Re: bogosity in blkelvset_ioctl

[Andrea Arcangeli <andrea _at_ suse.de>]

On Sat, 19 Aug 2000, Rik van Riel wrote:

>Heh, the new elevator is like 3 months (?) old...
>
>And it's _already_ suffering from bitrot?!

If bitrot means bit rotation I don't know what you mean with that.

It's sufferring one variable that gone away. max_bomb_segments doesn't
mean anything on 2.4.x.

The reason of the removal is very simple: people wants nominal performance
because none I/O benchmark out there cares about latency (just run dbench
and see where the first `+' shows up and see how long later the other +
shows up from the other threads...). I'm also wondering if dbench runs
faster by advacing a lot the progress of one thread w.r.t. the other
threads as we're doing as side effect right now. My guess is that is
reducing the pollution of the cache providing much better numbers. At
least tiobench doesn't have that very misleading issue because it doesn't
take advantage of the cache at all.

About making the number dynamic there's not such need, we just want to
avoid a 49hour stall, we reduce it to the order of seconds and we don't
need more tuning than that. The ioctl is there just in case somebody have
special needs (or troubles :).

Andrea

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