From mboxrd@z Thu Jan 1 00:00:00 1970 From: Nicolas Pitre Subject: Re: Preferring shallower deltas on repack Date: Mon, 09 Jul 2007 15:13:54 -0400 (EDT) Message-ID: References: <20070709044326.GH4087@lavos.net> <7v1wfixhvk.fsf@assigned-by-dhcp.cox.net> <7vir8tv8dr.fsf@assigned-by-dhcp.cox.net> <20070709185353.GL4087@lavos.net> Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=us-ascii Content-Transfer-Encoding: 7BIT Cc: Junio C Hamano , git@vger.kernel.org To: Brian Downing X-From: git-owner@vger.kernel.org Mon Jul 09 21:14:04 2007 Return-path: Envelope-to: gcvg-git@gmane.org Received: from vger.kernel.org ([209.132.176.167]) by lo.gmane.org with esmtp (Exim 4.50) id 1I7ygK-0005WR-GH for gcvg-git@gmane.org; Mon, 09 Jul 2007 21:14:00 +0200 Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754234AbXGITN5 (ORCPT ); Mon, 9 Jul 2007 15:13:57 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1753749AbXGITN5 (ORCPT ); Mon, 9 Jul 2007 15:13:57 -0400 Received: from relais.videotron.ca ([24.201.245.36]:27143 "EHLO relais.videotron.ca" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752876AbXGITN4 (ORCPT ); Mon, 9 Jul 2007 15:13:56 -0400 Received: from xanadu.home ([74.56.106.175]) by VL-MH-MR001.ip.videotron.ca (Sun Java System Messaging Server 6.2-2.05 (built Apr 28 2005)) with ESMTP id <0JKX009EVER6QXW0@VL-MH-MR001.ip.videotron.ca> for git@vger.kernel.org; Mon, 09 Jul 2007 15:13:55 -0400 (EDT) In-reply-to: <20070709185353.GL4087@lavos.net> X-X-Sender: nico@xanadu.home Sender: git-owner@vger.kernel.org Precedence: bulk X-Mailing-List: git@vger.kernel.org Archived-At: On Mon, 9 Jul 2007, Brian Downing wrote: > On Mon, Jul 09, 2007 at 09:39:44AM -0700, Junio C Hamano wrote: > > > OK here it is. And results on the GIT repo and another patalogical test > > > repo I keep around are actually really nice! Not only the pack itself > > > is a bit smaller, but the delta depth distribution as shown by > > > git-verify-pack -v is much nicer with the bulk of deltas in the low > > > depth end of the spectrum and no more peak at the max depth level. > > > > Looks obviously correct. Brian, it would be very interesting to > > see what Nico's patch does to your dataset. > > Nico's patch makes the overall statistics look better, but the > version.lisp-expr file still goes 593 levels deep, as opposed to about > 65 with my patch. (That's better than 980 with stock Git, though.) Tuning for such an extreme without impacting normal cases is rather hard. My patch was meant to be used on top of yours. Is that what you tested? Also I'd suggest you do not use a max depth of 1000. It is simply insane and might possibly make the existing logic less effective. Even for runtime pack access you want it to be reasonably short, say 100 maximum, or even the current default of 50. Any improvements you might come with (like automatic depth determined on replay cost) should be compared with that default which is known to work well already. Nicolas