A fast flexible simulation platform for multi-core systems
Gavara, Phanendra Sandeep Naidu
MetadataShow full item record
As the popularity and demand grow, high-performance systems are adopting more processing cores. IBM Roadrunner, a petascale supercomputer, has 130K+ cores. No wonder, designing and/or analyzing such a gigantic system is very complicated and it costs vast amount of money. Modeling and simulation is attracting researches from many organizations and different disciplines to conduct research on expensive multi-core systems in an inexpensive way. Various multicore simulators have been proposed with various specific purposes. For example, MIT Hornet multicore simulator provides cycle-accurate performance of multicore chips. We feel the need for a fast flexible simulation platform capable of guiding to select the right system parameters (like number of cores) for optimal performance/power ratio when a set of applications is given. The proposed multicore simulation platform takes various system parameters as inputs during the runtime. It is able to analyze the sequential and parallel executions of the target workload. It generates detailed log-file indicating operations and system status for the entire execution time. This simulation results are evaluated using traditional check-point method and comparing with results published in [Chaturvedula, 2011]. There are opportunities to add functionalities to the proposed multicore simulation platform to facilitate in-depth analysis of various crucial multicore issues including core allocation and cache optimization.
Thesis (M.S.)--Wichita State University, College of Engineering, Dept. of Electrical Engineering and Computer Science