Chapter 9

SANDY - Late January 1961

We reconvened our meeting on Monday two weeks later. I was surprised that Steve was the only person from the agency this time. He said, "Sam and Murray are tied up with some other stuff, and Dr. Edwards was supposed to come, but was asked to take on a special rush project. You'll just have to make do with me."

Paul announced that a new group had been created to do the initial design and specification of Reaper - by now everyone was using that name. He said, "We have a small core group for now, but they are good. In addition to Sandy, Stew Green, who had been in my VFL group, is joining us, along with Chet Howard who comes to us from the I-box group. The I-box emphasis on overlapped operations is going to fit right into our streaming concept."

I knew Chet slightly. He was an older engineer, but sharp with a lot of creativity and a good sense of humor.

Paul then said, "Most of you don't yet know Sarah Wilson. She is a Senior Programmer, with a lot of experience working on pattern recognition algorithms; we hope her experience and expertise will be very useful. She's also worked on programming language and compiler design, and I think her participation will help us with the design of the instruction mechanism for Reaper.

"Finally, Chad Wilkinson from the tape area will not formally be part of the group at this time, but he has been assigned to work with us whenever we need his expertise."

Steve then started to outline the indexing and operation requirements. He said,

"We see the data as essentially continuous strings of characters. We think what will work well is having two independent input strings, and a third output string. We can specify the starting address of each string and then proceed through the data using indexing. In many cases, the indexing is fairly straightforward, but there are cases where the results of some of the operations on the data change the indexing increment, or even require that we go back earlier in a string."

I asked him, "Do you have any examples of how this works, and what are the operations you need to perform on the data?"

Steve replied, "Yes, here's what we've put together so far on the operations and conditionals, and a few representative examples of the indexing algorithms and the conditional factors we need to factor in." as he handed out a short memo.

We all took a few minutes to absorb this, and then I said, "I think we need to put together a first cut at how we could program the Reaper. We know a few things - first, you're going to want to initiate a Reaper program and have it run concurrently with a normal Stretch program. This implies that we need some additional Stretch instructions to initiate the program, say with an initial set of control words. Obviously, you also will need to be able to determine when it is complete. Secondly, it looks like we need to think in terms of a sequence of control words for each stream, with some branching capability. How about letting each control word group specify the address of the next group, but using a form of indexing on the basis of the results of the operation to modify that address, so that you have a limited branching mechanism? Obviously, we need to spend a lot of thought and time getting this down."

Steve said, "Sandy, that's excellent! Let's come back to that, but we also need to address the external storage issues."

Paul said, "Good idea. Let's break for lunch and talk about storage after lunch."

Steve and I sat down together for lunch. By now it seemed like a very natural thing to do. I told him, "Remember, I owe you a dinner. Does seafood sound good to you? We could go to the Anchor Inn."

Steve said, "I'd like that. I told Murray that I was going to be quite unhappy if I couldn't rent a car this trip, and he reluctantly agreed, and that he'd approve it from here on out. I guess after he experienced Howard Johnson's firsthand, he realized that it wasn't just a frivolous request on my part. At any rate, I'll drive if you give me directions."

After lunch, Chad took the floor and reviewed the Hypertape specifications, and even passed one of the cartridges around for us to look at.

Steve asked Chad, "Well, what can you do for us in terms of data transfer rate and capacity?"

Chad said, "I'm not sure how much we can extend the Hypertape design. Even if we got twice the performance out of it, if I remember correctly, it doesn't do the job for you, is that right?"

Steve replied, "No, that doesn't do it. But I've been thinking about this, and have a couple of ideas. First, if you were to make the cartridge wider, you could increase the data rate and capacity significantly. Also, can you conjure up some sort of automatic mechanism, sort of like a jukebox, that could quickly change cartridges? This could give us reasonably fast access to a huge volume of data. We can determine in the program which cartridges we will need next, so could have them preloading into available drives. That way, the time to access and load a cartridge would not be critical."

Chad said, "Actually, we were talking last week about a wider cartridge, and we think that we probably could do that. I'm going to want some time to think about that jukebox idea and how we might get it to work."

That evening, Steve and I had another very pleasant dinner. I was beginning to realize that time with Steve was always very enjoyable; he had a lot of interesting things to talk about. He was interested in hearing my descriptions of how things worked at IBM.

"It seems that the lab management feels a need to keep shaking things up. The formation of our group was a very minor example. Last year there was a major restructuring of the development organizations - Stretch was almost the only project that wasn't affected. They say that the half-life of a lab director is only about eighteen months, and then he disappears into the division headquarters bureaucracy which is downstate in White Plains."

Steve said, "Yes, there's a lot of that in the military and in the civilian government agencies. I think we have some good friends in high places, since we've been left relatively untouched. We certainly don't have a problem getting the funding we need."

The next two days, while Chad was meeting with his group, we went back to the structure of the instruction mechanism. We started with the concept of a set of control words that were preloaded into Reaper, and then a "Start Streaming" instruction that would initiate operations in the box. We were stuck on how to design a control word structure that would handle the indexing and conditional branching that we needed. We also knew that we had to enumerate the various types of operations and decision mechanisms. Finally, we agreed to work on this independently over the next two weeks, and then get back together again in Poughkeepsie.