The B5K architecture is the prototype for what modern systems aught to be like. If I ever strike it rich, my next business would likely be reviving the B5K for enterprise. Think TCO,
enhancements to support dynamic languages, etc. Oh, and I'd use pascal style strings (32bit length field minimum).
The idea of an instruction set designed for the application is enlightened. Obvious yes, but not obvious enough for people to follow through with it.
EDIT: Also, this and the Carmack keynote in one day makes HN worth all the noise. This is what I came here for.
Knuth coded the whole Algol compiler in three months, on a computer on which other programmers and a payroll processing had higher priority and received for that money that a top programmer earned for a year at that time:
Reading further in the conference transcript I noticed one other major figure who makes an appearance: Ken Iverson. Here's what Barton says:
There are lots of funny might have beens in connection with this. You didn't ask the question, why didn't the machine have index registers? [...] Well, I think that what happened really was something like this. We had Iverson as a consultant and I think Duncan MacDonald must have had a lot to do with arranging that. He was a consultant in engineering at the time, but that time when I was a Burroughs employee, I was asked to contribute some money from my budget. I agreed to doing that, and I remember going to a series of talks by Iverson, and my reaction was, "Wow!" He was still at Harvard at the time; he had an unpolished system (it was many years before the book was published), but I sat in that lecture thinking you can do the kind of things he's talking about in a machine with very little more hardware than it takes for three index registers. I used to argue against index registers on the grounds that ALGOL indexing was more general, but in the back of my mind I was hoping that the machine could have vector operations. I wasn't very good at convincing anybody of that; in fact, I failed again with the 6700 program. I failed. One of the customers convinced Burroughs to do a retrofit on vector operations at that time -- I don't think a very good job, either. But that's a perception I have that has nothing to do with the record. I bet Paul cannot produce a single item that will show that there was ever a suggestion that vector operations be built into the machine. Am I right?
This is a real gem. I've been thankful to have a great career in computers the last 20 years and everyday I'm reminded how little I actually know about our history. We really do stand on the shoulders(and success) of giants.
The idea of an instruction set designed for the application is enlightened. Obvious yes, but not obvious enough for people to follow through with it.
EDIT: Also, this and the Carmack keynote in one day makes HN worth all the noise. This is what I came here for.