From mboxrd@z Thu Jan 1 00:00:00 1970 From: Daniel Hofmann Subject: Re: Forcing Ceph into mapping all objects to a single PG Date: Thu, 31 Jul 2014 02:29:22 +0200 Message-ID: <53D98DE2.7080802@trvx.org> References: <53CD93B4.6020500@trvx.org> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Return-path: Received: from trvx.org ([46.38.250.125]:44375 "EHLO mail.trvx.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753569AbaGaA3f (ORCPT ); Wed, 30 Jul 2014 20:29:35 -0400 Received: from [192.168.1.115] (HSI-KBW-046-005-197-128.hsi8.kabel-badenwuerttemberg.de [46.5.197.128]) by mail.trvx.org (Postfix) with ESMTPSA id 22EE41A254CB for ; Thu, 31 Jul 2014 02:29:33 +0200 (CEST) In-Reply-To: <53CD93B4.6020500@trvx.org> Sender: ceph-devel-owner@vger.kernel.org List-ID: To: ceph-devel@vger.kernel.org Follow up: I wrote a small Python script for evaluating the object to placement mapping's quality. I'm using a Chi-Squared statistical hypothesis test. Here is the script and also an example output for a few rados benchmark writes: > https://gist.github.com/daniel-j-h/d7d87dfe5de3c5bbfd0f Maybe you're interested in checking your cluster's mapping quality. More on the method and the output's meaning: > https://en.wikipedia.org/wiki/Chi-squared_test > http://en.wikipedia.org/wiki/P-value I'm still interested in a cluster-specific mapping as discussed in the other mailinglist posts. Cheers, Daniel