Minimizing the stretch when scheduling flows of divisible requests
We consider the problem of scheduling comparisons of motifs against biological databanks. This problem lies in the divisible load framework with negligible communication costs. Thus far, very few results have been proposed in this model. We first explain the relationship between this model and the preemptive uni-processor one. After having selected a few relevant metrics (max-stretch and sum-stretch), we survey, improve, and complete the existing results for the uni-processor model. Then we derive algorithms for the multi-processor case and we extensively study their performance in realistic scenarios. Our study clearly suggest an efficient heuristic for each of the two metrics, though a combined optimization is in theory not possible in the general case.