Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S261471AbVAMTuq (ORCPT ); Thu, 13 Jan 2005 14:50:46 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S261473AbVAMTrk (ORCPT ); Thu, 13 Jan 2005 14:47:40 -0500 Received: from one.firstfloor.org ([213.235.205.2]:60099 "EHLO one.firstfloor.org") by vger.kernel.org with ESMTP id S261442AbVAMTn6 (ORCPT ); Thu, 13 Jan 2005 14:43:58 -0500 To: "Justin M. Forbes" Cc: sander@humilis.net, Kernel Mailing List , "Sergey S. Kostyliov" Subject: Re: NUMA or not on dual Opteron References: <200501121824.44327.rathamahata@ehouse.ru> <20050113094537.GB2547@favonius> <41E6472B.5020701@imag.fr> <20050113170733.GA14524@linuxtx.org> From: Andi Kleen Date: Thu, 13 Jan 2005 20:43:54 +0100 In-Reply-To: <20050113170733.GA14524@linuxtx.org> (Justin M. Forbes's message of "Thu, 13 Jan 2005 11:07:33 -0600") Message-ID: User-Agent: Gnus/5.110002 (No Gnus v0.2) Emacs/21.3 (gnu/linux) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 1490 Lines: 32 "Justin M. Forbes" writes: > This is somewhat true. There are 2 types of dual opteron boards. Those in > the $200 US range only have one memory bank, which is attached to CPU0. > They operate as a single node, and may perform better with numa turned off. > Those in the $400+ range tend to have one bank per CPU and will certainly > perform better with numa on. They do usually have a bios option to > interleave the nodes which would show up as a single node, and probably > perform better with numa turned off, but a better solution is to turn off > the node interleave in bios and run the kernel with numa support. > Basically if you have 2 CPUs and only one memory bank, maybe turning numa > off will give better performance, but if you have one memory bank per CPU > numa should be on. [agreed, except:] There is no significant performance difference between NUMA on and off on the crippled boards. So best is to always enable CONFIG_NUMA on x86-64. The distribution kernels for x86-64 should ship in this configuration too. I would recommend to always disable node interleaved on the non crippled board with multiple memory banks. Most boards ship with this disabled by default, only a few enable it. -Andi - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/