Hello, I'm sorry to bother you, but I'm reading your paper and I came across a question that I would like to ask you.
Regarding the big-c experiment part, set up two queues, one configured with 95% of cluster resources dedicated to run short jobs, and the other configured with 5% of cluster resources dedicated to run long jobs. When the short job is submitted while the long job is running continuously, the short job queue will seize the resources of the long job queue at that time, is that right?
If so, the job-level preemption policy does not work at all, because each queue of the capacity scheduler will only run one task at the same time, and the others are in the blocking state, and the submitted short job can only preempt the one running in the long job queue. In this case, what is the point of proposing a job-level preemption policy?
Sincere thanks!
Hello, I'm sorry to bother you, but I'm reading your paper and I came across a question that I would like to ask you.
Regarding the big-c experiment part, set up two queues, one configured with 95% of cluster resources dedicated to run short jobs, and the other configured with 5% of cluster resources dedicated to run long jobs. When the short job is submitted while the long job is running continuously, the short job queue will seize the resources of the long job queue at that time, is that right?
If so, the job-level preemption policy does not work at all, because each queue of the capacity scheduler will only run one task at the same time, and the others are in the blocking state, and the submitted short job can only preempt the one running in the long job queue. In this case, what is the point of proposing a job-level preemption policy?
Sincere thanks!