High-speed protocol processing using parallel architectures

The authors study the effectiveness of different parallel architectures for achieving high throughputs necessary for processing signaling traffic in high speed networks. They consider shared memory and distributed memory parallel architectures for processing signaling messages. A key performance issue is the trade-off between load balancing gains and call record management overhead; arranging processors in large groups potentially yields higher load balancing gains but also incurs higher overhead in maintaining consistency amongst the replicated copies of the call records. They study this tradeoff and its impact on the choice of optimal parallel architectures for protocol processing. The results show that for shared memory architectures, which provide the maximal load balancing gains, organizing the processors in small groups optimizes the performance for a wide range of traffic loads. For distributed memory architectures, which do not inherently provide any load balancing, organizing the processors into small groups and using a simple distributed load balancing scheme yields modest performance gains even after call record management overheads are taken into account. A good architecture is a hybrid one using a distributed architecture in which each node is a small processing group with shared memory