Leveraging Programmable Data Plane For Compressing Forwarding Tables

The Forwarding Information Base (FIB) resides in the data plane of a routing device and is used to forward packets to a next-hop, based on packets\u27 destination IP addresses. The constant growth of a FIB forces network operators to spend more resources on maintaining memory with line-rate Longest Prefix Match (LPM) lookup in a FIB, namely, expensive and energy-hungry Ternary Content-Addressable Memory (TCAM) chips. In this work, we review two different approaches used to mitigate the FIB overflow problem. First, we investigate FIB aggregation, i.e., merging adjacent or overlapping routes with the same next-hop while preserving the forwarding behavior of a FIB. We propose a near-optimal algorithm, FIB Aggregation with Quick Selections (FAQS), that minimizes the FIB churn and speeds BGP update processing by more than twice. In the meantime, FAQS preserves a high compression ratio (at most 73\%). FAQS handles BGP updates incrementally, without the need of re-aggregating the entire FIB table. Second, we investigate FIB (or route) caching, when TCAM holds only a portion of a FIB that carries most of the traffic. We leverage the emerging concept of the programmable data plane to propose a Programmable FIB Caching Architecture (PFCA), that allows cache-victim selection at the line rate and significantly reduces the FIB churn compared to FIB aggregation. PFCA achieves 99.8% cache-hit ratio with only 3.3\% of the FIB placed in a FIB cache. Finally, we extend PFCA\u27s design with a novel approach of integrating incremental FIB aggregation and FIB caching. Such integration needed to overcome cache hiding challenge when a less specific prefix in a cache hides a more specific prefix in a secondary FIB table, which leads to incorrect LPM matching at the cache. In Combined FIB Caching and Aggregation (CFCA), cache-hit ratio is maximized up to 99.94% with only 2.5\% entries of the FIB, while the total number of route changes in TCAM is reduced by more than 40\% compared to low-churn FIB aggregation techniques