Re: direct (unbufferd) disk access
Matthew Jacob (mjacob@feral.com)
Mon, 28 Jun 1999 13:16:47 -0700 (PDT)
>
> On Mon, 28 Jun 1999 09:24:06 -0700 (PDT), Matthew Jacob
> <mjacob@feral.com> said:
>
> >> On Sat, 26 Jun 1999 00:06:53 -0400, Douglas Gilbert
> >> Given that we have clocked reasonably standard i386 hardware with fast
> >> disks at 50 or 60 MB/sec through the filesystem, I doubt that the
> >> indirect IO is the bottleneck in those cases.
>
> > Without cache pollution?
>
> Page cache pollution is a totally different issue. At 60MB/sec
> we aren't even close to the performance levels where memory bandwidth
> is an issue, so extra copies into the cache are irrelevant from a
> bandwidth point of view. Of _course_ raw IO has less impact on
> memory, but that's another point entirely.
What I mean was: raw I/O is an application policy statement that says
"leaving this data in the buffer cache is not what I want". You certainly
do have some overhead in throwing stuff away if it's sitting in the buffer
cache.
-
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/