From mboxrd@z Thu Jan 1 00:00:00 1970 From: Linus Torvalds Subject: Re: Name hashing function causing a perf regression Date: Sun, 14 Sep 2014 20:37:32 -0700 Message-ID: References: <54136EEF.4050209@fb.com> <201409151149.IBB76963.FtLOFOQSHMFVOJ@I-love.SAKURA.ne.jp> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Cc: Josef Bacik , Andi Kleen , linux-fsdevel , Al Viro , Christoph Hellwig , Chris Mason To: Tetsuo Handa Return-path: Received: from mail-vc0-f182.google.com ([209.85.220.182]:56518 "EHLO mail-vc0-f182.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752307AbaIODhd (ORCPT ); Sun, 14 Sep 2014 23:37:33 -0400 Received: by mail-vc0-f182.google.com with SMTP id le20so2918525vcb.13 for ; Sun, 14 Sep 2014 20:37:32 -0700 (PDT) In-Reply-To: <201409151149.IBB76963.FtLOFOQSHMFVOJ@I-love.SAKURA.ne.jp> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Sun, Sep 14, 2014 at 7:49 PM, Tetsuo Handa wrote: > > I care, at least as of RHEL6 (2.6.32) kernels. Some users are using tmpfs > in order to avoid dentry cache bouncing bomb (a series of simple stat() calls > take about 10 seconds when there are 300,000,000+ unused dentries, P31 of > http://I-love.SAKURA.ne.jp/tomoyo/LCJ2014-en.pdf ). I don't know performance > on recent kernels. Hmm. Probably not hugely different. It's still fairly easy to generate a lot of negative dentries. Do you have an actual test program that does this and is annoying? Do people actually do that? It shouldn't be impossible to keep negative dentries in check without getting rid of them *entirely*. They do speed up lots of common operations (things like $PATH lookups etc, negative caches really are useful). Linus