Todd Lipcon <t...@lipcon.org> wrote: > I've seen this kind of bimodal performance behavior even in native apps. On > some builds, a benchmark will consistently run in 6 seconds, and on the > next build, it will go back to 5 seconds.
I have heard of similar effects with a single build where the variation between runs is due to differences in page mapping, e.g. in this link where the author discusses "Sometimes the OS hands us a range of physical pages that works well for our workloads, other times it doesn’t." https://www.pvk.ca/Blog/2012/07/30/binary-search-is-a-pathological-case-for-caches/ Tony. -- f.anthony.n.finch <d...@dotat.at> http://dotat.at/ - I xn--zr8h punycode Fitzroy: Southwesterly veering northwesterly later in northwest, 4 or 5 increasing 6 at times. Moderate or rough, occasionally very rough at first in north. Occasional rain. Good, occasionally moderate.
