From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Original-To: caml-list@yquem.inria.fr Delivered-To: caml-list@yquem.inria.fr Received: from mail3-relais-sop.national.inria.fr (mail3-relais-sop.national.inria.fr [192.134.164.104]) by yquem.inria.fr (Postfix) with ESMTP id D10AFBBAF for ; Sun, 21 Nov 2010 20:26:48 +0100 (CET) X-IronPort-Anti-Spam-Filtered: true X-IronPort-Anti-Spam-Result: AmsKAFwB6UxKfVM0gWdsb2JhbACGQI1lhikBhzZNCBUBAQsJCgcTAx9GonyJY4IYg08uiFkBAQMFgm2CWQSKXoV9 X-IronPort-AV: E=Sophos;i="4.59,232,1288566000"; d="scan'208";a="67776748" Received: from mail-gw0-f52.google.com ([74.125.83.52]) by mail3-smtp-sop.national.inria.fr with ESMTP; 21 Nov 2010 20:26:48 +0100 Received: by gwb11 with SMTP id 11so923702gwb.39 for ; Sun, 21 Nov 2010 11:26:47 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:received:received:in-reply-to :references:date:message-id:subject:from:to:cc:content-type; bh=n0APTMCH9WjjCxlsw6Ky/ej6Tuvf44jXegCMelvZD3k=; b=XIppfjvuPUNrdvkQldobRmspix3CpjuOO8zB2IAcC6/cJYG2/V8jyoHntXLt7c+NvE VUkW8yS7IGW1o2nxGc3tn+p6+JWGUe3xnCfl02pBllCG4irXFsf5q16vhNnOgdlt3vHf 18evBKDJ/TXsXyqoj5nvaz1f+qGaOQTTv5qqQ= DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type; b=pxSQJ1iZE8Iy3eezsG0e5xawwtJzNdxSprzjc4YS1OmuJFlIUUXj5h4hCWlOQClfxw aeXnQ/TCgIRpWkgSIECm+Le6x7SUaBxRJzeqjcknUQtMGyn33wY8N+HunDKHRLie1UE5 t+fB8AlyU8iTIdEPRqREeiFDZGynijKV3Mbfo= MIME-Version: 1.0 Received: by 10.90.27.14 with SMTP id a14mr4527463aga.113.1290367606808; Sun, 21 Nov 2010 11:26:46 -0800 (PST) Received: by 10.91.154.3 with HTTP; Sun, 21 Nov 2010 11:26:46 -0800 (PST) In-Reply-To: References: <4CE68FAB.6020102@elehack.net> Date: Sun, 21 Nov 2010 21:26:46 +0200 Message-ID: Subject: Re: [Caml-list] Optimizing garbage collection From: Eray Ozkural To: Alexandre Pilkiewicz Cc: Michael Ekstrand , caml-list@yquem.inria.fr Content-Type: multipart/alternative; boundary=001636163d99d560fa0495951e6b X-Spam: no; 0.00; eray:01 ozkural:01 eray:01 ozkural:01 speedup:01 speedup:01 bilkent:01 bilkent:01 garbage:01 wrote:01 wrote:01 heap:01 heap:01 caml-list:01 minor:01 --001636163d99d560fa0495951e6b Content-Type: text/plain; charset=ISO-8859-1 Hello there Alexandre, On Sun, Nov 21, 2010 at 8:13 PM, Alexandre Pilkiewicz < alexandre.pilkiewicz@polytechnique.org> wrote: > Hi > > 2010/11/20 Eray Ozkural : > > Yes, the default minor heap size was indeed too low, I've been trying to > set > > it to a higher value, now testing with the OCAMLRUNPARAM settings you > > recommended. It did result in some speedup, but not an awful lot, it's > > important to profile it as you say. > > Can you tell us how high you set it? I would recommend at least > 524288, and even something like 3000000 if you really need to (I'm > talking in words here) > I've set it to 4M and I think it's worked wonders, the collection time is no more so significant in gprof output (surprisingly) at least now I can identify the real bottlenecks! Indeed the 5-instr long fast path is quite fast. Due to the peculiarities of my code, it didn't result in much speedup but I've solved this problem, I can't believe I've overlooked the Gc parameters, I should probably be setting them from within the code. A bit embarrassed about it actually :) I've been thinking whether some kind of doubling strategy would work for the minor heap size. What do you think? Best, -- Eray Ozkural, PhD candidate. Comp. Sci. Dept., Bilkent University, Ankara http://groups.yahoo.com/group/ai-philosophy http://myspace.com/arizanesil http://myspace.com/malfunct --001636163d99d560fa0495951e6b Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable
Hello there Alexandre,

On Sun, Nov 21, 2010 at 8:1= 3 PM, Alexandre Pilkiewicz <alexandre.pilkiewicz@polytechnique.org&= gt; wrote:
Hi

2010/11/20 Eray Ozkural <examach= ine@gmail.com>:
> Yes, the default minor heap size was indeed too low,= I've been trying to set
> it to a higher value, now testing with the OCAMLRUNPARAM settings you<= br> > recommended. It did result in some speedup, but not an awful lot, it&#= 39;s
> important to profile it as you say.

Can you tell us how high you set it? I would recommend at least
524288, and even something like 3000000 if you really need to (I'm
talking in words here)
=A0
I've set it t= o 4M and I think it's worked wonders, the collection time is no more so= significant in gprof output (surprisingly) at least now I can identify the= real bottlenecks! Indeed the 5-instr long fast path is quite fast. Due to = the peculiarities of my code, it didn't result in much speedup but I= 9;ve solved this problem, I can't believe I've overlooked the Gc pa= rameters, I should probably be setting them from within the code. A bit emb= arrassed about it actually :)

I've been thinking whether some kind of doubling st= rategy would work for the minor heap size. What do you think?
Best,

--
Eray Ozkural, PhD candidate.=A0 Com= p. Sci. Dept., Bilkent University, Ankara
http://groups.yahoo= .com/group/ai-philosophy
h= ttp://myspace.com/arizanesil ht= tp://myspace.com/malfunct

--001636163d99d560fa0495951e6b--