Hello.... :) > This is a structure that contains information about scheduling, for a > single process, by the look of it. correct, this is for single process / thread. IIRC, these fields can be accessed on /proc/<pid> schedstat. Make sure you have compiled your kernel with CONFIG_SCHEDSTATS=y check Documentation/sched-stats.txt for further informations :) > cpu_time gives a measure of the amount of time the process has spent > running on the given cpu (Presumably there's one of these structures > for each cpu, for each process). That's probably given in jiffies or > something, I dunnno =) IIRC this cpu_time field is cumulative time for a process being executed throughout all the available processor.... I also saw this sched_info structure is also assigned on runqueue definition....so each CPU is also tracking how long it has executed all the runnable processes since booting.... > run_delay gives the amount of time spent waiting for access to the > given cpu, in order to run (in a blocked/sleeping state, as it were) > - That means, run_delay + cpu_time gives the total real time that > process has existed, presumably. IMO, this is wrong. run_delay is the interval between a task inserted on a runqueue and marked as TASK_RUNNING until it is actually running (executed on a processor), so the previous state isn't blocked (TASK_STOPPED) or sleep (TASK_INTERRUPTIBLE / TASK_UNINTERRUPTIBLE). Follow the enqueue_task() function and you will see what I truly mean :) > pcnt gives the number of timeslices that the process has had on the > cpu (A timeslice is a period of time that a process is allowed to > run, before surrendering the CPU to another process.. The amount of > time is given by the TIMESLICE constant, and it's implemented by > using a timer IRQ set to that amount of time). maybe, to avoid confusion, I prefer to interpret "pcnt" as the "how many times a certain task has been executed until this very moment". >Similarly, > last_queued gives a measure of how long it's been since the process > was queued for the CPU. I don't know what this value would be when > the process is waiting for IO; It may just equal the last time the > process was waiting, and perhaps it would be updated once the IO > completes, ready to run the process again. no, last_queued is a timestamp, so this is not an interval :) the complete defintion is "the time a task enters a runqueue and thus marked as TASK_RUNNING". And yes, it is measured in jiffies :) to Mike: not bad, man.... :) you receive A grade IMHO :) regards Mulyadi -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
