Scheduling mechanism on virtual machines for supporting latency-sensitive tasks

In virtual machine environments, latency-sensitive tasks are difficult to support in a timely manner, especially when many domains have boosted priorities. In this paper, we describe the problem with MultiBoost and present a low-latency scheduling mechanism that uses VCPU shaping. The key idea of this paper is to adapt VCPU characteristics into a scheduling policy by investigating and measuring the behavior of each domain. We evaluate our prototype in terms of latency with diverse workloads. Our experiment results show that the proposed method effectively allocates CPU resources for low-latency tasks.