From owner-freebsd-isp Wed Jan 8 02:17:07 1997 Return-Path: Received: (from root@localhost) by freefall.freebsd.org (8.8.4/8.8.4) id CAA25147 for isp-outgoing; Wed, 8 Jan 1997 02:17:07 -0800 (PST) Received: from labinfo.iet.unipi.it (labinfo.iet.unipi.it [131.114.9.5]) by freefall.freebsd.org (8.8.4/8.8.4) with SMTP id CAA25139 for ; Wed, 8 Jan 1997 02:17:02 -0800 (PST) Received: from localhost (luigi@localhost) by labinfo.iet.unipi.it (8.6.5/8.6.5) id KAA09365; Wed, 8 Jan 1997 10:33:01 +0100 From: Luigi Rizzo Message-Id: <199701080933.KAA09365@labinfo.iet.unipi.it> Subject: Re: dedicating bandwidth? To: bminazzi@w3page.com (Blaine Minazzi) Date: Wed, 8 Jan 1997 10:33:01 +0100 (MET) Cc: isp@freeBSD.org In-Reply-To: <32D35748.5E407947@w3page.com> from "Blaine Minazzi" at Jan 8, 97 01:13:41 am X-Mailer: ELM [version 2.4 PL23] Content-Type: text Sender: owner-isp@freeBSD.org X-Loop: FreeBSD.org Precedence: bulk > > > anyway, i would like to dedicate bandwidth between my machine, and the > > > ethernet machine(s). meaning that the ethernet can not take the full > > > ppp bandwidth and also that my machine can not take it all so that there's > > > something left for the ethernet users... > ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > > I know this is not an answer, but I have to ask. > Do you allready have a problem with bandwidth usage? > > Many times I see people throwing Time/Effort/Money at something that > they percieve _may_, _could_be_ , _If the moon is full_ , and I hold my > tounge just right, problem. Or, they are not familiar enough with what > the operating system is already designed to do. Good point. In this specific case, though, it is true that it is easy to saturate a PPP link pumping data from an Ethernet. The TCP congestion avoidance policies do not help, since the window for each TCP connection fills up until the max size (e.g. 16KB), say you have 4 parallel connections (netscape ?), then all these packets pile up in your send queue easily amounting to delays of several seconds (up to 10 or more). This is just something the current IP implementation is not able (and designed to) deal with properly. At some point (2.1R) our TCP even had a nasty feature which completely disabled slow start on a 'local' (intended as a class A,B,C) network. Having a small amount of reserved bandwidth helps a lot in this case (or you can assign priorities depending on type of traffic, such as some version of iijppp is supposed to do, and some actually do). This is not to criticize FreeBSD, it's just that all systems have bugs or missing features and ours is no exception (luckily, otherwise there would be no need for people hacking it :). Luigi