From iWarp to Knight's Landing: James Reinders leaves Intel Intel's HPC director and evangelist James Reinders is leaving Intel after 27 years - or as he puts it, 10,001 days - accepting the firm's offer of early retirement for long-standing employees. Reinders describes how he joined Intel in 1989 to work on a VLIW (Very Long Instruction Word) processor called iWarp, designed to be …
Real Computer Scientists knew the future was parallel, even in late 1970s.
In the 1980s
We had the Transputer.
Ivor Catt.
Occam
Co-routines et al. built in to Modula-2.
Despite James Reinders, CAR "Tony" Hoare and others we have not made much progress since 1980s with "modern" multicore CPUs depending too much on OS level multithreading and bottlenecked by shared I/O and shared external RAM.
A beer for James Reinders.
All interesting ideas, but not really Computer Science
- the Transputer, interesting, but extendable instructions were trashed by RISC
- Occam - write only code, helped by a folding editor
- co-routines, buy into that a little, I ported a modula 2 compiler to the interesting Intel hyper cube (iPSC), but difficult to implement efficiently
- Tony Hoare, again interesting stuff but a little bit of a dead end
- Catt, OK, don't know enough to comment.
Occam was fine, if limited, as long as you realised it was effectively assembler. And all that formal methods stuff was definitely computer science.
Go, the big language with the cool start ups, has a concurrency model built around CSP. Personally I think CSP has yet to reach its full potential and all this messing around with explicit threads and locks has put held us back.
Having used Occam for several years to build quite large code bases (thanks Edinburgh SCC) I must disagree. It was a relatively low level language (in terms of base abstractions) but it was no more an assembly language than C is. The problem - a fundamental one is that of understanding the behaviour of codes of any real complexity. Hence the ongoing problems of maintenenance, re-use and tuning.
Few - actually I would say no programmers can get their heads around even moderately complex parallel systems in terms of their structure and behaviour.
In terms of CS, well I am sure that some might argue that it was mathematics, rather than engineering (in that most of the efforts to take pure and relatively simple structures and engineer real complex systems did not demonstrate leaps forward on the experiment, model, change loop). But I'm not suggesting that there was a lot of very good work, just that it seems to have been a dead end. As was other fascinating concurrent work on things like lazy evaluation systems.
A lot of extremely smart people have tried to push the work forward, but with limited success.
Even competing work (such as Robin Milners CCS - Milner as also awarded a Turing award as well as Hoare, so the work is not exactly obscure) has had limited, but IMHO more widespread influence on the understandability of concurrent systems.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
