Efficient max-min and proportional fair constrained multiresource scheduling
We consider the problem of scheduling a group of heterogeneous, distributed processes, with different relative priorities and service preferences, to a group of heterogeneous virtual machines. Assuming linearly elastic IT resource needs, we extend prior results on proportional fair and max-min fair scheduling to a constrained multiresource case for a fam- ily of fairness criteria (including our recently proposed Per- Server Dominant-Share Fairness). Performance comparison is made by illustrative numerical example. We conclude with a discussion of scheduling problems for a public cloud with heterogeneous instances and servers.
READ FULL TEXT