MVILAIN@G.BBN.COM (Marc Vilain) (01/21/88)
BBN Science Development Program
AI Seminar Series Lecture
DISTRIBUTING BACKWARD-CHAINING DEDUCTIONS TO MULTIPLE PROCESSORS
Vineet Singh
Stanford University, and SPAR
(VSINGH@SPAR-20.SPAR.SLB.COM)
BBN Labs
10 Moulton Street
2nd floor large conference room
10:30 am, Friday January 29th
The talk presents PM, a parallel execution model for backward-chaining
deductions. PM exploits more parallelism than other execution models
that use data-driven control and non-shared memory architectures. The
talk also presents an application-independent, compile-time allocation
strategy for PM that is both fast and effective. Effectiveness is
demonstrated by comparing speedups obtained from an implementation of
the allocator to an unreachable upper bound and speedups obtained from
random allocations. The resource allocator uses probabilistic
techniques to predict the amount of communication and the parallelism
profile; this should be useful for other allocation strategies as well.
-------