From mboxrd@z Thu Jan 1 00:00:00 1970 From: Anthony Liguori Subject: Re: open/stat64 syscalls run faster on Xen VM than standard Linux Date: Mon, 28 Nov 2005 13:37:08 -0600 Message-ID: <438B5C64.5090505@us.ibm.com> References: <907625E08839C4409CE5768403633E0B0EAAC5@sefsexmb1.amd.com> <438B3B88.8040702@cs.uchicago.edu> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <438B3B88.8040702@cs.uchicago.edu> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: xuehai zhang Cc: "Petersson, Mats" , Kate Keahey , Xen Mailing List , Tim Freeman List-Id: xen-devel@lists.xenproject.org xuehai zhang wrote: > So, the benchmark experiments I've done so far suggests XenLinux using > loopback files as VBD backends shows better performance (faster > execution) on part of the system calls like open and stat64, but it > shows worse performance (slower execution) on other system calls like > write than the standard Linux. Does this mean different applications > may have different execution behaviors on VM than on the standard > Linux? In other words, some applications run faster on VM and some > slower, comparing with the physical machine? I really want to stress here that your results do not mean system calls are faster in Xen (or with a loopback device). You're seeing the effects of caching which are probably sub-optimal for overall performance. A more meaningful benchmark here is to take a syscall intensive workload, and measure the thoroughput. You'll likely find that the syscall performance is actually worse because when you start missing the cache you're going to get huge delays in read-ahead. Be wary of microbenchmarks, they often are way off from real-world scenarios. As for write performance, in general, read and write performance are going to be measurably slower in Xen than on bare-metal Linux. It almost has to be by definition. The hope is that the difference in performance will be neglible especially considering the improved thoroughput from greater utiliziation of the underlying hardware. Regards, Anthony Liguori