A Study of the Impact of Network Traffic Pacing from Network and End-User Perspectives

As the demand on the Internet bandwidth keeps increasing, all-optical network architectures emerge as a promising solution to high-speed telecommunication networks. However, the performance of optical routers/switches is sensitive to the statistics of Internet traffic due to their limited buffer size, which weakens the optical routers' capability of absorbing Internet traffic surges (known as burstiness), resulting in a high packet drop rate and, hence, lowering the quality of service. As a natural solution to the Internet traffic burstiness, pacing has long been studied and a variety of pacing techniques have been proposed in the literature. Nevertheless, the effect of pacing has not been conclusive. In this paper, we aim to answer the following questions: (1) in what situation pacing is beneficial, (2) does pacing harm short-lived flows, (3) does pacing affect the established fairness? We adopt in our simulation a newly proposed queue length-based pacing technique as a cure to the burstiness. The experimental results show that with appropriate parameter settings, pacing (1) significantly improves the performance of a moderately loaded small-buffer network, (2) brings little negative impact on short-lived flows in a nearly drop-free environment, and (3) is able to maintain the same fairness as achieved on the current Internet. These observations are of significant importance to a potentially wide adoption of small-buffer all-optical networks.