Re: mmap() is slower than read() on SCSI/IDE on 2.0 and 2.1

Chris Wedgwood (chris@cybernet.co.nz)
Sat, 19 Dec 1998 09:51:01 +1300


On Fri, Dec 18, 1998 at 09:05:26PM +0100, Rik van Riel wrote:

> Some form of read-behind will give us optimal performance
> when the tiles are a lot smaller (say 1/4th) than the VM
> 'stripe size' we are using.

Still wearning my "I'm not sure its always a good idea to second
guess the application hat" -- do we actually have any applications
that would make use of this?

The only one I can think of is that I have large databases (well, 1
GB or so) that have sequentially grown and at time, we do queries
that requires us walking backwards though the data here, but I don't
think it uses mmap.

I know other `theorietical applications' exist that produce this
pattern, but then I can find `theorietical applications' that have
any kind of screwy access pattern.

> The actual disk I/O, of course, is still done in forward
> direction :)

True... but if this is an issue, gobs of disk and RAID can help,
especially when it comes to random seeks.

-cw

-
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/