Re: [PATCH 1/9] readahead: introduce PG_readahead

From: Fengguang Wu
Date: Mon Jun 11 2007 - 22:52:24 EST

Hi Rusty,

On Tue, Jun 12, 2007 at 11:04:54AM +1000, Rusty Russell wrote:
> On Thu, 2007-05-17 at 06:47 +0800, Fengguang Wu wrote:
> > plain text document attachment (mm-introduce-pg_readahead.patch)
> > Introduce a new page flag: PG_readahead.
> >
> > It acts as a look-ahead mark, which tells the page reader:
> > Hey, it's time to invoke the read-ahead logic. For the sake of I/O pipelining,
> > don't wait until it runs out of cached pages!
> Hi Fengguang!
> I've been reading your patches, and I have some (possibly dumb!)
> questions.
> For this patch: why set a bit in the page, rather than keep a value
> inside the "struct file_ra_state"?

Good question. I should have documented it in the patch ;)

The short answer:
there can be multiple read streams per fd, i.e. interleaved reads.

file_ra_state can not easily track all of the streams. Solaris zfs
does it by managing a list in struct dnode. While PG_readahead plus
the context based readahead( makes
another solution. The two solutions are comparable in complexity.
The context based readahead is a bit more feature rich: it is sensible
to memory pressure, and supports an unlimited number of streams. Note
that random reads can be regarded as a huge number of one-shot streams,
which can interfere with zfs's list-based readahead states badly.


To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at