Date: Tue, 10 Oct 1995 15:26:30 -0700 From: Poul-Henning Kamp <phk@freefall.freebsd.org> To: current Subject: phkmalloc & Olliviers numbers. Message-ID: <7574.813363990@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
Thanks for the numbers Ollivier, I have looked at them and I like them :-) I read the negative numbers in the "waste" to mean that the final sbrk isn't by definition the highest one, though I cannot know for sure I belive that is the only sane explanation. Anyway, if that is the case, then that would explain the relative big waste of phkmalloc, since it as default keeps 400K (100 pages) before starting to hand things back with sbrk(2). The reasoning for this is that sbrk(2) is relative expensive to do, you have to search and fiddle page-tables, and you have to bzero pages, so a pagein may actually be cheaper sometimes, and since phkmalloc doesn't touch these pages, they will not be faulted in until they are needed. You can try to modify this if you want, search for "malloc_cache = 100 <<". Apart from that I'm flattered :-) Can you try to run a test where you allocate a couple of megs more than you have RAM for, so that you start to page ? That is when phkmalloc >really< wins over gnu & the old bsd mallocs, and I'd like to see how the "DL" and "MR" mallocs compare on this metric. I guess if anything phkmalloc may lead to a slight increas in swap usage, but a drop in swap activity, but I defy anybody to measure it, and if you do anyway, you should have spent your time doing something useful instead :-) Thanks ! Poul-Henning
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?7574.813363990>