|
From: | Vadim V. Zhytnikov |
Subject: | Re: [Gcl-devel] GCL memory allocation and GC problems |
Date: | Wed, 28 Jan 2004 13:01:20 +0300 |
User-agent: | Mozilla/5.0 (Windows; U; Windows NT 5.1; ru-RU; rv:1.5) Gecko/20031006 |
Camm Maguire ?????:
Greetings! BTW, Vadim, if you have any suggestions on *preallocation* level adjustments, please let me know. (room) on starupt of ansi gcl shows most of the pages pushing the maximum boundaries. Take care,
Sorry for some delay. After all I was able to run some Maxima + current GCL CVS tests watching how results depends on initial preallocation. And they are quite nice. Mostly timing is virtually independent of initial preallocation - difference is around 10%. This is due to new growth parameters. On the other hand I see at least one test in which preallocation doubles performance. In this specific case Maxima/GCL does some long calculations which do not need a lot of memory - it results in a large number of GC without new allocation since free % remains below new allocation threshold. In this particular situation preallocation allows to reduce number of GC. Maybe it is worth to make initial allocation for CONS, FIXNUM, STRING and CFUN proportional to MAXPAGES? But I'm not sure about precise numbers yet. Best wishes, Vadim -- Vadim V. Zhytnikov <address@hidden> <address@hidden>
[Prev in Thread] | Current Thread | [Next in Thread] |