Re: The future of CSP? Fault tolerant implementations?

["G. S. Stiles" <dyke.stiles@xxxxxxxxxxx>]

Greetings -

I just started up my fault-tolerance course, and the quote below struck a
cord...replication being a foundation of tolerance...

>Your ideas are brillant. :-)
>The more threads, the better.
>The less nonsense like synchronisation, the better... :-)

>Your approach solves one of the major problems caused
>by synchronisation: deadlock. When there is no synchronisation,
>there is no deadlock. Excellent!

If we have large numbers of threads, unsynchronized, each with a substantial
number of operations, the number of possible interleavings is astronomical.
Some interleavings may be ok, some not.

Suppose we launch multiple copies of each thread, and assume each thread
can check its results for correctness. With enough copies, we should get at
least one thread that yields a correct result. Threads that yield incorrect
results swallow those results and continue on (maybe restarting with the
correct result).

If threads can't self-check, copies of the same thread could check against
each other.

In either case different threads (thread groups) do not slow each other down.

Probabilities seem important here.

Free-range threads?

Looks like a good term project for the course...

Dyke.


--
======================================================================
Dyke Stiles
Professor
Department of Electrical and Computer Engineering
Utah State University
4120 Old Main Hill
Logan UT 84322-4120
Voice: +1-435-797-2840                          FAX:   +1-435-797-3054
Work:                http://www.engineering.usu.edu/ece/research/rtpc/
Play:  http://www.engineering.usu.edu/ece/research/rtpc/utah/utah.html
======================================================================