Re: elevator algorithm bug in ll_rw_blk.c

Mark Hahn (hahn@coffee.psychology.mcmaster.ca)
Mon, 16 Nov 1998 09:41:40 -0500 (EST)


> With that kind of knowledge, what kind of improvements could be made to
> the driver?

actually, it's the true geometry that a disk scheduler wants,
along with a model for how long a given seek will take.
hdparm gives neither: the geometry is false, because all non-ancient
disks are zoned, and it doesn't mention seeking (which is a
nonlinear function in any case), or head-switching time.

fortunately, none of this matters much, since seeking is still basically
monotonic with distance in _blocks_. this is why the current request
sorting (even before the fix) can easily deliver maximal bandwidth.
it basically only errs when you generate a new request at a lower
absolute address (thus deferred by the sort), that in truth is in
the current track, and would be very fast to fetch. that's a fairly
obscure pattern...

regards, mark hahn.

-
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/