From mboxrd@z Thu Jan 1 00:00:00 1970 From: Leonid Podolny Subject: multipath performance problem Date: Thu, 17 Mar 2011 15:08:01 +0200 Message-ID: <4D8207B1.1060005@xtremio.com> Reply-To: device-mapper development Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; Format="flowed" Content-Transfer-Encoding: 7bit Return-path: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: dm-devel-bounces@redhat.com Errors-To: dm-devel-bounces@redhat.com To: dm-devel@redhat.com List-Id: dm-devel.ids Hi, I'm not sure it's a correct place to write to, but I'm having a weird issue with the multipath which I'm unable to solve on my own. There is a SAN device that exports each LUN 8 times. I measure performance using btest (http://http://sourceforge.net/projects/btest/). Basically, it writes 4kb chunks of data from multiple threads and gathers statistics -- throughput in IOPs, average latency and such. When I compare results of writing into /dev/sdXXX's vs. writing into /dev/dm-XXX's, I see that the multipath devices have about 4 times worse throughput (in IOPs), but about 2 times less latency. In addition, the SAN device allows to see how many commands are currently being processed, and when I work via multipath, this number is many times lower. All the above shows that I am unable to properly saturate the SAN device when working through multipath. I tried disabling the queue in multipathd. What do I miss? -- Cheers, L.