Lightweight Determinism in Large Scale Networks

In networking, determinism expresses the ability to deliver all data packets (no exceptions) to their intended destinations within tight time intervals. A growing variety of network applications and architectural paradigms require determinism: high-performance computing, power-grid control, industrial automation, and the disaggregation of mobile radio-access-network and core-network functions in edge-cloud data centers are just a few examples. End-to-end queuing latency is the primary metric that a network must control in order to realize determinism, since hard tight bounds on queuing latency also eliminate congestive packet losses and minimize latency jitter. IEEE and the IETF have been specifying traffic management techniques for tight control of the queuing latency that require new packet switching hardware and complex configuration procedures and also exhibit poor scalability and dynamicity. In this article we propose an alternative, lightweight approach to determinism that minimizes the impact on the network infrastructure and yet exceeds the performance of all techniques standardized so far.