3 ms·
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. O
by inopinatus 3y ago
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.
- retrocryptid 3y agoCan 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.