CPU cycles per packet (CPP)

In order to measure the Central Processing Unit (CPU) cycles per packet (CPP) spent in each network stack component, we first use the Linux perf tool [37] to count the total CPU cycles consumed in a 60-second packet transmission ($Cycl{e}_{total}$), which is repeated 5 times. We also use perf to trace the function calls and measure the percentage of the overall CPU cycles spent in the corresponding function ($Cycl{e}_{percentage}$). With the total number of packets sent in a 60-second packet transmission ($N_{packet}$), we can calculate the CPP of a specific function call as follows: $CPP=\frac{Cycl{e}_{total}}{N_{packet}\times Cycl{e}_{percentage}}$

Reference List

1. Qi, S., Kulkarni, S. G., & Ramakrishnan, K. K. (2020). Assessing container network interface plugins: Functionality, performance, and scalability. IEEE Transactions on Network and Service Management, 18(1), 656-671.