Summary

  • RFC 1201 let a datagram span up to 120 ARCNET frames, each carrying at most 504 octets of client data.
  • The 60,480-octet maximum did not settle a network's usable MTU; RFC 1201 required that limit to be configurable and shared among its implementations.

Begin with the frame, not the headline number

A long ARCNET frame was 512 octets, but 512 was not the amount available to IP. The hardware header, software header and fixed-frame padding occupied part of the format; software saw up to 504 octets of client data in a long frame. The short frame was 256 octets. Three awkward client-data lengths—250, 251 and 252—required an exception frame. Those details mattered because ARCNET's hardware did not simply carry an arbitrary-length packet.

The earlier RFC 1051, published in 1988, mapped IP and ARP onto an older ARCNET packet format. Where some stations lacked extended-frame support, it advised an IP MTU of 253 octets and encouraged IP-level fragmentation. RFC 1201, published in 1991 and replacing RFC 1051, moved the split into ARCNET's data-link layer. A split flag identified an unfragmented packet, its first fragment, or a later fragment; a sequence number tied the fragments to one packet. The receiver could rebuild the link-layer packet before passing the IP datagram onward.

A ceiling is not a network setting

With as many as 120 fragments and 504 client-data octets in a long frame, the format could describe a 60,480-octet packet. The RFC immediately called that length impractical and required each implementation to make its maximum packet size configurable. Nodes on one ARCNET network therefore needed a compatible setting. Implementations had to handle datagrams of at least 576 octets and were strongly encouraged to handle 1,500; neither requirement says that a particular network configured, transmitted or delivered a packet of that size.

The smaller 504-octet value avoided ARCNET data-link fragmentation. That could simplify reassembly, but it meant more IP packets for every node along a path. The RFC suggested advertising a small size with TCP's Maximum Segment Size option or an MTU-discovery mechanism, citing RFC 1063. The distinction is useful: a frame limit, a configurable interface maximum, a path constraint and a delivered datagram are four different facts. RFC 1191 later describes path-MTU discovery for IPv4; it does not turn RFC 1201's ARCNET maximum into a measured path result.

Reassembly had costs of its own

RFC 1201 says fragments are sent in order and recommends reserving enough space for the complete packet when its first fragment arrives. If one fragment arrived out of order, a receiver could abandon that packet. It could also discard an incomplete reconstruction after a few seconds without another fragment. Yet a repeated fragment was different: ARCNET hardware might deliver a frame successfully while its acknowledgement failed to reach the sender, which could retransmit it. Reassembly should ignore that duplicate instead of throwing away the packet.

The link layer's receipt was therefore not proof that a whole IP datagram had been reconstructed. The RFC described how software should manage ordering, storage and repeats; it did not report a capture, a successful application exchange, or how often any of those cases occurred. RFC 791 defines IP's own datagram and fragmentation model, while RFC 826 supplies the ARP context. RFC 1201's contribution was the ARCNET-specific boundary between those layers.

Coexistence did not mean compatibility

RFC 1201 reports that five companies agreed in 1989 to use the ARCNET data-link format referenced by the new specification. Its IP, ARP and RARP protocol identifiers were 212, 213 and 214. The predecessor RFC 1051 used different values. Both encapsulations could exist on one ARCNET, the RFC says, but a station using one could not communicate with a station using the other. A shared physical medium had not reconciled their packet formats.

The RFC Editor now catalogs RFC 1201 as STD 46, an Internet Standard. That publication history establishes a recorded specification, not the share of ARCNET stations that implemented it. The history visible in the document is narrower and more useful: a fixed-frame medium led to a new fragmentation layer, while a maximum packet size remained a local network agreement and a changed protocol identifier left an explicit compatibility boundary.

Sources