🚀 go-pugleaf

Path: gamma!wb
From: wb@gamma.UUCP (Bill Beblo)
Newsgroups: mod.computers.apollo
Subject: Re: TCP/IP Problems
Summary: We have experienced the same problems. Apollo is aware of what
	 is happening, but is unable to guarantee a fix for SR9.5
Message-ID: <371@gamma.UUCP>
Date: 3 Apr 87 20:40:42 GMT
References: <630*northover@cascade.carleton.cdn>
Organization: Bellcore, Livingston, NJ
Lines: 50

In article <630*northover@cascade.carleton.cdn>, northover@cascade.carleton.CDN (Stephen Northover) writes:
> 
>     Hi.  I have just installed TCP/IP on our Apollo ring and have come across
> the following bizzare behavior.  From the gateway node, it is possible to access
> external sites.  From any node on the ring, it is possible to access any other
> node on the ring.  From a non-gateway node, it is NOT possible to access any external
> sites.  The error message indicates a timeout.
> 
>     Has anybody else seen this?
> 
> Thanks in advance,
>     Steve Northover
> <northover@cascade.carleton.cdn>

I am running a ring of 24 Apollo nodes with a COM-ETH gateway on a DSP160.
After installing TCP/IP version 2.1, I noticed that we'd be unable to make
connections via the gateway after a few days.  The network to which the Apollo
ring was gatewayed was connected via point-to-point link (using local
serial line IP code) to another network.  Thinking the local code might
be playing a factor, I lived with reinitializing TCP/IP on the gateway
every few days since this link was going to disappear soon.

We replaced the link with a Vitalink Translan bridge to a backbone network.
Immediately, our MTTF on the Apollo gateway dropped to 5 minutes.
Assuming a great increase in the number of broadcast packets received
with the bridge replacing the point-to-point link and
seeing a huge (>3000) number of rwho packets in the Rcv-Q when the
gateway crashed, I decided not to run the rwhod on the gateway.
MTTF increased to a few hours.  At this point we called in Apollo.

We discovered that the routed on the Apollo gateway was crashing.
Although it would continue to receive routing packets from the network
and maintains it's internal routing table, Apollo routed would cease
to send out it's routing information.  It was failing with some sort
of an I/O error which Apollo reps clearly saw.  Although the Apollo gateway
could get to the requested node, the node could not get back to Apollo,
hence the connection time out.

We are currently running with static routing on the Apollo and
static additions for the Apollo gateway on our non-Apollo nodes.
Apollo cannot guarantee that this problem would be resolved in
TCP/IP 3.0 which will be released in conjunction with SR9.5.
As we will be installing SR9.5 soon after it arrives, I will
post an update soon after the installation.

Bill Beblo
Bell Communications Research
290 West Mt. Pleasant Ave.,  Rm 1D-148
Livingston, New Jersey 07039
(201) 740-4421