From mboxrd@z Thu Jan 1 00:00:00 1970 From: John Kacur Subject: Re: Calculate average latencies on the fly Date: Tue, 22 Nov 2016 10:43:27 -0500 (EST) Message-ID: <477381688.1371680.1479829407972.JavaMail.zimbra@redhat.com> References: Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Cc: Piotr Gregor , Clark Williams , Linux RT Users To: Tracy Smith Return-path: Received: from mx6-phx2.redhat.com ([209.132.183.39]:49291 "EHLO mx6-phx2.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755131AbcKVPnf (ORCPT ); Tue, 22 Nov 2016 10:43:35 -0500 In-Reply-To: Sender: linux-rt-users-owner@vger.kernel.org List-ID: ----- Original Message ----- > John, > > Latency of 100,000 seems to me to be a bug when doing floating point long > division, which potentially accumulates small errors. > > Is the issue in the cyclictest calculation or in the kernel? Or, the FPU? > > If in the kernel, where do we need to trace to check for increased latency > when doing floating point long division? > > Piotr, needs to trace where the latency is occurring and fix. Any > recommendations on a possible fix at this stage? > > Thx, Tracy > Sure, you make some good points, it could be a cyclictest bug, it could be in the kernel, he also said he was doing this in a virtualized system, it could be somewhere in that stack. The first step of course is to see if you can come up with a way to reliably reproduce it. Thanks John