IBM J. Res. Dev

Experience with building a commodity intel-based ccNUMA system

View publication


A 16 way cache-coherent nonuniform memory access (ccNUMA) Intel system consisting of four commodity four-processor Fujitsu Teamserver SMPs connected by a Synfinity cache-coherent switch was built. Results from a performance-evaluation study confirm the success of the combined hardware/software approach for performance tuning in computation-intensive workloads. The results also show that the poor local-memory bandwidth in the commodity Intel-based systems is often the main contributor to poor scalability and performance.