From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755653AbZHYSLk (ORCPT ); Tue, 25 Aug 2009 14:11:40 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1755626AbZHYSLj (ORCPT ); Tue, 25 Aug 2009 14:11:39 -0400 Received: from mail-px0-f182.google.com ([209.85.216.182]:56094 "EHLO mail-px0-f182.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755641AbZHYSLi (ORCPT ); Tue, 25 Aug 2009 14:11:38 -0400 Message-ID: <4A94293A.2090103@vflare.org> Date: Tue, 25 Aug 2009 23:41:06 +0530 From: Nitin Gupta Reply-To: ngupta@vflare.org User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.1.1) Gecko/20090814 Fedora/3.0-2.6.b3.fc11 Thunderbird/3.0b3 MIME-Version: 1.0 To: Balbir Singh CC: akpm@linux-foundation.org, linux-kernel@vger.kernel.org, linux-mm@kvack.org, linux-mm-cc@laptop.org Subject: Re: [PATCH 4/4] compcache: documentation References: <200908241008.02184.ngupta@vflare.org> <661de9470908251003y3db1fb3awb648f9340cd0beb4@mail.gmail.com> In-Reply-To: <661de9470908251003y3db1fb3awb648f9340cd0beb4@mail.gmail.com> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 08/25/2009 10:33 PM, Balbir Singh wrote: >> +It consists of three modules: >> + - xvmalloc.ko: memory allocator > > I've seen your case for a custom allocator, but why can't we > > 1) Refactor slob and use it SLOB is fundamentally a different allocator. It looked at it in detail but could not image how can I make it suitable for the project. SLOB really does not fit it. > 2) Do we care about the optimizations in SLUB w.r.t. scalability in > your module? If so.. will xvmalloc meet those requirements? > Scalability is desired which xvmalloc lacks in its current state. My plan is to have a wrapper around xvmalloc that creates per-cpu pools and leave xvmalloc core simple. Along with this, detailed profiling needs to be done to see where the bottlenecks are in the core itself. > > What level of compression have you observed? Any speed trade-offs? > All the performance numbers can be found at: http://code.google.com/p/compcache/wiki/Performance I also summarized these in patch [0/4]: http://lkml.org/lkml/2009/8/24/8 The compression ratio is highly workload dependent. On "generic" desktop workload, stats show: - ~80% of pages compressing to PAGE_SIZE/2 or less. - ~1% incompressible pages. For the speed part, please refer to performance numbers at link above. It show cases where it help or hurts the performance. Thanks, Nitin