MQFURP: An Overprovision Strategy Supporting Performance Interference Management in Cloud

Cloud computing has become a prevailing trend within the computing landscape, enabling users to access computational resources on-demand through the deployment of virtual machines on physical servers. To enhance resource utilization, cloud administrators commonly employ a technique known as resource overprovisioning, where more virtual resources are allocated to clients than what is physically available. This practice optimizes the utilization of computing resources and is frequently adopted by well-known virtual machine monitors (VMMs) like KVM to maximize the efficiency of cloud data centers. Resource overprovisioning strategies are designed to support more virtual machines (VMs) with limited resources. However, a challenge arises in the form of performance interference among VMs co-located on the same physical host. This interference issue may be exacerbated by resource overprovisioning, which aims to accommodate more VMs within the same physical machine. The performance degradation results from contention over shared resources among co-hosted VMs, leading to a violation of VM quality-of-service (QoS) standards. Resource contention occurs when the demand for resources being shared surpasses the incoming request as a result of co-hosting virtual machines. In our paper, we introduce a resource overprovisioning strategy called the Multi-level QoS Framework Using Resource Pooling (MQFURP). This strategy aims to establish a multi-level QoS environment within a cloud by utilizing Linux control groups. These control groups enable the differentiation of QoS levels for virtual machines, allowing for better and more flexible control over resource management. Our evaluation was conducted within our private cloud infrastructure, which is built on OpenStack. The results of our evaluation demonstrate that MQFURP successfully ensures that each VM meets its performance requirements even when co-hosted with other virtual machines on the same physical machine.