[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: TCP/TCP Gateways
Subject: Re: TCP/TCP Gateways
Organization: University of Colorado, Boulder
Cc: Slug.R20.Utexas.EDU < OOPS!
In article <3404@sigi.Colorado.EDU> you write:
>I have run this problem by the German Lispm users group, and the
>We have a number of Symbolics 3670's running 7.1 with IP-TCP version
>52.16. These Symbolics hang on an Ethernet, which is interfaced
>through several unix gateways to other Ethernets, which are in turn
>interfaced to others, etc. etc. TCP runs on all of these.
>Problem: Even though I believe I have set up the namespaces right, I
>cannot get the Symbolics to route to machines not directly on the
>Ethernet to which it is connected. I can try to establish a
>connection from the other direction, and this gets routed to the
>Symbolics, but since the Symbolics doesn't somehow find the host
>acceptable, the connection gets refused.
what class addresses? are you using subnet masks?
>How I set it up:
>1. There is a separate (namespace,site,network) triple for each
> individual physical
> It didn't work any better when all
> the hosts were in one namespace, the way I set it up initially.]
We did not find this necesary.
>2. Each gateway host has an entry for both networks, and the services
> - gateway ip internet-gateway
> - gateway ip internet-gateway-prime
The first looks fine, I am ignorant about the second
>3. Hosts on the same network can establish connections with each other with
> no problem, so they have all the end user right services.
>Yet when I try to connect to a host on the other network, I get an
>error message stating that the connection service I am requesting is
>not supported by the remote host, followed by a lists of the services
>supported by the local and remote hosts (both of which contain the
>Has anyone succeeded in getting a non--trivial network configuration
>to work with the Symbolics software?
Ours is working fine here at CU
>What's the secret? Tracing through the code, it seems that a new
>network protocol is needed, (when I substitute the TCP/Chaos protocol
>in, it does the routing ok, and then crashes because the streams don't
>have the right properties), however certain TCP hackers (who only know
>Unix) around here tell me this is nonsense --- that routing takes
>place in the IP layer, and I should never see it.
I think that there is also a TCP-level gateway, but I'm pretty sure
it's only on ()'s.
We have a multi-subnetted network here at CU, and our ()'s are on
a physical subnet along with some suns, and we gateway onto other
parts of out network (and the rest of the internet) by indicating that
our sun gateway offers INTERNET_GATEWAY service. We have only one
namespace, and we have no problems. I set this up over a year ago, so
some parts of it are fuzzy by now, but I would be glad to answer any
specific questions. We are also using subnet masks now, but prior to
Gen 7, we were using class C addresses and didn't need to do subnet
addressing since we gave a separate network number to each physical
For a while after we switched over to Class B and subnet masking, we
had some problem with reaching some machines across gateways, because
they didn't understand subnet masks, and didn't know to route through
the gateways. As it turns out, when we thought we couldn't contact
them, we were actually getting TO them but they couldn't get any
acknowledgment back to us. This doesn't seem to match what you
described above, though.
I would beat on your unix gateways a little bit, too, to make sure
that their routing tables are set up right. (are there non-()'s on
your physical net which can get across the gateways?) The fact that
you can get to the ()'s from the other net but not back seems to rule
this out, but...
hang in there, cause it does work. I'm at a loss with what you've
said so far? Anyone else see something obvious?