An extended task management to improve QoS of cloud computing services in NUMA systems

For Cloud computing services, efficient task management to sustain I/O requests with latency outliers reduced is an important design requirement for QoS. However, the existing task management techniques do not adequately meet this requirement on modern server systems, especially NUMA machines. In this paper, we propose an efficient task management scheme for NUMA systems. Unlike existing one in Linux, the proposed scheme considers the storage/network asymmetry of modern NUMA machines in managing network and storage I/O requests. By exploiting the storage/network asymmetry in task scheduling, the proposed scheme reduces the tail latency and improves the QoS of Cloud computing services in a NUMA system. The experimental result shows that the proposed scheme improves QoS by up to 21%.