Strange scheduling behavoir in SMP (kernel 2.2.14)

[Jeffrey B. Siegal]

> That's why if you have a background task and two interactive tasks on a
> 2-CPU machine, you really cannot avoid the background task jumping
> between CPUs unless you are willing to leave a CPU idle at times (and
> that, in general, is a very bad thing to do.)

No, in general for scheduling problems, leaving a resource idle can be a
useful thing to do.  In specific cases it can be a bad thing.

We discussed this last year relating to disk I/O scheduling.  If you have just
completed an I/O on the top end of a disk, and the next queued I/O is on the
bottom end of the disk, it can make sense to wait a short time to see if
another request arrives for the top end before seeking all the way down to the
bottom.

This situation is similar.  It depends on the pattern of requests and the
relative cost of a state change.  As caches get bigger and faster the cost
increases and it becomes more likely to be beneficial to wait.  Similarly,
disk transfer rates have been increasing a lot faster than seek times, making
anticipatory scheduling more likely to be correct.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo en vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/