PlanB: Optimizing the IPv6 LPM Algorithm with AVX-512
Discover how PlanB revolutionizes the IPv6 LPM lookup algorithm, improving performance and efficiency.
The growing demand for speed and efficiency in modern networks has pushed developers to innovate lookup algorithms for IPv6. A recent significant advance is PlanB, an algorithm that optimizes the Longest Prefix Match (LPM) lookup for IPv6, using a linearized B+-tree and exploiting AVX-512 instructions to improve performance.
Innovation in the LPM Algorithm for IPv6
Traditionally, LPM lookup algorithms for IPv6 face challenges related to handling variable-length prefixes and the need for efficient lookup operations. PlanB addresses these challenges by transforming the two-dimensional LPM problem into a one-dimensional search over elementary intervals, unifying the search across prefix values and lengths. This innovative approach reduces computational complexity and improves the overall efficiency of the algorithm.
The Linearized B+-Tree Structure
To implement this unified search, PlanB adapts the linearized B+-tree structure to the specific needs of LPM. This flat-array-based structure allows lookup operations to be managed efficiently, reducing data access time and improving overall performance. Adopting this structure represents a significant step forward compared with traditional approaches, offering a more scalable and high-performing solution for managing IPv6 prefixes.
Integration with AVX-512 for Advanced Performance
To maximize performance, PlanB integrates advanced techniques such as vectorization, batching, branch-free logic and loop unrolling, fully exploiting the parallelism of modern CPUs. In particular, the use of AVX-512 instructions allows more data to be processed in parallel, reducing processing times and increasing search speed. This approach is especially advantageous on fifth-generation AMD EPYC processors, which offer a native 512-bit data path for AVX-512, further improving performance compared with previous implementations based on 256-bit data paths.
Benchmarks and Concrete Results
PlanB's evaluations show exceptional performance: on a single core, the algorithm reaches 390 million lookups per second (MLPS) using real IPv6 FIBs on AMD processors, and scales up to 3.4 billion lookups per second (BLPS) on a 12-core system. These results outperform state-of-the-art software-based schemes such as PopTrie, CP-Trie, Neurotrie and HBS by 1.6× to 14×. Such improvements are crucial for applications that require fast, scalable processing, such as BGP routing in high-speed networks.
Implications for BGP Routing and Modern Networks
Improved performance in the LPM algorithm has a direct impact on BGP routing, since more efficient IPv6 prefix lookup can reduce convergence times and improve network stability. In addition, adopting PlanB can help reduce the power consumption of network devices, since more efficient operations require fewer computational resources. This is particularly relevant in a context where energy efficiency is a growing priority for network infrastructure.
Conclusion
PlanB represents a significant advance in the field of IPv6 LPM lookup algorithms, offering a more efficient and scalable solution for managing IPv6 prefixes. The combination of an innovative data structure with optimization for modern CPU architectures, including integration with AVX-512, makes PlanB an ideal choice for applications requiring high performance and reliability. Its implications for BGP routing and for the energy efficiency of modern networks are considerable, suggesting a potential large-scale positive impact in the telecommunications sector.
- Optimizing the LPM algorithm for IPv6: Transforming the two-dimensional LPM problem into a one-dimensional search over elementary intervals, improving computational efficiency.
- Adapting the linearized B+-tree structure: Implementing a flat-array-based data structure to manage lookup operations efficiently, reducing data access times.
- Integration with AVX-512 for advanced performance: Using AVX-512 instructions to process more data in parallel, fully exploiting modern CPU parallelism and improving performance on fifth-generation AMD EPYC processors.
- Benchmarks and concrete results: Exceptional performance with 390 MLPS on a single core and 3.4 BLPS on a 12-core system, outperforming state-of-the-art software-based schemes by 1.6× to 14×.
- Implications for BGP routing and modern networks: Improved convergence times and network stability, with potential energy-efficiency benefits for network infrastructure.