version-specific, as the main
|
|
reason to change versions is changes in the output format. For those wishing
|
|
to write their own scripts, the fields are described here.
|
|
|
|
CPU statistics
|
|
--------------
|
|
cpu<N> 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
|
|
|
|
NOTE: In the sched_yield() statistics, the active queue is considered empty
|
|
if it has only one process in it, since obviously the process calling
|
|
sched_yield() is that process.
|
|
|
|
First four fields are sched_yield() statistics:
|
|
1) # of times both the active and the expired queue were empty
|
|
2) # of times just the active queue was empty
|
|
3) # of times just the expired queue was empty
|
|
4) # of times sched_yield() was called
|
|
|
|
Next four are schedule() statistics:
|
|
5) # of times the active queue had at least one other process on it
|
|
6) # of times we switched to the expired queue and reused it
|
|
7) # of times schedule() was called
|
|
8) # of times schedule() left the processor idle
|
|
|
|
Next four are active_load_balance() statistics:
|
|
9) # of times active_load_balance() was called
|
|
10) # of times active_load_balance() caused this cpu to gain a task
|
|
11) # of times active_load_balance() caused this cpu to lose a task
|
|
12) # of times active_load_balance() tried to move a task and failed
|
|
|
|
Next three are try_to_wake_up() statistics:
|
|
13) # of times try_to_wake_up() was called
|
|
14) # of times try_to_wake_up() successfully moved the awakening task
|
|
15) # of times try_to_wake_up() attempted to move the awakening task
|
|
|
|
Next two are wake_up_new_task() statistics:
|
|
16) # of times wake_up_new_task() was called
|
|
17) # of times wake_up_new_task() successfully moved the new task
|
|
|
|
Next one is a sched_migrate_task() statistic:
|
|
18) # of times sched_migrate_task() was called
|
|
|
|
Next one is a sched_balance_exec() statistic:
|
|
19) # of times sched_balance_exec() was called
|
|
|
|
Next three are statistics describing scheduling latency:
|
|
20) sum of all time spent running by tasks on this processor (in ms)
|
|
21) sum of all time spent waiting to run by tasks on this processor (in ms)
|
|
22) # of tasks (not necessarily unique) given to the processor
|
|
|
|
The last six are statistics dealing with pull_task():
|
|
23) # of times pull_task() moved a task to this cpu when newly idle
|
|
24) # of times pull_task() stole a task from this cpu when another cpu
|
|
was newly idle
|
|
25) # of times pull_task() moved a task to this cpu when idle
|
|
26) # of times pull_task() stole a task from this cpu when another cpu
|
|
was idle
|
|
27) # of times pull_task() moved a task to this cpu when busy
|
|
28) # of times pull_task() stole a task from this cpu when another cpu
|
|
was busy
|
|
|
|
|
|
Domain statistics
|
|
-----------------
|
|
One of these is produced per domain for each cpu described. (Note that if
|
|
CONFIG_SMP is not defined, *no* domains are utilized and these lines
|
|
will not appear in the output.)
|
|
|
|
domain<N> 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
|
|
|
|
The first field is a bit mask indicating what cpus this domain operates over.
|
|
|
|
The next fifteen are a variety of load_balance() statistics:
|
|
|
|
1) # of times in this domain load_balance() was called when the cpu
|
|
was idle
|
|
2) # of times in this domain load_balance() was called when the cpu
|
|
was busy
|
|
3) # of times in this domain load_balance() was called when the cpu
|
|
was just becoming idle
|
|
4) # of times in this domain load_balance() tried to move one or more
|
|
tasks and failed, when the cpu was idle
|
|
5) # of times in this domain load_balance() tried to move one or more
|
|
tasks and failed, when the cpu was busy
|
|
6) # of times in this domain load_balance() tried to move one or more
|
|
tasks and failed, when the cpu was just becoming idle
|
|
7) sum of imbalances discovered (if any) with each call to
|
|
load_balance() in this domain when the cpu was idle
|
|
8) sum of imbalances discovered (if any) with each call to
|
|
load_balance() in this domain when the cpu was busy
|
|
9) sum of imbalances discovered (if any) with each call to
|
|
load_balance() in this domain when the cpu was just becoming idle
|
|
10) # of times in this domain load_balance() was called but did not find
|
|
a busier queue while the cpu was idle
|
|
11) # of times in this domain load_balance() was called but did not find
|
|
a busier queue while the cpu was busy
|
|
12) # of times in this domain load_balance() was called but did not find
|
|
a busier queue while the cpu was just becoming idle
|
|
13) # of times in this domain a busier queue was found while the cpu was
|
|
idle but no busier group was found
|
|
14) # of times in this domain a busier queue was found while the cpu was
|
|
busy but no busier group was found
|
|
15) # of times in this domain a busier queue was found while the cpu was
|
|
just becoming idle but no busier group was found
|
|
|
|
Next two are sched_balance_exec() statistics:
|
|
17) # of times in this domain sched_balance_exec() successfully pushed
|
|
a task to a new cpu
|
|
18) # of times in this domain sched_balance_exec() tried but failed to
|
|
push a task to a new cpu
|
|
|
|
Next two are try_to_wake_up() statistics:
|
|
19) # of times in this domain try_to_wake_up() tried to move a task based
|
|
on affinity and cache warmth
|
|
20) # of times in this domain try_to_wake_up() tried to move a task based
|
|
on load balancing
|
|
|
|
|
|
/proc/<pid>/schedstat
|
|
----------------
|
|
schedstats also adds a new /proc/<pid/schedstat file to include some of
|
|
the same information on a per-process level. There are three fields in
|
|
this file correlating to fields 20, 21, and 22 in the CPU fields, but
|
|
they only apply for that process.
|
|
|
|
A program could be easily written to make use of these extra fields to
|
|
report on how well a particular process or set of processes is faring
|
|
under the scheduler's policies. A simple version of such a program is
|
|
available at
|
|
http://eaglet.rain.com/rick/linux/schedstat/v10/latency.c
|