diff options
Diffstat (limited to 'Documentation/sched-stats.txt')
| -rw-r--r-- | Documentation/sched-stats.txt | 153 |
1 files changed, 153 insertions, 0 deletions
diff --git a/Documentation/sched-stats.txt b/Documentation/sched-stats.txt new file mode 100644 index 0000000..6f72021 --- /dev/null +++ b/Documentation/sched-stats.txt @@ -0,0 +1,153 @@ +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 |
