This page contains the data set and some of the results and heuristics of the papers
These data are produced from a set of sparse matrices from the DWT collection, available at the Matrix Market.
See the codes to create the data.
The data files are named according to the following convention: t[T]p[P]r[rho]ETC[X].I where
P T E %First line: procs, num tasks, tot number of edges (2*num_comm)). etc(t1, p1) etc(t2, p1) ... etc(tT, p1) etc(t1, p2) etc(t2, p2) ... etc(tT, p2) ... ... ... ... etc(t1, pP) etc(t2, pP) ... etc(tT, pP) vi comm(1,i) vj comm(1,j) ... %The neighbors of task 1. Here vi and vj are its neighbors with communication amout next to the id of the neighbor. vk comm(2,k) vl comm(2,l) ... %The neighbors of task 2. ... vy comm(T,y) vz comm(T,z) ...%The neighbors of task T.
Distances between processors (for P=2,3,4,8 processors, the PxP principal submatrix is used):
0 1 3 2 5 5 4 3 1 0 2 1 4 4 3 2 3 2 0 1 4 4 3 2 2 1 1 0 3 3 2 1 5 4 4 3 0 2 1 2 5 4 4 3 2 0 1 2 4 3 3 2 1 1 0 1 3 2 2 1 2 2 1 0
Optimal Cost | ||
---|---|---|
Data file | Homogeneous Comm. | Heterogeneous Comm. |
t59p8r1.4ETC3.1 | 5243 (sol) | 5915 (sol) |
t72p8r1.0ETC3.1 | 2508 (sol) | 2850 (sol) |
t87p4r1.4ETC3.1 | 11756 (sol) | 12275 (sol) | t162p3r1.0ETC3.2 | 22674 (sol) | 22674 (sol) (Yes the same cost, the same assigment; but different number of nodes explored in A*) |
You can find the codes of the heuristics discussed in the paper "Bora Uçar, Cevdet Aykanat, Kamer Kaya, and Murat Ikinci, Task assignment in heterogeneous computing systems, Journal of Parallel and Distributed Computing, Vol. 66, No. 1, pp.32--46, 2006," here. This tgz file also includes the necessary codes to create the data files (see also above).
There are two files and five directories:
Each directory contains a suitable Makefile.