To coalesce or not to coalesce

System performance of Gigabit network hosts can be severely degraded due to interrupt overhead caused by heavy incoming traffic. One of the most popular solutions to mitigate such overhead is interrupt coalescing in which a single interrupt is generated for multiple incoming packets. This is opposed to normal interruption in which an interrupt is generated for every incoming packet. In this paper we investigate the performance of interrupt coalescing analytically and compare it with that of normal interruption. We consider two types of coalescing (viz. count-based and time-based). The system performance is studied in terms of throughput, CPU availability for user applications, latency and packet loss.