> takes 45 seconds CPU time due to the O(clean * dirty) algorithm in
> truncate_inode_pages().  The machine is locked up for the duration.
> The patch reduces this to 20 milliseconds via an O(clean + dirty)
> algorithm.

Unfortunately, it's _not_ O(clean + dirty).

> +		while (truncate_list_pages(&mapping->clean_pages, start, &partial)) {
> +			spin_lock(&pagecache_lock);
> +			complete = 0;
> +		}

Cool. Now think what happens if pages with large indices are in the
very end of list. Half of them. You skip clean/2 pages on each of
clean/2 passes. Hardly a linear behaviour - all you need is a different
program to trigger it.

Now, having a separate pass that would reorder the pages on list,
moving the to-kill ones in the beginning might help.

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