Article 41069 of comp.unix.solaris: Path: mri.com!newsfeed1.aimnet.com!nntp04.primenet.com!news.shkoo.com!nntp.primenet.com!news.mathworks.com!newsfeed.internetmci.com!news.internetMCI.com!news-admin From: Brad Glidewell Newsgroups: comp.unix.solaris Subject: Re: route default to itself Date: Fri, 02 Aug 1996 10:16:59 -0500 Organization: internetmci.com Lines: 57 Distribution: inet Message-ID: <32021BEB.960@atm.ns.mci.com> References: <3202076F.658D@jpmorgan.com> NNTP-Posting-Host: atm.ns.mci.com Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Mailer: Mozilla 3.0b5a (X11; U; SunOS 5.5.1 sun4u) Ling Wang wrote: > > This new env has their routing set up this way in /etc/init.d/inetinit > > route add default $IP 0 > > where $IP is its own IP address. > > This causes no route table to be kept except basically this one entry > (besides multicast and localhost) > > default $IP U 4 19709 le0 > > When I ping something a few hops away, I see the source machine > issuing an ARP of designation IP, and the designation host > issuing a RARP over the router, not of its MAC address, but > the routers. > This is called PROXY ARP. > I do not see this method mentioned anywhere. > > Is this a common way of doing routing? > Is the routers configured for passing broadcast ARPS/RARPS? Yes, this is not routing on the workstation, but ARP. The ROUTER is doing the routing. > What special config does the routers have to have in order > for this to happen. The router has to support PROXY ARP. > If there is multiple routers on the subnet, doesn't The routers use ICMP to resolve who has the best metric for each route. This router will respond to ARP for the destination IP address. They also can send an ICMP redirect message if needed. > both routers try to forward the ARP? how is that resolved? > > The only advantage of this over RIP seems to be less mbufs > usage and no route table overflow on very large networks. The biggest advantage is the removal of RIP overhead which can choke a sun in a large network. Other benifits include the reduced RIP updates and possible conflict with real routers. > > Any info would be appreciated. Unfortunately, this is poorly documented but is a very good way to set up a system especialy in a multihomed environment. I have used this method for years. You will notice that the ARP cache can get rather large if your system connects to many nodes such as a NMS will contact possible thousands of nodes. Some SUNOS needs a patch for the ARP table size limmit.