From owner-freebsd-hackers Wed Mar 29 13:24:36 1995 Return-Path: hackers-owner Received: (from majordom@localhost) by freefall.cdrom.com (8.6.10/8.6.6) id NAA29536 for hackers-outgoing; Wed, 29 Mar 1995 13:24:36 -0800 Received: from hutcs.cs.hut.fi (root@hutcs.cs.hut.fi [130.233.192.2]) by freefall.cdrom.com (8.6.10/8.6.6) with SMTP id NAA29528 for ; Wed, 29 Mar 1995 13:24:30 -0800 Received: from shadows.cs.hut.fi by hutcs.cs.hut.fi with SMTP id AA03501 (5.65c8/HUTCS-S 1.4 for ); Thu, 30 Mar 1995 00:23:51 +0300 From: Heikki Suonsivu Received: (hsu@localhost) by shadows.cs.hut.fi (8.6.10/8.6.10) id AAA19947; Thu, 30 Mar 1995 00:23:56 +0300 Date: Thu, 30 Mar 1995 00:23:56 +0300 Message-Id: <199503292123.AAA19947@shadows.cs.hut.fi> To: Mark Hittinger Cc: freebsd-hackers@freefall.cdrom.com In-Reply-To: Mark Hittinger's message of 29 Mar 1995 18:02:56 +0300 Subject: Re: Mail... (fwd) Organization: Helsinki University of Technology, Otaniemi, Finland Sender: hackers-owner@FreeBSD.org Precedence: bulk Each FreeBSD machine might have such a limit. The idea here would be to build a "cluster" design to handle your growth by adding machines. Do you really want more than 256 people on one box simultaneously? No, but on average each user consumes 3-4 pty's for screen, xterms or things like that. That's the average on our systems (we run out of 64 with less than 15 users). We currently utilize something like a cluster (users can log into 3 machines). The servers aren't similar. Both 1.1.5.1 ("stable" servers and routers) and 2-current ("unstable" servers, new routers), in addition to Linux (terminal servers) and older junk we have gathered during our existence (old suns). This is a mess, really, but seems to be unavoidable these days (we probably have to add NetBSD, LITES and Hurd next :-). Using a terminal server front end will allow you to switch users based on incoming protocol type, load, or software version. (you also don't have to futz with serial drivers or burden your box with serial interrupts). 386-33 with 16 V34's connected with a cyclades card doesn't seem to cause any visible load (load average stays under 15%). Haven't tried to server PPP yet, but will do so soon. This is with Linux, 8M, dedicated to be a terminal server. One philosophy of design that you definetly do not want to use is the "one giant box does it all" plan. This is an old mainframe plan and every plan has a flaw! In the 80's the centralized versus distributed debate was fought and the centralized guys lost. We have been trying to implement this, but we would like to get rid of all different dedicated hardware, just have free unix systems on all jobs, including terminal servers and routing. If we need more power, the PC shop next door will be happy to provide us with a couple more pc's. Not yet there, unfortunately. -- Heikki Suonsivu, T{ysikuu 10 C 83/02210 Espoo/FINLAND, hsu@cs.hut.fi home +358-0-8031121 work -4513377 fax -4555276 riippu SN