Datacenter Networking Nexus

Before the megabit: A trip through vintage datacenter networking

When it was all about the baud rate

The world of datacenter networking is crammed with exotic technology and capabilities beyond the imaginings of administrators charged with running big iron decades ago. However, while it might have been a slower and more proprietary time, it was also perhaps a little simpler.

Dr Andrew Herbert, Chairman of the Board of Trustees at the UK's National Museum of Computing (TNMOC), is very familiar with the Elliott 900 minicomputer series and noted that in the late 1960s and early 1970s, the world of networking was very different from how it is today.

While the architecture of the 900 series had no support for partitioning memory (requiring cooperation for multi-user activity), and many ran without any operating system at all, there was an optional NPL interface.

Herbert remembered it as "a sort of parallel port," likely used mainly for controlling instruments, but "NPL were, of course, networking pioneers."

Donald Davies and his team at National Physical Laboratory (NPL) came up with packet switching in 1965 (although Paul Baran had independently devised a similar concept a few years earlier as distributed adaptive message block switching).

"Mainframe manufacturers," recalled Herbert, "defined their own proprietary network protocol stacks, e.g. IBM System Network Architecture, Digital's DECNet.

"These generally ran over leased lines between datacenters, or in star configurations linking smaller branch office minicomputers to a central mainframe. They were mutually incompatible, although many vendors sold adapters to hook up to other vendors' networks.

"Users found this frustrating, especially when combining businesses that had made different technology choices."

A volunteer at TNMOC said: "All the major computer manufacturers had a proprietary polled, half-duplex protocol in the 1970s – these were very efficient when used on 'multidrop' leased lines in the USA, but similar lines weren't generally available in Europe."

"IBM's polled protocol was Bisync, Univac's was 1004, and I think ICL's was C01." ICL used a protocol called C01 primarily to manage communication between the mainframe and its terminals.

Then there was the challenge of connecting to resources in the datacenter in an era when dial-up was the only option.

The earliest terminals were little more than electromechanical teletype machines running at 110 baud, and the modem used could be as simple as an acoustic coupler into which a telephone handset could be placed. Another TNMOC volunteer told us that some of the UK's major banks (such as Barclays) had a mighty leased 1,200-baud analog network connection for their terminals. However, "terminal" underplays the devices somewhat. "These were, for the time, very intelligent, full-blown computers in their own right," he said, "and whilst they were of limited use when the connection to the computer center was down, they could still function."

"If the leased line failed, they had a backup 600 baud dial-up (where, for security reasons, the computer center always dialed the branch)."

Herbert continued: "In the late 1970s, wired local area network technologies emerged (Cambridge Ring, Token Ring, Ethernet) to connect terminals and computers in the same building/campus together. Ethernet came to dominate.

"Then there was the tussle in the 1980s and 1990s between proprietary network stacks, 'Open Systems Interconnection' network protocols, and academia-led TCP/IP developed from ARPANet. TCP/IP came out top."

And the rest is history.

The UK's National Museum of Computing has several working systems rescued from the datacenters of yesteryear, including an ICL 2966 from the 1980s and, of course, the Elliotts from the 1960s. To scratch that networking itch and go back to a time when performance was more leisurely, there's always the NPL gallery. ®