On Thu, Aug 08, 2019 at 08:08:17AM +1000, Dave Chinner wrote:
> On Wed, Aug 07, 2019 at 04:03:16PM +0100, Mel Gorman wrote:
> > > > inode cache shrinker is not being called when large batches of pages
> > > > are being reclaimed. In roughly the same time period that it takes
> > > > to fill memory with 50% pages and 50% slab caches, memory reclaim
> > > > reduces the page cache down to just dirty pages and slab caches fill
> > > > the entirity of memory.
> > > > 
> > > > At the point where the page cache is reduced to just the dirty
> > > > pages, there is a clear change in write IO patterns. Up to this
> > > > point it has been running at a steady 1500 write IOPS for ~200MB/s
> > > > of write throughtput (data, journal and metadata).
> > 
> > As observed by iostat -x or something else? Sum of r/s and w/s would
> 
> PCP + live pmcharts. Same as I've done for 15+ years :)
> 
> I could look at iostat, but it's much easier to watch graphs run
> and then be able to double click on any point and get the actual
> value.
> 
> I've attached a screen shot of the test machine overview while the
> vanilla kernel runs the fsmark test (cpu, iops, IO bandwidth, XFS
> create/remove/lookup ops, context switch rate and memory usage) at a
> 1 second sample rate. You can see the IO patterns change, the
> context switch rate go nuts and the CPU usage pattern change when
> the page cache hits empty.

And now attached. :)

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com