This paper presents the concept of virtual DBA, a method we propose to virtualize upstream capacity scheduling in PONs, which provides multiple independent virtual network operators with the ability to precisely schedule their upstream traffic allocation. After a brief introduction on the evolution of access network sharing, we present our virtual DBA architecture, detailing its main components. We then provide a summary of the work done in this area both from theoretical and practical implementation perspectives. In this paper, we propose a novel stateless algorithm for merging multiple independent virtual bandwidth maps based on priority classes and analyze its performance in terms of efficiency of capacity allocation and latency. Through our results we discuss the existence of a trade-off between traffic load and grant size distribution vs. efficiency and latency. We find that, differently from residential single-tenant application, when PONs are used for low latency and multi-tenant applications, the system has better overall performance if grants are allocated in small size. In addition, our analysis shows that for high priority, strict latency services, our proposed merging algorithm presents delay performance that is independent of the traffic distribution considered.

We propose and compare algorithms to allocate upstream PON capacity, where multiple virtual operators generate independent frame-level allocation over shared infrastructure. Our fragmentation-based approach shows the ability to limit latency increase to a few microseconds.

We present a stateful DBA hypervisor for shared PONs capable of meeting specific flow-level service level agreements, supporting services requiring strict latency and high availability. We report considerable improvement compared to standard stateless priority-based mechanisms.

No abstract