C2QoS: CPU-Cycle based Network QoS Strategy in vSwitch of Public Cloud

The network Quality-of-Service (QoS) strategy in vSwitch aims to guarantee the Service-Level-Agreement (SLA) for the concurrent Virtual Machines (VMs) residing on a particular server platform. Different from a hardware switch, the traffic forwarding tasks in vSwitch are completed by processes/threads running on some dedicated CPU cores. Existing vSwitch QoS strategies, inherited from the solutions in the hardware switch, are interface-based and ignore the isolation of these IO-dedicated CPU resources among VMs. As the result, they cannot ensure VMs’ SLA targets due to the resource contention. In order to resolve the issue, we propose a CPU-Cycle based QoS (C2QoS) strategy, that contains a CPU-Cycle based Token-Bucket (C2TB) mechanism and a Hierarchical Batch Scheduling (HBS) mechanism. The C2TB apportions the IO-dedicated CPU resources to each VM for ensuring bandwidth, while the HBS schedules the VMs’ forwarding tasks on these CPU cores to guarantee hierarchical latency. We implement the C2QoS strategy on the Data Plane Development Kit accelerated Open vSwitch (OVS-DPDK) platform. Experimental results show that compared with existing strategies, the influence of CPU resource congestion on bandwidth is eliminated and that on latency is reduced by 80%.