Name
Abstract | ||
|---|---|---|
During the implementation of an event driven simulation package, we faced with a stack management problem. Conventionally, in a muti-threaded system a separate stack is associated to every thread created. Naturally, the memory requirement to support the threads increases as the number of threads increases. At some point, the system may run out of usable memory; furthermore, even if there is sufficient memory, the management of such large memory can severely undermine the performance of the system. Alternatively, the stack sharing scheme is commonly used. The system maintains only one stack and all threads share this common stack. In order that this scheme works, the old content of the stack must be saved and the new content be restored everytime when a thread switching occurs. This overhead can be intolerable for large applications (in our case simulation) which require thousands of threads. In this paper, the implementation of a hybrid stack sharing scheme which combines the advantages of the two conventional stack management schemes are described. |
Year | DOI | Venue |
|---|---|---|
1994 | 10.1145/326619.326822 | SAC |
Keywords | Field | DocType |
genetic algorithms,optimization,data parallelism | Computer science,Task parallelism,Parallel computing,Thread (computing),Data parallelism,Genetic algorithm | Conference |
ISBN | Citations | PageRank |
0-89791-647-6 | 2 | 0.40 |
References | Authors | |
1 | 2 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Kam-Fai Wong | 1 | 1718 | 176.33 |
| Benoît Dageville | 2 | 203 | 25.83 |