Summary

  • RFC 1574, published in February 1994, specified a minimum OSI/CLNP debugging set: echo, a traceroute using packet Lifetime, and a routing-table dump for hosts and routers providing OSI service.
  • A node that understood CLNP but not the new echo selector could return an error when the request set the ERROR report flag. That error located a responding hop; it did not claim a completed echo or a working path to the final destination.

What does an operator do when the network is broken but the tools for finding the break disagree with one another? Susan Hares and Cathy Wittbrodt described precisely that position in RFC 1574. As OSI network protocols appeared alongside TCP/IP, they said, network-layer debugging tools were scarce, and existing implementations often could not interoperate. Their answer was a short list of things a host or router providing OSI service had to let an operator find out.

The first was whether a node could be heard. The companion RFC 1575 defined an echo function for CLNP, the connectionless OSI network layer. RFC 1574 used its request and response selectors, 1E and 1F in hexadecimal, and required the requester to set the ERROR report flag. A newer node could answer with an echo. An older CLNP node might not recognize the selector, but it was supposed to try sending an error packet back. The supposedly failed ping was already information: the probe had reached a machine capable of complaining.

That distinction matters in a fault call. An echo response says the target handled the new request. An unsupported-option error says a CLNP node received something and reported that it could not handle it. Silence says less; the reply may have been lost, filtered or never generated. RFC 1574 asked the ping utility to show each round-trip time, aggregate minimum, maximum and average times, and the actual error code. Folding an error into a generic red lamp would throw away the one clue the network gave back.

Next came the path. The memo used CLNP's Lifetime field much as IP traceroute uses TTL. Send an echo request with Lifetime 1, and the first hop should return a Lifetime-expired error. Send another with 2, and the next hop may answer. Increasing the value exposes a sequence of responses until the destination answers or a reported error changes the diagnosis. The code matters: A0 and A1 identify expiration conditions, B0 signals an unsupported option, while 80 and 81 report an unreachable or unknown destination. A line of responding hops is evidence about those hops, not a promise that the application at the far end works.

The RFC also allowed a partial source route, a way to ask a probe to pass through a chosen intermediate point. But ISO 8473 told a node that did not support that option to forward the packet as if the option had not been selected. A packet arriving at the destination could therefore be a trick answer to the operator's question: it travelled, but perhaps not through the requested place. The trace had to report source-route errors such as unknown address or unacceptable path instead of decorating a successful arrival as proof of the chosen route.

Finally, the operator needed to know what the device itself believed. RFC 1574 required every OSI host or router to provide some out-of-band way to dump its routing tables. Routes might come from ES-IS, IS-IS, IDRP, a static entry or another source. The memo suggested a show osi routes command with filters, plus access through SNMP or CMIP where available. It acknowledged that the then-current CLNS MIB did not expose everything the operator wanted. A table dump was the third witness: more specific than an advertisement that the device was alive, less conclusive than traffic reaching its destination.

The document was an Informational specification of tools, not a report of a named outage or a measured worldwide deployment. A dated trace from a particular CLNP implementation would show how its echo errors, hop replies and route dump behaved in practice. The useful historical move was already in the proposed contract: an error could be a useful answer if the tool kept its exact meaning. The danger was to promote that answer into a verdict about the whole path.