[comp.protocols.tcp-ip] Network Connectivity - Part 2

AI.CLIVE@MCC.COM (Clive Dawson) (11/04/88)

After sending my previous message in which I wondered what would
cause only some hosts on net 10 to be able to communicate with
a certain other hosts, I began doing some systematic testing
to try and determine what the connectivity was really like.

In the process, I discovered the answer to my own question.  It
turns out that all the prime gateways which my TOPS-20 system
knew about were down!  Some years ago, the NIC assigned certain
MILNET gateways to each Arpanet host.  Up until yesterday,
these gateways have always been sufficient to provide all the
redirect info required.  But since all the Milnet gateways apparently
went down because of this virus, my system had nobody left to
ask.

I've added a couple of other non-Milnet prime gateways to
my INTERNET.GATEWAYS file, and all is well.  I suspect there
may be other TOPS-20 systems out there with the same problem,
since the practice of having only one or two entries in
the gateways file was pretty widespread for a while.

Clive
-------

fserr@ALEXANDER.BBN.COM (Frederick E Serr BBNCC 20/666 x2474) (11/05/88)

I would urge TOPS-20 administrators NOT to add more than one or two
other gateways to their gateways file, if any.  The MILNET Mailbridge
Gateways were explicitly turned off yesterday to slow the spread of 
the virus.  As you yourself said, their normal reliability is such
that one primary gateway and one backup entry was enough to keep
your host "connected" for several years without running into this
problem.  Presumably, turning off the Mailbridges because of external
circumstances will continue to be a rare event.

The reason for only having one or two entries in the file in the first
place is to reduce the amount of ICMP overhead traffic on Arpanet and 
Milnet.  TOPS-20s periodically "ping" all the gateways on their list.
A few years back, this was adding up to a substantial fraction of all
the traffic on the Arpanet.  While one can argue that adding one more
gateway to one host will result in only a small amount of traffic, it
still seems worth avoiding, when the problem one is fixing is so rare.

Fred Serr
Network Analysis Dept.
BBN Communications

WANCHO@WSMR-SIMTEL20.ARMY.MIL ("Frank J. Wancho") (11/05/88)

Clive,

I discovered the same problem last night caused by the fact that the
prime gateways were down (either deliberately taken down, looped off
the net or powered down by the NOC) in an effort to compartmentalize
the spread of the worm (not virus).  It is not just the TOPS20 hosts
which depend on these gateways for ICMP REDIRECTs - all hosts using
the prime gateways do.  Once I added a couple of other gateways,
everything started flowing again everywhere except across the MIL-ARPA
gateways (as I believe was the intent in shutting off those gateways).

The lesson to be learned from this is that we should have at least one
other non-MIL-ARPA prime gateway in the list of assigned gateways
should it become necessary to isolate those two nets from each other
in the future.

--Frank

SRA@XX.LCS.MIT.EDU (Rob Austein) (11/05/88)

Clive,

Right.  Here's the rest of the story.  XX and Score are configured
differently than MCC or NIC, because while all four are multi-homed
hosts, XX and Score are not the sole contact points between the nets
to which they are connected.  Eg, XX is connected to nets 10 and 18,
but so is GW.LCS.MIT.EDU.  A multi-homed TOPS-20 in this configuration
needs more entries in its INTERNET.GATEWAYS file than just its IMP's
two assigned "mailbridge" gateways, eg, XX's also lists GW,
KLUDGE.AI.MIT.EDU, SLUDGE.LCS.MIT.EDU, and SEWAGE.MIT.EDU.  Yesterday
XX ended up routing all of its "default" traffic via GW.LCS.MIT.EDU
after it gave up on IMP44's assigned gateways.  I assume something
similar happened to Score.

Is the "eager pinger" problem believed to be fixed?  If I remember
correctly, that's where the practice of having only two live core
gateways known per host came from.  The NIC has a file online (in
NETINFO:, I think) explaining this, but it was never clear to me if
the underlying problem was really fixed or simply toned down to an
acceptable level if everybody stuck to the two gateway rule.

--Rob
-------