summaryrefslogtreecommitdiff
path: root/Documentation/sched-stats.txt
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 15:20:36 -0700
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 15:20:36 -0700
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /Documentation/sched-stats.txt
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
Diffstat (limited to 'Documentation/sched-stats.txt')
-rw-r--r--Documentation/sched-stats.txt153
1 files changed, 153 insertions, 0 deletions
diff --git a/Documentation/sched-stats.txt b/Documentation/sched-stats.txt
new file mode 100644
index 000000000000..6f72021aae51
--- /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