From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Tue, 18 Sep 2001 10:53:01 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Tue, 18 Sep 2001 10:52:51 -0400 Received: from mail2.aracnet.com ([216.99.193.35]:52753 "EHLO mail2.aracnet.com") by vger.kernel.org with ESMTP id ; Tue, 18 Sep 2001 10:52:35 -0400 From: "M. Edward Borasky" To: "Neulinger, Nathan" , Subject: RE: How much performance hit from running SMP kernel on UP box? Date: Tue, 18 Sep 2001 07:54:08 -0700 Message-ID: MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit X-Priority: 1 (Highest) X-MSMail-Priority: High X-Mailer: Microsoft Outlook IMO, Build 9.0.2416 (9.0.2911.0) Importance: High In-Reply-To: <6CAC36C3427CEB45A4A6DF0FBDABA56D86D74D@umr-mail03.cc.umr.edu> X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2600.0000 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org This question as it stands is more or less meaningless, as are the rumored values of 5-10%. There are two types of performance measures commonly used: throughput -- the number of operations of a given type completed in a given amount of time -- and response time -- the time it takes to complete a given operation. To some extent, there are tradeoffs between the two. As load increases, all other things being equal, throughput will increase and so will response times for the operations. So I think the question you need to be asking is, for any specific benchmark or application, "How does throughput differ between the two environments with all other things -- load and response time, for example -- being equal?" or "How does response time differ between the two environments with all other things being equal?" The other point I think needs to be made is that, while performance of the *kernel* and its mechanisms is important to the folks on this list, it is *not* for the most part what people buy computers for. For the most part, people buy computers to accomplish *tasks*, for example running an e-commerce business, processing astronomical images, editing documents, creating electronic music, communicating with the Internet or operating a manufacturing production line. How the *application code* performs, and *how the kernel manages competing demands for resources from applications and their users* are what matters. In other words, an inefficient kernel is a bad thing, but inefficient applications on top of a perfect kernel are much much worse. -- M. Edward (Ed) Borasky http://www.aracnet.com/~znmeb mailto:znmeb@aracnet.com Stand-Up Comedy: Because Man Does Not Live By Dread Alone