348 European Journal of Operational Research 57 (1992) 348-354
North-Holland
Theory and Methodology
Task scheduling with interprocessor
communication delays
Philippe Chr6tienne
Universit~ Pierre et Marie Curie, Laboratoire de Mdthodologie et Architecture des Systkmes Informatiques,
Tour 55-65 B320, Place Jussieu 4, 75252 Paris Cedex 05, France
Received July 1989; revised April 1990
Abstract: Distributed memory architectures raise new and complex scheduling problems. In this paper,
we first define a basic distributed memory computer scheduling problem issued from an ideal architec-
ture. By proving that the corresponding decision problem is NP-complete, we show that unlike shared
memory computer scheduling problems, these new problems do not become easy when the processor
limitation constraint is removed. Finally, we improve the knowledge of the borderline between the easy
and difficult subproblems of the basic one by giving some polynomial special cases.
Keywords: Scheduling, complexity, algorithms, interprocessor communication delay
Introduction
To efficiently execute an application program
on the processors of a distributed memory ma-
chine, three difficult problems have to be solved.
First, the application program must be split into a
set of tasks. Then, each task must be assigned a
processor (this is the well-known placement prob-
lem). Finally, a starting time must be assigned to
each task so that all the precedence and resource
constraints are satisfied.
In this paper we take as input a set of tasks
with precedence constraints and interprocessor
communication delays and minimize the make-
span of the corresponding scheduling problem
[1].
In Section 1 we specify the data of the schedul-
ing problem (called FDPS) we get when the num-
ber of processors is limited. We then consider the
case (called IDPS) when there is no limitation on
the number of processors. In Section 2, we prove
that the IDPS decision problem is NP-complete;
this result shows that, unlike shared memory mul-
tiprocessor scheduling problems [2,4], a dis-
tributed memory computer scheduling problem
does not necessarily become an easy problem
when there is no resource limitation. In the last
section, we present two polynomial special cases
of IDPS.
1. The two problems FDPS and IDPS
A generic instance f FDPS is specified in terms
of the following five parameters (I, U, p, v, m),
where
I = (1 ..... n} is a finite set of n tasks;
G = (I, U) is a directed acyclic graph;
Pi, i~I, is the processing time of task i,
whichever processor executes it;
vi~ , (i, j) ~ U, is the communication time from
task i to task j (i.e., the time needed by task i to
transfer data to task j if tasks i and j are as-
signed distinct processors);
0377-2217/92/$05.00 © 1992 - Elsevier Science Publishers B.V. All rights reserved