Abstract
There are at least three dimensions of overhead to be considered by any parallel job scheduling algorithm: load balancing, synchronization, and communication overhead. In this work we first study several heuristics to choose the next to run from a global processes queue. After that we present a mechanism to decide at runtime weather to apply Local process queue per processor or Global processes queue per job, depending on the load balancing degree of the job, without any previous knowledge of it. © Springer-Verlag Berlin Heidelberg 2008.
Cite
CITATION STYLE
Utrera, G., Corbalán, J., & Labarta, J. (2008). Dynamic load balancing in MPI jobs. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 4759 LNCS, pp. 117–129). Springer Verlag. https://doi.org/10.1007/978-3-540-77704-5_10
Register to see more suggestions
Mendeley helps you to discover research relevant for your work.
