Summary

  • MILNET service was established on 4 April 1983; the logical ARPANET/MILNET split was completed on 4 October 1983; the physical partition was completed on 6 September 1984.
  • The separation moved operational users behind new boundaries while leaving controlled gateways for limited exchange. The documents establish the sequence, but do not measure whether it reduced intrusions, outages or operating costs.

One word, several cutovers

On 4 October 1983, newspaper accounts described the Defense Department as separating ARPANET into a military network and a research network. The date was real. It was not the date every wire moved. The distinction appears in the contemporaneous report itself: UPI called that day’s separation “logical” and said physical separation was still planned for 1984. A later Defense Department review gives the fuller chronology: MILNET service began on 4 April 1983, software-enforced separation was completed on 4 October, and the physical partition was completed on 6 September 1984.

That chronology changes the story from a ceremonial rename into an operations program. “MILNET split” can refer to a service, a rule about which hosts belonged to which community, controls at gateways and terminal access points, or separate physical infrastructure. Those boundaries did not arrive together. Reading every use of “split” as one event erases the work that made the separation executable.

The policy context preceded the October date. In 1982 the Defense Department cancelled AUTODIN II, a planned communications system, in favor of ARPANET-derived designs. A 1985 National Research Council report lists security, architecture, survivability and cost among the reasons. It describes MILNET as a network for unclassified Defense Department operational data, with ARPANET continuing to support research organizations. Most of ARPANET’s capacity was assigned to MILNET.

That was a new operating purpose built from a research network’s installed capacity, not a claim that the Internet itself had been divided into “civilian” and “classified” halves.

From operating service to a logical boundary

The first documented milestone is the service date. The Defense Data Network review says MILNET service was officially established on 4 April 1983. That gave the operational segment a name and a place in the Defense Data Network before the logical partition was complete. The available sources do not say that every subscriber or interface changed on that day.

The next milestone was the logical split. RFC 881, published in November 1983 as a plan for domain-name deployment, puts “The ARPANET/MILNET Logical Split” on 4 October. It is a schedule, not a detailed after-action report. The same schedule places MILNET gateway access controls and TAC user access controls around 15 February 1984, explicitly marking that date approximate. The ordering matters: separation of communities could be enforced in software while the physical partition remained unfinished.

A Washington Post report published on 4 October attributes the move to Defense Department concerns about access to sensitive information and disruptions from research experiments. It also reports that a senior official denied any break-in had occurred. The distinction is important. The article records a stated security rationale and a denial; it does not establish a prior compromise. The report describes mail bridges between the networks, while UPI’s same-day account says the logical separation was followed by a planned physical split and later mail-only communication.

Those were contemporary descriptions of a plan, not evidence that every restriction operated exactly as forecast.

The 1985 Defense Data Network review supplies the most specific completion record. It says seven Internet gateways had been installed to permit interchange between ARPANET and MILNET and would remain after logical separation. It then dates the physical partition to 6 September 1984. Physical separation therefore did not mean that all communication stopped. It meant a different network arrangement, with controlled gateways retained as a bounded connection.

The boundary reached the interfaces

The change was more than a label or an address number. RFC 952 assigns network number 10 to ARPANET and 26 to MILNET. That is evidence of separate network identities in the host-table system; it is not by itself proof of physical separation or complete isolation.

A later technical record describes a deeper interface difference. RFC 3109 says the Department of Defense required a byte-oriented X.25 interface for MILNET’s packet switches, while research ARPANET machines continued using the older 1822 host-to-IMP interface under Internet protocols. This is a retrospective account, but it shows why the physical phase mattered: operational separation reached the equipment boundary as well as the software and naming layers.

MILNET was also growing while the changes were underway. RFC 942 reports fifteen additional subscribers in the second half of 1983 and says future growth depended partly on trunk and cryptographic-equipment availability. It does not provide traffic volumes, subscriber identities, utilization, or evidence that the growth was caused by the split. Its evidence is enough to show that the new network was an expanding operational service, not merely a second name for the same research project.

Why accounts give different years

A few later summaries state that ARPANET and MILNET split in 1984. RFC 1302, a 1992 history of network information services, uses that broad year. RFC 980, in 1986, distinguishes the early-1983 NCP-to-TCP transition from the later DDN creation and ARPANET/MILNET split. Neither account contradicts the more granular chronology: “1983” can refer to service and logical partition; “1984” can refer to physical partition or to the completed two-network arrangement.

The split was not the January 1983 TCP/IP flag day. Protocol conversion changed how ARPANET hosts communicated; the MILNET program reassigned users, access boundaries and network infrastructure. The changes were related, but they answered different operational questions. A system can run TCP/IP and still lack a separate physical backbone. A network can be logically partitioned while still exchanging traffic through controlled gateways.

The sources do not provide a complete topology, gateway policy, access-control audit, or incident series. They cannot establish that the change prevented a breach, eliminated experimentation-related disruptions, improved availability, or lowered costs. What they do establish is the sequence: a service boundary, a logical boundary, subsequent access controls and a later physical partition—with limited interconnection deliberately retained.

Sources