include "../eng/header.html" ?>
H. Gustafsson, F. Svensson, L. Abeni, T. Cucinotta. "State Synchronization for Fault-Tolerant Cloud RAN Scheduling: A Design Space Exploration," in Proceedings of the 5th International Workshop on Real-Time and Cyber-Physical Cloud (RT-Cloud 2026), 38th European Conference on Real-Time Systems (ECRTS 2026), July 7th, 2026, Lund, Sweden. Best Paper Award
Cloud Radio Access Network (RAN) deployments move baseband processing onto commercial off-the-shelf servers managed by cloud orchestration stacks. The MAC scheduler - the most latency-sensitive component in the baseband - is a stateful service updating its internal state thousands of times per second. Since it is required to recover from faults within tens of milliseconds, it must be able to easily synchronize its internal state with one or more replicas. This paper explores the problem of state synchronization in a Radio Access Network (RAN) scheduler considering three different aspects: change detection (so that only the fractions of the state that actually changed can be sent to other nodes), state transfer, and redundancy model. Several techniques for performing these three operations have been tested and compared, with different combinations implemented in a scheduler mock-up that faithfully reproduces the state layout and workload of a 5G NR or a potential 6G RAN scheduler, allowing evaluation of per-TTI synchronization overhead, dirty-data volume, and end-to-end failover latency. The experimental results show that the granularity of change detection and the latency of state transfer dominate the overhead of state synchronization. On the other hand, the choice of redundancy model primarily affects failover time rather than steady-state cost.
Download paper include "../eng/tail.html" ?>