Understanding the behavior of Pthread applications on non-uniform cache architectures

Why is it important? As number of cores in a processor scale up, caches would become banked Keeps individual look-up time small. Allows parallel accesses by different cores. Present shared programming model assumes a flat memory. Unaware application can have sub-optimal performance Conclusion Programming model needs to change For any heterogeneous memory hierarchy. Architecture, OS, compiler and application developer should work together Significant performance gains can be achieved. ? Without increasing system complexity. As complexity of memory hierarchy grows, optimizations like these will be critical.