On Wed, 29 Apr 2020 10:18:06 +0100 Jonathan Cameron <Jonathan.Cameron@xxxxxxxxxx> wrote: > On Wed, 29 Apr 2020 09:49:54 +0200 > SeongJae Park <sjpark@xxxxxxxxxx> wrote: > > > On Tue, 28 Apr 2020 17:17:13 +0100 Jonathan Cameron <Jonathan.Cameron@xxxxxxxxxx> wrote: > > > > > On Tue, 28 Apr 2020 15:23:42 +0200 > > > SeongJae Park <sjpark@xxxxxxxxxx> wrote: > > > > > > > On Tue, 28 Apr 2020 13:27:04 +0100 Jonathan Cameron <Jonathan.Cameron@xxxxxxxxxx> wrote: > > > > > > > > > On Mon, 27 Apr 2020 14:04:27 +0200 > > > > > SeongJae Park <sjpark@xxxxxxxxxx> wrote: > > > > > > > > > > > From: SeongJae Park <sjpark@xxxxxxxxx> > > > > > > > > > > > > Introduction > > > > > > ============ > > > > > > [...] > > > > Appreciate your explanations and suggestions. > > You are welcome. > > Out of interest, do you have any comparative data on how 'accurate' the resulting > estimates are vs a more precise heatmap from a memory trace? No, I don't have such data. I'm only comparing the big trends of heatmap, working set sizes analyzed from the recorded access pattern and the DAMOS performance results for each version using my human eye, to check regression. > > I'm looking at gathering such data but much happier to leverage your work if > you've already done it! That would be great. If I get such data later, I will let you know. I will be also very happy if you could get it first and share with me. Maybe we could make and use another variant of DAMON, which uses page-size regions only and disable the adaptive regions adjustment. It will be also useful for overhead comparison. Actually, I heard that my previous colleague made this variant for the comparison based on a prototype of DAMON. I will also consider extending DAMON to support such variant. Also, if you need the heatmaps, analyzed working set size distribution, and/or the record file itself for each version of the patchsets, please let me know. Thanks, SeongJae Park
