Funded Projects

Characterization, Evaluation and Development of High Performance Network Services on Multi-Core Architectures.

Principal Investigator’s Organization (PIO):
Al-Khawarizmi Institute of Computer Science, UET Lahore
Principal Investigator (PI):
Dr. Waqar Mahmood, Dr. Abdul Waheed
Project Details:
Start Date 01-Jan-2008
Duration 18 months
Budget PKR 11.76 million
Status Project Successfully Completed
Progress Report View Progress Report
Publications View Publications
Thematic Area Telecommunication
Project Website
Executive Summary

Multi-core systems are coming into the mainstream and the demand for high performance networking is on the increase. There are multiple research efforts in individual area of parallel and multi-core architecture, high performance networks, and performance evaluation. However, hardly any research effort combines all three areas. This aspect is gaining important due to the technological trends in processor and network architectures. There is a need for micro-kernels that target multi-core processor and high performance networking architecture to help attain their optimized performance. This project is focused on researching and developing micro-kernels targeting multi-core processor architecture as well as high performance networking architecture to leverage their high performance capabilities. The key objectives of this technical project are: • Development of a benchmarking suite for multi-core systems by integrating and automating the developed micro-benchmarks with appropriate experimental design. • Establishment of a platform for research on multi-core processor performance characterization and evaluation of CPU-memory subsystem performance disparity for high-throughput networking applications. Such a facility is expected to be the first of its kind in Pakistan to support research and development in high performance networking. • Development of expertise in the area of High Performance Computing and High Performance Networking that will provide a level playing field to collaborate and contribute in this groundbreaking research and produce quality publications. The project initiated with the research and learning of the state-of the art micro-benchmarks for CPU and Networks. The memory and networking micro-benchmarks for multi-core system were developed followed by setting up of a state-of-the-art multi-core processor based system as well as 1Gbps networking equipment. These micro-benchmarks help evaluate the performance of cache and memory subsystem, CPU and high performance networks, to identify and minimize performance gaps and bottlenecks to attain sustainable maximum performance of multi-core systems and high performance networks collectively. This measurement based study provides the basis for efficiently parallelizing the network applications with memory, cache, and interconnection subsystem constraints of the target multi-core systems. Lastly these micro-benchmarks are tuned for 1Gbps networks and multi-core systems and performance measurement based research will be conducted using these micro-benchmarks to identify and analyze performance gaps and bottlenecks. A benchmarking suite for multi-core systems is also developed by integrating the developed micro-benchmarks. The key benefits of this project are given below: • The tools developed as a result of this project replace, existing, multiple server solutions with a single multi-core processor based products at approximately less than half the cost with no server administration requirement. This product replaces the existing software based traffic generators. • The benchmarks developed, as a result of this project, for multi-core system can be used by end users to evaluate the performance baseline for a system and compare with other similar system for procurement decisions in IT departments. • The test-bed setup, developed as an outcome of this project can be used for further academic research and development as well as service provisioning to industry or organizations that require system and platform tuning, performance evaluation and profiling, and application tuning and parallelization.