routed

[Vernon Schryver]

In article <9104232358.AA08192 at koko.pdi.com>, shoshana at pdi.UUCP (Shoshana Abrass) writes:
> 
>   I just noticed that we have routed running on all our SGI's. After
>   checking the  file /etc/init.d/network, I found that there was no
>   'chkconfig' option for routed - ie, SGI thinks it should run 
>   everywhere.

There is a 'chkconfig' option in 4.0, as well as 3.3.3.  If you buy FDDI,
you get it as well.  One can always edit the /etc/init.d/network
configuration file, tho that is best avoided.

>   My question is: why? Isn't routed something of a CPU hog on a
>   complex network? and isn't its functionality superseded by gateways
>   that run dynamic routing protocols?
> 
>   -shoshana
>   pdi!shoshana at sgi.com


No, routed is not expensive for a non-gateway.  The IRIX routed includes
some simple hacks that keep it from waking up more often than every 30
seconds on a simple host regardless of the number of active RIP
broadcasters on your network.

RIP and routed are as "dynamic" as a routing protocol might want to be.  If
anything, RIP is too "dynamic."  Many of the old routed problems have been
helped with recent 4.3BSD changes for split-horizon, delayed updates,
poisoning, and (I think) hold-down.

The troubles with the "functionality [of] gateways that run dynamic routing
protocols" include:
  1. how does a host discover them?  That is, "gateway discovery" is
	not a solved protocol, even if is trivial in concept.
  2. there is more than 1, 2, or even 3 flavors of such "dynamic functionality".
  3. they often do not understand RIP
  4. those that do understand RIP generate bogus garbage

Common solutions to these problems are:
  a. use a static default route, but that is a big hassle to keep accurate
      in a big network.
  b. broadcast a default RIP route, but that has troubles with 3 & 4 above.

People imply many bad things about RIP and routed.  I've begun to have
doubts about many of those people.  The network within Silicon Graphics has
grown to medium size over the years.  It now spans the Atlantic and the
Pacific.  Most of it uses RIP and routed, and most of the gateways among
the approximately active 80 networks are IRIS's.  The worst routing
problems are with the fancy boxes from other vendors.  (You might predict
that for several reasons.)

We ship a version of gated, which may be the sort of thing you are
referring to.  There are rumors that a future version of gated from Cornell
will support OSPF.  (The port of the Cornell source is easy and getting
easier.)  I don't know about IS-IS.  Policy based routing seems no more
than talk now.


Vernon Schryver,   vjs at sgi.com

P.S.  I hope I don't offend anyone by noting that I have found the words
  "functionality" and "technology" are often synonyms for male bovine
  excretia.  Have you recently been talking to a sales critter who told
  you its fancy router was perfect for your network?