From mboxrd@z Thu Jan 1 00:00:00 1970 From: =?UTF-8?B?Um9nZXIgUGF1IE1vbm7DqQ==?= Subject: Analysis of PV(H) build times Date: Mon, 8 Jun 2015 13:10:35 +0200 Message-ID: <5575782B.20109@citrix.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Received: from mail6.bemta14.messagelabs.com ([193.109.254.103]) by lists.xen.org with esmtp (Exim 4.72) (envelope-from ) id 1Z1uxN-0008RG-Vt for xen-devel@lists.xenproject.org; Mon, 08 Jun 2015 11:11:06 +0000 List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xen.org Errors-To: xen-devel-bounces@lists.xen.org To: xen-devel List-Id: xen-devel@lists.xenproject.org Hello, I've done a small analysis of PV and PVH build and launch times using both xl and a custom builder [1] based on init-xenstore-domain. The purpose of this analysis is to measure the build time of PV(H) guests. In order to do this rdtsc calls have been added to the start of the main xl function and the xen_start_kernel function in Linux (first C function called by the kernel). The time measured is not only xl build time, but the time since the xl create call until the guest starts executing. The results can be found at: https://docs.google.com/spreadsheets/d/1xePTI6r1INAIQR901P_lFHFT9ZpduSLpRnKFDl_iB-Q/edit?usp=sharing A couple of conclusions can be obtained from the results: - xl/libxl overhead is very minimal compared to the simple loader (~10%). This is probably due to the extra computing done in libxl and the fact that the simple loader doesn't populate xenstore paths. - Most of time is spent in xc_dom_parse_image (~70%) for both PV and PVH domain build. - PVH build time is greater than PV by ~10%, due to xc_dom_boot_mem_init being more expensive for PVH domains. Roger. [1] http://xenbits.xen.org/people/royger/sbuilder.c