Hacker News Reader

Comment by Earw0rm | Hacker News Reader

I'm talking about server to server message propagation, not client/server - agreed, the Usenet clients were much faster and had better feature-sets than any web forum running on Netscape or the early versions of IE. That part was great, but messages could take hours to even days to propagate across the whole network. The sync-reply-upload work flow was indeed useful - especially so in countries where you had to pay by the minute to be online.

View on HN

Propagation time was never the problem. I ran a major European node in the 90s and our propagation times were in the seconds. We absolutely obsessed about it. Only dialup leaf nodes took longer, but that was and remains a feature, not a bug.

I've been using NNTP as a Gossip-type protocol for eventually-consistent pub/sub telemetry, monitoring, and notifications between partially-connected nodes in an unstable mesh ever since. Just not in public.

Let’s be careful to separate the message protocol from the system it supported. Usenet’s carcinogenic inability to self-regulate at scale arose from the absence of distributed authority and negative feedback loops in the control plane, something we’d aim to build in algebraically and cryptographically if designed today. Alas, VC backing directs most research in this subsector towards systems enabling consumer surveillance.

Can confirm. I ran a small node for a reasonably sized university in the early 90s. Replication times were in the tens of seconds. Could be in the minutes if a perfect storm of congestion and server overloading occurred. But my memory was more than 99% of the time latency was pretty low (seconds to tens of seconds.) Not bad for a single VMS cluster that serviced 20k+ clients.

That reminds me of the days of FidoNet [0] and Barren Realms Elite [1] where the inter-BBS propagation delay made your BBS's attacks on another BBS even more tense and exciting, and well, it felt realistic that sending your forces to another "planet" took time to happen and for results to be reported.

[0]: https://en.wikipedia.org/wiki/FidoNet

[1]: https://www.johndaileysoftware.com/products/bbsdoors/barrenr...

Woah, I remember playing BRE, which must have been close to 30 years ago. Impressed the author still has a site for it much less that you can purchase it!

I'm going to call "mostly wrong" on the server-to-server propagation.

By 2000 the NNTP mesh was very interconnected and fast. Sites competed[1] to get messages propagated fast. Especially text messages.

Any site hooked into the mesh was going to be offered any article multiple times within seconds of it being posted.

Slow articles would happen but only during outages or with sites that were not following good practice.

[1] Via http://top1000.anthologeek.net/

Absolutely - the federated case (I was rather active on the tolkien, c64 and various computing groups) was terrible. We used to use email lists for real discussion with plain text messages (shame that isn't a thing anymore).

It was an excellent experience in the non-federated case.

Reply on news.ycombinator.com