Re: 2.1.91 swap performance: jerky.

Dr. Werner Fink (werner@suse.de)
Fri, 27 Mar 1998 15:15:01 +0100

>
> What I observe is rather strange, using 'top' to watch things:
>
> the "cached" value suddenly starts getting bigger,
> as the system nearly halts while dumping stuff en masse
> to the swap partition.
>
> "cached" climbs to about 32MB of my 64MB total SDRAM,
> and then a flurry of reverse activity kicks in..
> the "cached" value drops again, as pages are read back
> in and response returns to the previously comatose system.
>
> "cached" levels off again around 2-5MB, until I start some
> new big application, whence the whole scenario repeats.

Could you try to change in linux/mm/filemap.c:shrink_mmap() the
new statement line 250 and following?

>From

/* Refuse to swap out all buffer pages */
if ((buffermem >> PAGE_SHIFT) * 100 < (buffer_mem.min_percent * num_physpages))
goto next;

To

/* Refuse to swap out all buffer pages */
if (bh && (buffermem >> PAGE_SHIFT) * 100 < (buffer_mem.min_percent * num_physpages))
goto next;

... then we catch the inode cache even if the buffer_mem.min_percent is
reached.

Werner

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu