[ih] Overlay networks (1980s SRI Reconstitution Protocol)

[John Day]

Inline below

> On Aug 22, 2025, at 03:01, Greg Skinner via Internet-history <internet-history at elists.isoc.org> wrote:
> 
> Forwarded for Barbara.  Also, I have a couple of comments.
> 
> There was a Sun workstation in the plane running the host RP implementation that also displayed some (rudimentary) network connectivity status.
> 
> In response to John Day’s comment about “partition” being an exercise in sloppy thinking, that use was consistent with its use in IENs 135 and 146, which are cited in the final report.  Did anyone (else) involved in the Internet Project feel that use was sloppy?

As I said in my previous email, a network partition is when network failures isolate two subnets of a network so that there can be NO communication between them. There is no connectivity between the two subnets.

Some background. We were the ARPANET node at Illinois. The jumping off point between the East and West Coast subnets. In the early days of the ARPANET, we saw several network partitions when both cross-country links were down. The East and West Coast hosts couldn’t communicate because there was no connectivity. The introduction of a third cross-country link seemed to solve the problem. However, that led us in the mid-70s to do considerable research on the issues of network partitions and the fact that from inside the network, it was impossible to distinguish a number of hosts being down from a partition, and the issues of reconciling differences in databases once a partition was repaired.

It is clear from reading the SRI documents and from discussions with people that participated in it, that this was a case of connectivity still existing in the network. (How else could any protocol solve it?) These weren’t network partitions, but cases of a link being down. If connectivity still existed as the document show, then this was a matter of simply running the routing algorithm again. This was not in any means a special case or a network partition.

If the problem they were solving was something else, they should have said so.

So do I consider this sloppy thinking.  Yes, this was clearly sloppy thinking . . . or worse.

Take care,
John Day
> 
> Greg
> 
>> On Thursday, August 21, 2025 at 10:06:02 PM PDT, Barbara Denny <b_a_denny at yahoo.com> wrote:
>> 
>> 
>> Having trouble reaching the mailing list....
>> 
>> Within a packet radio network, the mobility of the nodes could cause a network to partition and then eventually coalesce (repair). Jammers were also considered a threat besides capture or failure of nodes.  In the SAC experiments the airplane functioned as a repeater and I think there was a host (sun workstation) in the plane too. We did plan for the plane to fly a specific path to create the topologies we wanted. Banking of the airplane did cause some short term unexpected outages but these weren't long enough for the system to invoke the components to handle a partition.
>> 
>> The diagrams in the report show we did have multiple entry/exit points (RP gateways) in the networks to allow us to create a path to a destination through the remaining pieces and networks if it existed.
>> 
>> The name Reconstitution Protocol might be confusing some of you. There was more than a single protocol in the approach.
>> 
>> barbara
>> 
>> On Thursday, August 21, 2025 at 01:48:59 PM PDT, Jack Haverty via Internet-history <internet-history at elists.isoc.org> wrote:
>> 
>> 
>> The ICCB discussed the "reconstitution" need back in the early 1980s.  
>> The concept was that the Internet could be used to re-establish 
>> host-host connectivity when an underlying network, e.g., the ARPANET, 
>> was partitioned.  This was thinking beyond research to operational 
>> military usage, when the underlying network, i.e., the Defense Data 
>> Network, might be partitioned due to enemy activity.
>> 
>> The mechanism would require multiple paths through the Internet. That 
>> situation existed at the time in the research installations. For 
>> example, the WBNET provided connectivity between the US east and west 
>> coasts, and could serve as a backup to the ARPANET connectivity.   
>> Similarly, SATNET provided connectivity to Europe, and the X.25 public 
>> network provided a second path.
>> 
>> I don't know any details of what was actually tested but the notion was 
>> that the relevant gateways could notice that one of their underlying 
>> networks (e.g., the ARPANET) had become partitioned. They could detect 
>> that situation by their sudden inability to communicate with other 
>> gateways on a particular underlying network.
>> 
>> A similar procedure had been used internally within the ARPANET, to 
>> detect and route around circuit failures.  Gateways talked to each other 
>> much as IMPs did, to detect failures and route around them. But they 
>> assumed at that time that networks were either working or not working.  
>> In the ARPANET, circuits never "partitioned", and the gateway system had 
>> used the same assumptions about its underlying "circuits", i.e., the 
>> underlying networks.  "Reconstitution" was possible in the Internet, 
>> where it wasn't in the ARPANET.
>> 
>> The gateways could then exchange routing information, using the paths 
>> between gateways that still were functional, to begin to treat the 
>> broken ARPANET as two separate networks.   Routing changes would then 
>> direct Internet traffic as needed to re-establish communications between 
>> host pairs on the broken network(s).
>> 
>> This of course required hosts to be using TCP/IP, and also a robust 
>> connectivity within the Internet, so that there were alternate paths to 
>> be used.
>> 
>> The underlying "broken" network was not itself patched back together.  
>> But assuming the individual pieces of that network continued 
>> functioning, traffic through the Internet could be resumed, passing 
>> around the breaks.  The ARPANET pieces would continue to work even when 
>> partitioned.
>> 
>> I recall an incident in the early 1980s when an operator from the 
>> ARPANET NOC stuck his head in my office and frantically asked if the 
>> Internet could be used to get to some west-coast computer from our 
>> east-coast location, without using the ARPANET.  Couldn't happen - there 
>> was no such "reconstitution" mechanism in place at the time. Curious, I 
>> wandered down to the NOC to see what was going on.  Some combination of 
>> buggy software and possibly an errant backhoe had partitioned the 
>> ARPANET, and the operators were trying to figure out how to put it back 
>> together.  Turned out bugs and backhoes had similar effects to bombs and 
>> missilies.
>> 
>> "Reconstitution" was about re-establishing communications through the Internet,  not fixing broken underlying networks.
>> 
>> /Jack Haverty
>> 
>> 
>> 
> 
> -- 
> Internet-history mailing list
> Internet-history at elists.isoc.org
> https://elists.isoc.org/mailman/listinfo/internet-history
> -
> Unsubscribe: https://app.smartsheet.com/b/form/9b6ef0621638436ab0a9b23cb0668b0b?The%20list%20to%20be%20unsubscribed%20from=Internet-history