From mboxrd@z Thu Jan 1 00:00:00 1970 From: xuehai zhang Subject: Re: question about disk performance in domU Date: Mon, 21 Nov 2005 16:41:53 -0600 Message-ID: <43824D31.4080600@cs.uchicago.edu> References: <6C21311CEE34E049B74CC0EF339464B924B51D@cacexc12.americas.cpqcorp.net> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <6C21311CEE34E049B74CC0EF339464B924B51D@cacexc12.americas.cpqcorp.net> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: "Santos, Jose Renato G" Cc: keahey@mcs.anl.gov, Tim Freeman , xen-devel@lists.xensource.com List-Id: xen-devel@lists.xenproject.org Renato and Tim, Thank you for your replies. Is there anyway to flush the dom0 file buffer before running hdparm on domU each time? Xuehai Santos, Jose Renato G wrote: > >>>-----Original Message----- >>>From: xen-devel-bounces@lists.xensource.com >>>[mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of >>>Tim Freeman >>>Sent: Monday, November 21, 2005 12:45 PM >>>To: xen-devel@lists.xensource.com >>>Cc: keahey@mcs.anl.gov; xuehai zhang >>>Subject: Re: [Xen-devel] question about disk performance in domU >>> >>> >>>So the "Timing buffered disk reads" show much higher >>>results. I see the DMA zone is larger in domU, but since >>>this is mapped to a loopback file I'm guessing the physical >>>disk performance should only be affected by what dom0's DMA >>>zone is set to on the node running the domU if that is all >>>that was going on. But dom0 performance seems comparable to >>>native linux. >>> >>>Is this huge "timing buffered disk read" difference >>>accurate? Does the domU benefit from some other cache of the >>>loopback file in dom0? > > > Yes. It seems to me that this is the effect of the file buffer > cache on dom0. hdparm flushes the file cache on domU > to make sure that there is no data in the file buffer cache > when measuring device access times (reported as buffered disk > reads). However when using VBDs mapped to files, the data is > also cached on dom0 file buffer. Therefore data is not coming > directly from the device but from dom0 file cache. Note how the > amount of data that is read in 3 sec increases at each step, > since data read in previous steps come from dom0 file cache. >