Re: [PATCH] Adaptive read-ahead v4

[WU Fengguang]

On Mon, Oct 17, 2005 at 07:30:34PM +0200, J?rn Engel wrote:
> On Sun, 16 October 2005 01:47:31 +0800, WU Fengguang wrote:
> >
> > @@ -131,6 +133,63 @@ struct page_state {
> >  
> >  	unsigned long pgrotated;	/* pages rotated to tail of the LRU */
> >  	unsigned long nr_bounce;	/* pages for bounce buffers */
> > +
> > +	unsigned long cache_miss;	/* read cache misses */
> > +	unsigned long readrandom;	/* random reads */
> > +	unsigned long pgreadrandom;	/* random read pages */
> > +	unsigned long readahead_rescue; /* read-aheads rescued*/
> > +	unsigned long pgreadahead_rescue;
> > +	unsigned long readahead_end;	/* read-aheads passed EOF */
> > +
> > +	unsigned long readahead;	/* read-aheads issued */
> > +	unsigned long readahead_return;	/* look-ahead marks returned */
> > +	unsigned long readahead_eof;	/* read-aheads stop at EOF */
> > +	unsigned long pgreadahead;	/* read-ahead pages issued */
> > +	unsigned long pgreadahead_hit;	/* read-ahead pages accessed */
> > +	unsigned long pgreadahead_eof;
> > +
> > +	unsigned long ra_newfile;	/* read-ahead on start of file */
> > +	unsigned long ra_newfile_return;
> > +	unsigned long ra_newfile_eof;
> > +	unsigned long pgra_newfile;
> > +	unsigned long pgra_newfile_hit;
> > +	unsigned long pgra_newfile_eof;
> > +
> > +	unsigned long ra_state;		/* state based read-ahead */
> > +	unsigned long ra_state_return;
> > +	unsigned long ra_state_eof;
> > +	unsigned long pgra_state;
> > +	unsigned long pgra_state_hit;
> > +	unsigned long pgra_state_eof;
> > +
> > +	unsigned long ra_context;	/* context based read-ahead */
> > +	unsigned long ra_context_return;
> > +	unsigned long ra_context_eof;
> > +	unsigned long pgra_context;
> > +	unsigned long pgra_context_hit;
> > +	unsigned long pgra_context_eof;
> > +
> > +	unsigned long ra_contexta;	/* accelerated context based read-ahead */
> > +	unsigned long ra_contexta_return;
> > +	unsigned long ra_contexta_eof;
> > +	unsigned long pgra_contexta;
> > +	unsigned long pgra_contexta_hit;
> > +	unsigned long pgra_contexta_eof;
> > +
> > +	unsigned long ra_backward;	/* prefetch pages for backward reading */
> > +	unsigned long ra_backward_return;
> > +	unsigned long ra_backward_eof;
> > +	unsigned long pgra_backward;
> > +	unsigned long pgra_backward_hit;
> > +	unsigned long pgra_backward_eof;
> > +
> > +	unsigned long ra_random;	/* read-ahead on seek-and-read-pages */
> > +	unsigned long ra_random_return;
> > +	unsigned long ra_random_eof;
> > +	unsigned long pgra_random;
> > +	unsigned long pgra_random_hit;
> > +	unsigned long pgra_random_eof;
> > +
> >  };
> 
> Without actually understanding what you're doing, wouldn't a struct
> for all those groups make sense?  I bet it can simplify the actual
> code as well.
Yeah, I did not expect the structure to grow so large.

It should be better to create a stand alone struct and create a
dedicated entry in /proc/. The data would be a nice hint for
administrators who care the performance.

Thanks,
Wu Fengguang
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/