From mboxrd@z Thu Jan 1 00:00:00 1970 X-Msuck: nntp://news.gmane.io/gmane.emacs.gnus.general/35704 Path: main.gmane.org!not-for-mail From: Karl Kleinpaste Newsgroups: gmane.emacs.gnus.general Subject: Re: Question about mail archive Date: 09 Apr 2001 20:55:31 -0400 Message-ID: References: NNTP-Posting-Host: coloc-standby.netfonds.no Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Trace: main.gmane.org 1035171404 4520 80.91.224.250 (21 Oct 2002 03:36:44 GMT) X-Complaints-To: usenet@main.gmane.org NNTP-Posting-Date: Mon, 21 Oct 2002 03:36:44 +0000 (UTC) Keywords: file,article,single,overviews,headers Return-Path: Original-Received: (qmail 14385 invoked by alias); 10 Apr 2001 00:55:26 -0000 Original-Received: (qmail 14380 invoked from network); 10 Apr 2001 00:55:25 -0000 Original-Received: from mesquite.slip.cs.cmu.edu (HELO cinnamon.vanillaknot.com) (128.2.207.11) by gnus.org with SMTP; 10 Apr 2001 00:55:25 -0000 Original-Received: (from karl@localhost) by cinnamon.vanillaknot.com (8.9.3/8.9.3) id UAA02620; Mon, 9 Apr 2001 20:55:31 -0400 Original-To: ding@gnus.org X-Face: "5(T0tZd{6}pd~YzBG8O/*EW,.]6]@`m^e;fv65W^Y&=d"M\1H}>T~4_.kcDD.O~y3k)a6 hR;Nmi>9|>Nm${2IpM0^RcUEa\jcq?KOP)C&~x51l~zCHTulL^_T|u0I^kB'z@]{`2YjQu In-Reply-To: (prj@po.cwru.edu's message of "09 Apr 2001 14:39:58 -0400") User-Agent: Gnus/5.090001 (Oort Gnus v0.01) XEmacs/21.2 (Urania) Original-Lines: 20 Xref: main.gmane.org gmane.emacs.gnus.general:35704 X-Report-Spam: http://spam.gmane.org/gmane.emacs.gnus.general:35704 prj@po.cwru.edu (Paul Jarc) writes: > The new version uses one file per message for NOV data I have to observe that the whole point of the development of NNTP's XOVER facility is because news servers were suffering so badly in the face of user agents entering a group and wanting header data for hundreds of articles at a time, requiring opening hundreds of article files in order to find those headers. System call tracing of NNTP reader servers revealed that a hefty percentage of their total time was being wasted by having the server open, read, and immediately close every single article file in the sequence, just to dig out a few headers, and that was all just for group entry, not including the user's actual article reading choices. Overviews were developed as a way of saving the most common, important headers in a single flat file, so that only a single open/read/close was needed, and some quite spiffy code was developed to make searching that flat file for a specified article range very easy and efficient. If you have resorted to per-message overviews, you have in effect utterly destroyed the very purpose for which overviews were created.