Messages in this thread | | | Date | Fri, 19 Feb 2016 15:20:00 -0500 | From | Johannes Weiner <> | Subject | Re: [PATCH] mm: scale kswapd watermarks in proportion to memory |
| |
On Fri, Feb 19, 2016 at 11:25:43AM +0000, Mel Gorman wrote: > On Thu, Feb 18, 2016 at 11:41:59AM -0500, Johannes Weiner wrote: > > In machines with 140G of memory and enterprise flash storage, we have > > seen read and write bursts routinely exceed the kswapd watermarks and > > cause thundering herds in direct reclaim. Unfortunately, the only way > > to tune kswapd aggressiveness is through adjusting min_free_kbytes - > > the system's emergency reserves - which is entirely unrelated to the > > system's latency requirements. In order to get kswapd to maintain a > > 250M buffer of free memory, the emergency reserves need to be set to > > 1G. That is a lot of memory wasted for no good reason. > > > > On the other hand, it's reasonable to assume that allocation bursts > > and overall allocation concurrency scale with memory capacity, so it > > makes sense to make kswapd aggressiveness a function of that as well. > > > > Change the kswapd watermark scale factor from the currently fixed 25% > > of the tunable emergency reserve to a tunable 0.001% of memory. > > > > On a 140G machine, this raises the default watermark steps - the > > distance between min and low, and low and high - from 16M to 143M. > > > > Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> > > Intuitively, the patch makes sense although Rik's comments should be > addressed. > > The caveat will be that there will be workloads that used to fit into > memory without reclaim that now have kswapd activity. It might manifest > as continual reclaim with some thrashing but it should only apply to > workloads that are exactly sized to fit in memory which in my experience > are relatively rare. It should be "obvious" when occurs at least.
This is a problem only in theory, I think, because I doubt anybody is able to keep a workingset reliably at a margin of less than 0.001% of memory. I'd expect few users to even go within single digit margins without eventually thrashing anyway.
It certainly becomes a real issue when users tune the scale factor, but then it will be a deliberate act with known consequences. That's what I choose to believe in.
> Acked-by: Mel Gorman <mgorman@suse.de>
Thanks!
| |