Date: Tue, 13 Aug 1996 09:58:20 +0900 (JST) From: Michael Hancock <michaelh@cet.co.jp> To: Terry Lambert <terry@lambert.org> Cc: freebsd-hackers@FreeBSD.ORG Subject: Re: Load-balancing box Message-ID: <Pine.SV4.3.93.960813093632.13759A-100000@parkplace.cet.co.jp> In-Reply-To: <199608121626.JAA25431@phaeton.artisoft.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 12 Aug 1996, Terry Lambert wrote: > Specifically, it can react when one of the machines goes down (but how > can it know?). It can assign an infinite cost to a server that doesn't respond. The operator can manually reset the server or there can be some kind of cost decay algorithm so that the redirector will try again later. > 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. Doesn't absolute duration only matter anyway? I don't think it matters that a faster server services more requests. > The fix is called "service anonymity" and requires a fundamental change > in the way addresses get resolved/routed, and the ability to actively > retarget session state (if any) from one server to another through > consensus of the servers in the cluster. Different costs can be assigned by ip/port pairs. I'm not sure what exactly the LocalDirector does as there's little information on their pages. I would imagine that the cost determination policy is where most of the experimentation would be needed. Mike Hancock
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.SV4.3.93.960813093632.13759A-100000>