Job Scheduling Strategies for Parallel Processing
th Thisvolumecontainsthepaperspresentedatthe12 workshoponJobSched- ing Strategies for Parallel Processing. The workshop was held in Saint-Malo, France, on June 16, 2006, in conjunction with SIGMETRICS 2006. This year, the presented papers covered a large variety of topics. The ?rst three papers address work?ow problems. “Provably e?cient two-level adaptive scheduling” by Yuxiong He et al. provides a theoretical analysis of a sched- ing approach for independent jobs consisting of threads, that are represented by a DAG. Job and thread scheduling are separately addressed with di?erent algorithms. The task graph is not known a priori in the paper “Scheduling - namically spawned processes in MPI-2” by Marcia ´ Cera et al., but processes are spawneddynamically. ThispaperisbasedonthefeaturesofMPI-2andevaluates its scheduler with the help of an experiment. The DAG of a Grid job is known at submission time in the problem discussed in the paper “Advance reservation policies for work?ows” by Henan Zhao and Rizos Sakellariou. Here, the tasks of this job are automatically scheduled on heterogeneous machines using advance reservationsuch that the overallexecution time frame of the user is obeyed. The proposed approach is again experimentally evaluated. The next three papers describe classical job scheduling problems that arise whenparalleljobsaresubmittedtoparallelsystemswithlittleornonodehete- geneity. The paper “On advantages of scheduling using Genetic Fuzzy systems” by Carsten Franke et al. presents scheduling algorithms that support arbitrary scheduling criteria. The algorithms are trained with recorded workloads using Fuzzy concepts. Their performances are evaluated by simulations with those workloads.