From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ted Ts'o Subject: Re: [PATCH] ext4: remove deprecated oldalloc Date: Sat, 8 Oct 2011 14:09:05 -0400 Message-ID: <20111008180905.GG7948@thunk.org> References: <1307453714-2741-1-git-send-email-lczerner@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Lukas Czerner , linux-ext4@vger.kernel.org To: Andreas Dilger Return-path: Received: from li9-11.members.linode.com ([67.18.176.11]:58473 "EHLO test.thunk.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753423Ab1JHSJK (ORCPT ); Sat, 8 Oct 2011 14:09:10 -0400 Content-Disposition: inline In-Reply-To: Sender: linux-ext4-owner@vger.kernel.org List-ID: On Thu, Aug 11, 2011 at 03:05:55PM -0600, Andreas Dilger wrote: > > That said, "orlov" is also mostly meaningless with flex_bg as well, since > there is very little real benefit/affinity from inodes being "close" to > their data blocks. We gain far more benefit from keeping the inodes > together than spreading them out and keeping them close to the data blocks. I'm going to be applying the patch remove the oldalloc allocator, but just for the record --- Orlov *does* matter (which is why it's a good thing the default allocator also uses an Orlov-like approach that's flex_bg aware). In fact, we saw significant latency drops that grew over time as more and more blocks got allocated. I traced it back to using bigalloc, which increased the average distance between the inode tables and the data blocks. So as you use larger bigalloc cluster sizes, which increases the block group size, it's a good idea to decrease the flex_bg size. Which is another way of saying that Orlov matters. - Ted