The original question was:
> SunOS 4.1 comes with dynamic routing enabled (in.routed is uncommented in
> rc.local), yet I have read in a couple of articles from Sun that for most
> desktop machines, dynamic routing is inefficient. Commenting out the
> route deamon in your rc.local and adding a default route would alleviate
> this problem, but would this introduce any new problems into our network?
> Does anybody use static routing, and if so, what type of problems have you
> come across? Does anybody have any statistics as too how ineffient dynamic
> routing is compared to static routing?
Summary:
I received over one hundred responses, all were appreciated. I have
selected a few to use as my summary.
From Katy Kislitzin, ktk@nas.nasa.gov -
Status: R
> One reason dynamic routing is inefficient is because it requires a
> seperate process to run. Which will use resources which might be
> better used by more interesting processes... Also, routed uses RIP,
> which is a broadcast based protocol.
>
> There are 2 issuses in deciding whether you want dynamic or static
> routing:
>
> 1. are there multiple paths off of your network/subnet? If so, you probably
> want your machines to silently choose an alternate path if one should
> fail. If there there is only one way out, you don't lose anything by
> having static routes.
>
> 2. do your machines move around within your network a lot? If you
> have machines which move from subnet to subnet and therefore need
> different static routes, you may want to use dynamic routing even if
> every subnet only has one choice to get out. The reason here would be
> administrative simplicty.
>
> If you have multiple ways out, another thing to consider is what you
> want to have happen when one path goes down. If you want it to be
> (relatively) transparent, use dynamic routing. If you want to know
> about it and change all the machines manually, use static routing.
>
> BTW, if you decide to use dynamic routing, you may want to look into
> using gated in lieu of routed. That's what we use, and it is supposed
> to be superior, although I confess I cannot tell you why.
From sam@john-bigboote.ICS.UCI.EDU -
Status: R
> I use the following rule. Use a default static route when:
>
> 1. there's only one gateway off of the net
>
> - or -
>
> 2. your machine is completely dependent on some other machine (like your
> server) and that machine is a gateway off of your network.
>
> Rule 1 is obvious. If there's only one way to get off of your network
> you'll always have to use it and if it goes down, running routed running won't
> help.
>
> Rule 2. If you're already completely dependent on some machine (like a file
> server) then if you always route through it and it goes down you won't care
> anyways because you won't be functional while it's down. This one isn't
> quite as hard and fast as rule 1 because if there's a faster gateway off of
> your network (like a 4/490 or some fast gateway box) you may want to
> use that box most of the time and only fall back on your server in
> emergencies.
From Tim Becker, becker@cs.rochester.edu -
Status: R
> We had used routed for awhile and then switched to a static default
> route pointer to our router (that eventually leads to the internet).
> I don't know how your internet access works, but I really doubt if you
> want all the RIP routes coming in from your router going to all your
> workstations. The number of RIP routes must be well over 1000 by now
> (it was about 600 2 years ago, when we stopped using routed). With
> that number of routes and the fact that RIP updates happen every 30
> sec, your machines will take a short dive every 30 sec. I think
> you'll notice it. In addition RIP is sent as broadcast packets, so
> even if you don't want to run routed, your workstation has to "look
> at" the packets anyway.
>
> If you have only one gateway out to the internet (or other networks
> that you care about), I can think of no significant disadvantage to
> using a static default route pointing to that gateway.
>
> If you have more than 1 gateway, you might want to use dynamic routes.
> The way I'd set this up would be not have the gateways RIP only
> "default" and not all the networks it knows about. I'd make one of
> the send default with a metric one higher than the other. Then if the
> first goes down, all your machines will automatically fall back to
> using the 2nd gateway. (I don't think there is an automagic way to do
> load balancing between the 2 gateways -- other than having half the
> workstations pointing to 1 and the other half pointing to the other
> gateway).
Thanks again,
Rich Eckert
Battelle Northwest Labs
d3e505@pnlg.pnl.gov
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:10 CDT