[ih] Overlay networks (1980s SRI Reconstitution Protocol)
John Day
jeanjour at comcast.net
Sat Aug 23 17:24:43 PDT 2025
It seems a stretch to call that recovering a partition. Although, the fancy name sounds more important.
I presume what you have in mind is that there alternative networks that can be brought to bear, but they aren’t readily available for routing to use. Just sounds like normal failure recovery.
Do you assume this is what the authors of the CAP Theorem had in mind? Because for me, it is fundamentally flawed.
But if it makes you happy, that is what is important.
Take care,
John
> On Aug 23, 2025, at 18:23, Tony Li <tony.li at tony.li> wrote:
>
>
>
>> On Aug 21, 2025, at 12:38 PM, John Day via Internet-history <internet-history at elists.isoc.org> wrote:
>>
>> This topic has come up several times here. Could someone please explain it to me.
>>
>> My understanding is that this was a protocol to repair network partitions.
>> If there is a network partition, there are two issues:
>> 1) it can’t be known that it is a partition until it is over. From within the network, a network partition is indistinguishable from a number of hosts being down for some other reason.
>> 2) If it is a network partition, there is no communication between the partitions so how can there be a protocol to repair it?
>
>
> The cases of practical interest are ones where one level of topology is partitioned and can be ‘healed’ by tunneling through a higher level network.
>
> In particular, in the day, CAnet was prone to partition due to insuffcient redundancy and budgetary constraints. The thought was to automatically restore connectivity by tunneling across NSFnet.
>
> Similarly, if an OSPF or IS-IS area partitioned, then tunneling between the two partitions over the backbone or L2 topology would restore connectivity.
>
> These all worked through interactions with the higher level, which addresses both of your points.
>
> The complexities of these proposals was daunting, at the time, and impetus for them died when Real Money started to show up at the ISPs.
>
> Regards,
> Tony
>
More information about the Internet-history
mailing list