123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153 |
- 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
|