Re: PATCH: rewritten bdflush

Zack Weinberg (zack@rabi.columbia.edu)
Wed, 12 May 1999 14:43:50 -0400


On Wed, 12 May 1999 14:32:28 -0400 (EDT), Steve Willer wrote:
>
>On Tue, 11 May 1999, Zack Weinberg wrote:
>
>> There should be no change in what bdflush actually does. I've been
>> running 2.2.7-as-patched for 13 hours and it works fine; there was a
>> six-hour compile job in that interval. I'm sure others can stress it
>> harder though, I'd be interested in results.
>
>I made a separate posting of this, but this one will have actual data in
>it.
>
>Your patch doesn't seem to scale up to heavy writes, like you might find
>in a transaction-oriented database. Running without update, the bdflush
>doesn't seem to be able to keep up with the I/O load (without making me
>very nervous about recoverability and the age of buffers). Ideally, the
>writes are done at a pace that's equal to a per-30-second sync, but spread
>over the full 30 seconds. In other words, the ideal is what update used to
>do. :-)

I think I know what is doing this, but I would like to be certain. The
max-writes-per-cycle choke didn't used to apply to the five-second
writebacks, now it does. I didn't think that would cause trouble, but
I appear to have been wrong. Can you please rerun your test after
doing

# echo "40 5000 64 256 5 3000 500 1884 2" >/proc/sys/vm/bdflush

This increases the maximum number of pages to write back per wakeup
from 500 to 5000.

zw

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