Authors: Sapna Prabhu; Rohin Daruwala
Addresses: Department of Electrical Engineering, Veermata Jijabai Technological Institute (VJTI), Mumbai, India ' Department of Electrical Engineering, Veermata Jijabai Technological Institute (VJTI), Mumbai, India
Abstract: In the age of high-performance computing, multi-core technology is seen to bridge the gap between good system throughput and increased power dissipation. Most multi-core architectures share hardware resources between cores, leading to cross-core interference which is non-trivial in many cases. Effective shared resource management is necessary to minimise the performance variability of threads due to cross-core interference as well as optimise resource usage. Many biological models have been proposed which model interactions between species. This paper uses a mathematical model, on the lines of the competitive Lotka-Volterra model for biological population growth, which models performance of the cores in a shared-memory multi-core system. It further formulates stability conditions for the multi-core system taking into account the workload characteristics and derives partition sizes to partition the shared cache to reduce the effect of cross-core interference for different workload types. Analysis of scalability of the model is also presented. Performance evaluations reveal that the LV-based mathematical model can be used to study the effects of resource sharing between cores, calculate partition sizes and is scalable.
Keywords: workload characterisation; inter-specific competition; stability; cross-core interference; resource management; LV competition model; scalability; cache partitioning; shared resources; multi-core systems; high-performance computing; hardware resources; mathematical modelling; competitive Lotka-Volterra model; shared memory; partition sizes; performance evaluation.
International Journal of High Performance Computing and Networking, 2016 Vol.9 No.3, pp.206 - 217
Available online: 13 Apr 2016 *Full-text access for editors Access for subscribers Purchase this article Comment on this article