Using Traffic Shaping to Improve Performance for Raw Data Transfer and Video Streaming

Abstract

As networks and data transmission are becoming increasingly important for computational science, we are working on developing a visualization application that uses high-speed networks to aid scientists in their scientific exploration. However, we have identified a major issue in the performance of data transmission when rate-based UDP is used in parallel from multiple senders to a single receiver. This affects both video and raw data transmission. In this paper we identify the source of the issue in the traffic bursts caused by uneven pacing of data at the senders. Our experiments confirm that adding precise pacing at the data transmission source addresses the issue, improving the performance of data transfer and video streaming.