Summary

  • RFC 1165 proposed NTP over the OSI Remote Operations Service in June 1990. Its connection-oriented model used association state—connected, connecting, accepting or disconnected—to describe peer reachability.
  • The existing packet-based NTP reachability register tracked replies across eight transfers. An established or broken connection gave the experiment a more immediate reachability signal; timing measurements and the NTP data fields continued to do the clock work.

A clock peer has two questions to answer

One question is whether another machine can be reached. Another is what its clock says, with what delay and uncertainty. Packet-based NTP had to infer the first answer from recent exchanges. RFC 1165 describes a shift register that records responses over the last eight transfers; after eight unanswered packets, the peer is deemed unreachable. It is a compact attendance record. A missed response pushes another mark into the row, and the row gradually stops giving the peer the benefit of the doubt.

Crowcroft and Onions asked what would happen if NTP travelled through an OSI connection rather than the familiar UDP service. Their experimental memo defined Remote Operations and an ASN.1 abstract syntax for that setting. The reasons ranged from offering time synchronization to OSI applications to testing whether the specification could be checked and code generated automatically. Connection-oriented services such as X.25 were also a laboratory for a sharper reachability signal: the connection was established, broken or still being established. The eight-exchange register no longer had to stand in for that particular fact.

The proposal gave the peer a small set of explicit states. Connected meant synchronization data was being exchanged. Connecting meant a connection attempt was under way. Accepting meant an incoming connection had arrived while transport and session negotiation still had work to do. Disconnected meant contact had failed. Setup was asynchronous because a distant connection could take seconds; holding the time daemon still while it waited would disturb exchanges with other peers. The clock did not need a dramatic new face. The door to the other clock needed a state of its own.

An association began with an A-ASSOCIATE.REQUEST and an indication from the remote side. After that, peers periodically invoked the relevant remote operation with NTP parameters. Information unlikely to change during the lifetime of a connection—protocol version or local clock precision, for example—could be learned at setup and omitted from each subsequent operation. RFC 1165 also considered doing authentication at connection establishment if that mechanism were in use. The NTP data fields themselves followed the earlier specification, which gave implementers a path to retain familiar internal records while translating them to ROS/ASN.1 encodings.

The new reachability evidence had a precise limit. A working connection says the peer can be contacted through that service at that moment. The quality of synchronized time still rests on the exchanges and their transit behavior. RFC 1165 made a pointed observation: variation in packet travel time matters more to clock synchronization than raw network speed alone. A slower path with steadier transit might serve time better than a faster path with wild delay swings. The authors called that thought “heresy.” The joke had a target: the reflex to equate a faster network with a better clock.

There was even a working bridge between worlds. RFC 1165 reported several UK sites using NTP over ROS and X.25 with an earlier version of the ROS specification. At least one site peered that way with UK sites and also used UDP to peer with US Internet sites. The memo described the initial experience as promising and its comparative testing as limited. Its useful historical evidence is the mixed arrangement itself: a time daemon could speak through both transports while keeping its NTP state recognizable. The experiment gave no need to declare a winner between network architectures.

For operators, the important change was where a decision came from. In a packet setting, eight recent responses supplied a history of possible reachability. In the connected setting, the transport association supplied present connection state and could carry setup information that later operations did not repeat. The same clock algorithm could then spend its attention on the measurements that actually express time. RFC 1165 is a small example of a large engineering habit: when a lower layer already knows a fact, let it report that fact accurately and stop asking an upper layer to infer it by watching the mail arrive.