Fine-grained Adaptive Interrupt Coalescing In SR-IOV Network

Trap-and-emulations are involved in handling I/O interrupts in virtual environment. Such procedure will cause heavy CPU overhead. To address this problem Huang et al.[37] proposed AIRC algorithm based on interrupt coalescing. AIRC controls interrupt rate dynamically based on the available CPU resource. AIRC can make full use of CPU resource to optimize network throughput. There are two main problems of AIRC: Firstly, network throughput is limited by physical device. Adjusting interrupt rate is no longer effective when network throughput reaches the upper bound in optimizing network bandwidth. Secondly, many parameters used in AIRC are hard to obtain. Additional experiments are involved and make the algorithm inefficient in practice.To address these challenges, this paper consists of: Firstly, a novel FAIRC algorithm based on AIRC is proposed. FAIRC take both network usage and CPU overhead into consideration. Secondly, we designed and implemented FAIRC in Xen and Linux environment. At last, experiments in real test bed are implemented to test performance of those algorithms.Experimental result shows FAIRC can maintain no less than 900MBps(90%of the upper bound) as bandwidth.