The Beginnings of Computer Networks
4.3 Delivery of the First IMP to UCLA - September 1969
The schedule called for UCLA to receive the first IMP from BBN since UCLA, under the supervision of Kleinrock, was creating the software and hardware needed to measure network performance. Since no network like the Arpanet had ever existed before, measuring its performance was crucial to understanding its operating characteristics and, by so doing, knowing when and how to expand the network without inducing catastrophic problems. In recognition of its critical role, UCLA became known as the Network Measurement Center or NMC.
In a meeting on March 21, 1969, Kleinrock reviewed the goals and objectives of the Arpanet work at UCLA. The following are excerpts from a memo he distributed on April 25th documenting what had to be done and assigning responsibilities.
"To: ARPA Employees Working on COMPUTER NETWORK RESEARCH Project
Goals and Objectives
Our point of departure is the stated objectives as quoted below in our proposal to ARPA:
"The main purpose of this contract shall be to engage in those activities which enable UCLA to participate as a viable, creative, and significant node in the ARPA experimental computer network. These activities include:
-implementation of hardware and software to couple the UCLA HOST to its IMP;
-research in the development and use of analytical and simulation models of computer networks and time-shared computer systems;
-development of software procedures for active participation in the initial and full ARPA net;
-develop software suitable for measuring those network parameters which allow one to understand network behavior, which allow control of network functions and which permit us to assess the validity of our mathematical and simulation models;
-carry out studies on the use of remote facilities in a network environment."
These objectives fit into what I see as the medium term development of computer research here at UCLA. Broadly speaking, one of our goals is to create science and engineering out of the 'art' portion of the field. This technology will lead to the intelligent design and prediction of existing and new computer systems."
Kleinrock ended the memo:
"We have a busy few months ahead of us running into late fall 1969. The real burden does fall upon UCLA since we are the first node in the net and it is important that we carry out our responsibilities in this role successfully. Following the creation of this network will come 'fun time.' That is, the use of the network and its vast power and challenge will provide considerable diversity of activity."
As NMC, UCLA’s responsibilities ranged from working with BBN to make sure the necessary measurement capabilities were designed into the IMPs, to providing constant network monitoring and analysis once the network was operational. Knowing Roberts wanted to expand the number of sites connected to Arpanet as quickly as possible, the NMC plan called for three months of tests with the first four IMP's prior to network expansion. John Postel, another graduate student working with both Estrin and Kleinrock on measuring computer performance, remembers:
"So we threw away all this stuff about probing inside another computer and began to think about how to measure networks."
In addition to his role in the group defining the host-to-host software, Crocker managed UCLA’s site preparation. The hardware needed to connect their computer(s) to the IMP was a high priority. Crocker looked into having an outside contractor build the connection, but a bid of $19,000 and six months prompted him to give the job to Mike Wingfield, a graduate student, who had a prototype working in less than six weeks for $4,000. The first IMP interface was a "box about a foot cube, maybe a little bigger, and elegant." Crocker soon wished everything else were proceeding as smoothly. He remembers:
"Well, we were having a little trouble with the software. No big surprise, getting our software ready, and I knew that BBN was struggling with some subtle timing problem in the IMP's and they were real worried about whether they were going to make their deadline. I looked at the calendar and I said: 'Oh well, September 1 is Labor Day. There's no problem, because that's a holiday, that's Monday.' So I reasoned like this, I said: ' First of all, BBN is running late. Second of all, September 1st is a holiday, so it'll slip sometime after that. They'll ship it out here and Honeywell will come in and try to get the machine working, and then BBN will send out a team and see if they can get their software working on top of that, so I've got an easy two weeks minimum.'
BBN fixed their problem, looked at the calendar and said:
'We're not going to miss the first date. Air ship the thing,' and it wound up on our loading dock on Saturday, the end of August, and on the same day somebody showed up and wheeled it in and plugged it in and the program kept running from when they had unplugged it, and they had a working IMP. So all the reasons why I thought I had two weeks collapsed to about minus two days, and there it was. That was a bad day. So that was September, 1969."
Kleinrock too recalls the timely arrival of their IMP:
"Well, to BBN's credit, they connected this thing up, and that day we had bits moving back and forth. I think the next day we had packets moving back and forth and messages worked. It came in and the damned thing worked!"
A milestone, one that even weeks earlier had seemed out of reach, had been achieved. BBN delivered the second IMP to SRI in early October, and UCSB and the University of Utah received theirs in November and December, respectively.