On Wed, 11 Feb 2009 19:02:14 +0100, Rob van der Heij <[email protected]> wrote:
...(deleted for brevity)... It's certainly a fair point that a complete picture is formed only with b oth actual measurements of "real world" representative workloads as well as t he more theoretical knowledge of how the code actually works internally. As the development team leader and design owner of the portion of CP involve d, I was speaking primarily from the latter perspective, I hope that was generally understood. I believe that what I stated was correct, however: although CP certainly does not "hold on" to old paging slots as tenacious ly as some folks believe, Linux is by no means obligated to inform CP every time it discards or remaps page contents, which limits CP's ability to fr ee the old paging slots. There's a lot of "it depends" in there, between th e code algorithms and the observed behaviors. My statements were (are) generally not based on assumptions, but rather o n how the code actually works. I don't need measurements to explain how th e code works. Of course, for any real world workload, how the design manifests in terms of externally visible behaviors is definitely the sort of "your mileage may vary" case where the measuring approach of course has immense value. - Bill Holder, z/VM Development, IBM Endicott
