Implementing Malleability on MPI Jobs
Gladys Utrera, Julita Corbalán, Jesús Labarta · 2004
Abstract. Parallel jobs characterized for having processes that communicate and synchronize each other frequently. A processor allocation strategy widely used in parallel supercomputers to optimize their performance is Space-Sharing, that is assigning a processors partition to each job for its exclusive use. The Moldability can reduce the queuing time by sizing jobs to the available resources at the start of execution. However, they will remain fixed even though the system load varies. Only Malleable jobs will be able to adapt to load changes. A way to simulate Malleabilty using Moldability is by applying the Folding technique. In this article we propose a processor allocation strategy to message-passing parallel jobs, based on Folding and Moldability concepts. This technique tries to decide the optimal initial number of processes, when to fold and the number of Folding times by analyzing the current and past system information. At processor level, we apply Co-Scheduling. We implement our proposal and compare to an implementation of Folding and some other Moldability from the bibliography. We evaluate them under several workloads with different job sizes, classes and machine utilization. Results show that the proposed technique adapts easily to load changes, and can obtain better performance than the rest evaluated, on workloads with high coefficient variation and especially with burst arrivals. 1