2002-10-07 20:10:43

by Andrew Morton

[permalink] [raw]
Subject: Re: The reason to call it 3.0 is the desktop (was Re: [OT] 2.6 not 3.0 - (NUMA))

Daniel Phillips wrote:
>
> This is easy to verify: say you have 100 MB of kernel source stored in, say,
> 50 different clumps on disk.

Disks use segmentation on their readahead buffers. Typically four-way.
So they will only buffer four different chunks of disk at a time.

If you're reading from 50 different places on disk, the disk keeps
invalidating readahead at the segment level.


2002-10-07 20:22:20

by Daniel Phillips

[permalink] [raw]
Subject: Re: The reason to call it 3.0 is the desktop (was Re: [OT] 2.6 not 3.0 - (NUMA))

On Monday 07 October 2002 22:14, Andrew Morton wrote:
> Daniel Phillips wrote:
> >
> > This is easy to verify: say you have 100 MB of kernel source stored in, say,
> > 50 different clumps on disk.
>
> Disks use segmentation on their readahead buffers. Typically four-way.
> So they will only buffer four different chunks of disk at a time.
>
> If you're reading from 50 different places on disk, the disk keeps
> invalidating readahead at the segment level.

Sure, and kernel-based physical readahead would not have that problem.
(Kernel-based physical readahead has its own problems, for example: how
do you determine that a given physical block is already cached in an
inode and so should be ignored as a readahead candidate?)

--
Daniel