> It is. Not unplugging the queue results in higher throughput
> when running a benchmark load, but seems to really harm
> system throughput (and cause stalls) in /real/ loads.
>
> This is most likely due to the fact that in most real life
> loads we have to write data and metadata all over the place
> and write clustering isn't as effective as in benchmark
> loads (where we write 500MB of "data" at full speed to one
> file).
Interesting.
This brings me to another point. We probably want some generic
interfaces in ll_rw_blk to unplug individual queues, where you can either
specify either the actual queue a kdev_t. Some of the places, (such as
__wait_on_buffer()) it might make more sense to unplug just the queue
related to the buffer in question rather than just unplug everything by
running tq_disk. On single-disk systems there wouldn't be any appreciable
difference, but if you had a lot of disks, it would probably work out to be
more efficient. Something to think about....
To fix this specific problem for 2.4, the easiest and safest thing to do
is to run tq_disk when we inject a SCSI ioctl (iff the queue is plugged).
Right now there is no way to remove individual entries from a task queue
without running the entire list, and calling the unplug routine while the
thing is still in the list would *appear* to be unsafe in that if the queue
were re-plugged the tq_disk list could be corrupted.
-Eric
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Thu Sep 07 2000 - 21:00:31 EST