On task tree executor architectures based on intel parallel building blocks

Abstract

Our aim was to optimize a SOA control system by evolving the architecture of the service component that transforms system models into task trees, which are then executed by the runtime library called the Task Tree Executor, TTE. In the paper we present the two novel TTE architectures that evolved from the previous TTE architecture and introduced finer grained parallelism. The novel architectures execute TTE tasks as more lightweight TBB tasks and Cilk strands rather than the OS threads, which was the case for the previous TTE architecture. The experimental evaluation based on time needed for TTE reliability estimation, by statistical usage tests, shows that these novel TTE architectures are providing the average relative speedup, RS, from 8x to 11x, over the original TTE, on a dual-core machine. Additional experiments made on eight-core machine showed that RS provided by TTE based on TBB scales perfectly, and goes up to 77x. [Projekat Ministarstva nauke Republike Srbije, br. III 44009 and br. TR 32031]