From owner-freebsd-hackers Mon Aug 12 14:01:56 1996 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id OAA20754 for hackers-outgoing; Mon, 12 Aug 1996 14:01:56 -0700 (PDT) Received: from phaeton.artisoft.com (phaeton.Artisoft.COM [198.17.250.211]) by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id OAA20737 for ; Mon, 12 Aug 1996 14:01:50 -0700 (PDT) Received: (from terry@localhost) by phaeton.artisoft.com (8.6.11/8.6.9) id NAA26078; Mon, 12 Aug 1996 13:56:00 -0700 From: Terry Lambert Message-Id: <199608122056.NAA26078@phaeton.artisoft.com> Subject: Re: Load-balancing box To: luigi@labinfo.iet.unipi.it (Luigi Rizzo) Date: Mon, 12 Aug 1996 13:56:00 -0700 (MST) Cc: terry@lambert.org, michaelh@cet.co.jp, freebsd-hackers@freebsd.org In-Reply-To: <199608121804.UAA07814@labinfo.iet.unipi.it> from "Luigi Rizzo" at Aug 12, 96 08:04:37 pm X-Mailer: ELM [version 2.4 PL24] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-hackers@freebsd.org X-Loop: FreeBSD.org Precedence: bulk > > The whole problem is that you connect to an address instead of a service > > in the first place -- this type of load balancing assumes all clients > > have statistically identical duration, and doesn't truly balance the > > load between the boxes. For example, if I have 3 boxes, and 300 incoming > > connections are made, and 200 of them leave, one box can have 100 > > connections and the other two can be idle. > > Statistically this is _very_ unlikely. You will not have perfect > balancing, but not this kind of unbalancing. Depends. DNS information can be caches (and generally is) despite setting a short expiration (ie: your suggested expiration is overridden). If you live near a terminal branch in the Internet, you are likely to have the remainder of your branch assigned one address and the rest of the Internet assigned another. This is only worse now that Internic is fighting to get address scoping set up so that ISP address ranges can be homogeneously routed without route flopping. Sprint is pretty much the only holdout still having serious route floppage. I would be interested in a monitoring system that would report connections per cluster member -- but, of course, if you had that, you would be able to implement load balancing a better way anyway. 8-(. Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.