Cross-Layer Pacing for Predictably Low Latency

Abstract

Low-latency networking has become a major research domain as to the increasing demand for wireless applications. A predictably low age of information is key to such applications (i.e., augmented reality, smart factories, and edge computing infrastructures) as they need to operate on the latest data. Despite variable channel properties (i.e., non-deterministic interferences) it must be ensured that especially end-to-end communication operates with lowest latency that is possible. To enable low-latency networking we propose X-PACE, a cross-layer pacing scheme. X-PACE provides low-latency communication under consideration of dynamics from the network protocol layers up to the application level. Bottleneck detection and mitigation techniques of X-PACE provide predictability and improve the end-to-end communication latency. Our evaluation demonstrates that X-PACE reduces the end-to-end tail latency by up to 54% and narrows the latency range by up to 91 %.