From: Keir Fraser Subject: Re: [Patch] support of lock profiling in Xen Date: Thu, 08 Oct 2009 10:17:29 +0100 Message-ID: References: <4ACDA516.6030800@ts.fujitsu.com> Mime-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <4ACDA516.6030800@ts.fujitsu.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Juergen Gross Cc: "xen-devel@lists.xensource.com" List-Id: xen-devel@lists.xenproject.org On 08/10/2009 09:38, "Juergen Gross" wrote: >> There's also the question of whether this really gets you useful extra info >> than can be digested from a xentrace run. I suppose at least the info is >> easier to understand, and not many people know how to drive xentrace. > > It helped me a lot for narrowing down my performance problem with a 8 vcpu > BS2000 system (this was the main reason I made the patch). > Finding such a problem in xentrace is quite a bit of work. xentrace produced > over 500 MB of data in 30 seconds, while reading the xenlockprof output > revealed the bottleneck in seconds. > > And I think the lock profiling tool is much better for evaluating > optimizations in Xen regarding lock conflicts, if the differences are only > a few percent. Okay, well the only thing I'm dead against in a lockmeter patch is requring dynamic registration and unregistration of individual locks. Largely because I see a never-ending stream of fix this unregistration, fix that unregistration, stretching out before me, because callers will forever be getting broken. I suppose most locks are either global or per-domain, so perhaps you could just have domain create/destroy hooks into lockmeter, to alloc/free an array of per-domain locks? Would that actually be an argument for a perfc-style central declaration of all profileable locks, so you could indicate which are global and which are per-domain? -- Keir