Adopt a better IP lookup algorithm to achieve 100+ Gbps throughput #365
Description
Currently, we are using a single global LPM table in GK blocks, which uses a variation of the DIR-24-8 algorithm. Through the evaluation in the paper titled DXR: Towards a Billion Routing Lookups per Second in Software, we know that the throughput of DIR-24-8 flattens as the number of parallel worker threads grows: the large amount of memory used causes frequent accesses to off-chip DRAM which saturates with parallel worker thread.
To achieve 100+ Gbps (100+ Mpps) throughput on a single server, we may have to adopt SAIL or Poptrie.