[Openmcl-devel] Quick HW question...

[Paul Krueger]

On Nov 16, 2010, at 12:30 PM, Spires, Shannon V wrote:

> We either need to find a way to build uniform-speed memory (including the non-volatile variety) or move to something like Cray's XMT architecture if we want Lisp to run really fast.

I was V.P. Software Engineering at Cray for a while and always thought that the XMT would have made a stupendous Lisp machine for the right sort of applications. For those who don't know, it's a massively multi-threaded architecture with the objective of having enough active threads that there is always at least one ready to run (i.e. not waiting on some memory reference) so processors are never stalled waiting on data. Although no single thread runs any faster than on any other platform, the net throughput can be substantially better than other architectures if you can keep enough threads active. This requires the right sort of application as well as a clever compiler.

I even started to look at what a port of CCL to the XMT might take. I was unable to convince the rest of the management team that there would be any significant demand for it, however, and dropped the idea. I am now retired and doing my own research (using Lisp of course). I still have in the back of my mind that if I get to the point of needing a really large-scale massively parallel Lisp machine that I might go back to Cray with the idea. There are, of course, other architectures that have similar objectives, but none that I am aware of that do it quite the way the XMT does.

Paul