| Summary: | Abstract This paper proposes the concept of compute resource disaggregation in centralized softwarised radio access networks. This approach allows individual allocation of processing functions to different servers depending on the nature and volume of their processing requirements. The benefits of the proposed approach are evaluated through experiments analyzing the baseband unit (BBU) processing requirements of LTE PHY using an open source suite for wireless systems and a purposely developed multistage linear programming modeling framework. To address the high computational complexity of integer linear programming (ILP), limiting real-time services, a heuristic algorithm was also developed. Our results show that when using the proposed approach, the overall system power consumption is reduced by 40% under high loading scenarios compared to the traditional solution where all BBU functions are hosted in the same physical servers. The heuristic results achieve relatively good agreement with the ILP results (the gap is less than 5%), particularly for low and high loading conditions. At the same time, the heuristic requires less than 0.3 s to identify the optimal allocation policy.
|
|---|
