mwang@watmath.UUCP (mwang) (08/07/84)
_D_E_P_A_R_T_M_E_N_T _O_F _C_O_M_P_U_T_E_R _S_C_I_E_N_C_E
_U_N_I_V_E_R_S_I_T_Y _O_F _W_A_T_E_R_L_O_O
_S_E_M_I_N_A_R _A_C_T_I_V_I_T_I_E_S
_S_Y_S_T_E_M_S _S_E_M_I_N_A_R
- Thursday, August 16, 1984.
Dr. E.A. Ashcroft of Stanford Research Institute will
speak on ``An Eduction Engine for Lucid - A Supercom-
puter Based Upon Demand-Driven and Data-Driven Evalua-
tion.''
TIME: 3:00 PM (Please note)
ROOM: MC 6007
ABSTRACT
This talk will be about a proposed multiprocessor
machine design that is being studied at SRI Interna-
tional. A relatively simple prototype of this machine
might have 256 processors. (In fact, when we get fund-
ing for actually building a machine, rather than just
doing simulations, this is the size of machine we will
start with.) The machine is designed to have Lucid as
its assembly language. We feel that the main problem
with previous attempts at designing and building mul-
tiprocessor machines is that the problem of programming
the machines is left as an afterthought. Lucid
(designed by Bill Wadge and myself) is a very high-
level language, yet it is simple enough to be able to
design a machine to match it.
We have called the machine an Eduction Engine because
we wanted to get away from the word ``dataflow'', which
is synonymous with data-driven evaluation, and which
has not lived up to its expectations. A Lucid machine
requires demand-driven evaluation. (The word ``educ-
tion'', according to the Oxford English Dictionary,
means, among other things, the action of drawing forth
results of calculations from the data.) This talk will
concentrate on a purely demand-driven machine, and the
talk tomorrow, by Jagan Jagannathan, will concentrate
on a hybrid machine that uses data-driven evaluation as
well, when it can't cause any trouble.
August 7, 1984