Re: [OPW kernel] Re: [RFC] mm:prototype for the updated swapoff implementation

[Josh Triplett]

On Wed, Feb 19, 2014 at 04:39:47PM -0500, Rik van Riel wrote:
> On 02/19/2014 04:27 PM, Andrew Morton wrote:
> > On Tue, 18 Feb 2014 16:35:22 -0800 Kelley Nielsen <kelleynnn@xxxxxxxxx> wrote:
> > 
> >> The function try_to_unuse() is of quadratic complexity, with a lot of
> >> wasted effort. It unuses swap entries one by one, potentially iterating
> >> over all the page tables for all the processes in the system for each
> >> one.
> >>
> >> This new proposed implementation of try_to_unuse simplifies its
> >> complexity to linear. It iterates over the system's mms once, unusing
> >> all the affected entries as it walks each set of page tables. It also
> >> makes similar changes to shmem_unuse.
> >>
> >> Improvement
> >>
> >> swapoff was called on a swap partition containing about 50M of data,
> >> and calls to the function unuse_pte_range were counted.
> >>
> >> Present implementation....about 22.5M calls.
> >> Prototype.................about  7.0K   calls.
> > 
> > Do you have situations in which swapoff is taking an unacceptable
> > amount of time?  If so, please update the changelog to provide full
> > details on this, with before-and-after timing measurements.
> 
> I have seen plenty of that.  With just a few GB in swap space in
> use, on a system with 24GB of RAM, and about a dozen GB in use
> by various processes, I have seen swapoff take several hours of
> CPU time.

And it's clear what the lower bound on swapoff should be: current amount
of swap in use, divided by maximum disk write speed.  We're definitely
not to *that* point yet; this ought to get us a lot closer.

- Josh Triplett
--
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/