Intro; Preface; Contents; Acronyms; 1 Introduction; 1.1 Improving Memory Locality with Sharing-Aware Mapping; 1.2 Monitoring Memory Accesses for Sharing-Aware Mapping; 1.3 Organization of the Text; 2 Sharing-Aware Mapping and Parallel Architectures; 2.1 Understanding Memory Locality in Shared Memory Architectures; 2.1.1 Lower Latency When Sharing Data; 2.1.2 Reduction of the Impact of Cache Coherence Protocols; 2.1.3 Reduction of Cache Misses; 2.1.3.1 Capacity Misses; 2.1.3.2 Invalidation Misses; 2.1.3.3 Replication Misses; 2.1.4 Reduction of Memory Accesses to Remote NUMA Nodes

2.1.5 Better Usage of Interconnections2.2 Example of Shared Memory Architectures Affected by Memory Locality; 2.2.1 Intel Harpertown; 2.2.2 Intel Nehalem/Sandy Bridge; 2.2.3 AMD Abu Dhabi; 2.2.4 Intel Montecito/SGI NUMAlink; 2.3 Locality in the Context of Network Clusters and Grids; 3 Sharing-Aware Mapping and Parallel Applications; 3.1 Parallel Applications and Sharing-Aware Thread Mapping; 3.1.1 Considerations About the Sharing Pattern; 3.1.1.1 Dimension of the Sharing Pattern; 3.1.1.2 Granularity of the Sharing Pattern (Memory Block Size)

3.1.1.3 Data Structure Used to Store the Thread Sharers of a Memory Block3.1.1.4 History of the Sharing Pattern; 3.1.2 Sharing Patterns of Parallel Applications; 3.1.3 Varying the Granularity of the Sharing Pattern; 3.1.4 Varying the Number of Sharers of the Sharing Pattern; 3.2 Parallel Applications and Sharing-Aware Data Mapping; 3.2.1 Parameters that Influence Sharing-Aware Data Mapping; 3.2.1.1 Influence of the Memory Page Size; 3.2.1.2 Influence of Thread Mapping; 3.2.2 Analyzing the Data Mapping Potential of Parallel Applications

3.2.3 Influence of the Page Size on Sharing-Aware Data Mapping3.2.4 Influence of Thread Mapping on Sharing-Aware Data Mapping; 4 State-of-the-Art Sharing-Aware Mapping Methods; 4.1 Sharing-Aware Static Mapping; 4.1.1 Static Thread Mapping; 4.1.2 Static Data Mapping; 4.1.3 Combined Static Thread and Data Mapping; 4.2 Sharing-Aware Online Mapping; 4.2.1 Online Thread Mapping; 4.2.2 Online Data Mapping; 4.2.3 Combined Online Thread and Data Mapping; 4.3 Discussion on Sharing-Aware Mapping and the State-of-Art; 4.4 Improving Performance with Sharing-Aware Mapping; 4.4.1 Mapping Mechanisms

4.4.1.1 Source Code Changes4.4.1.2 Offline Profiling at the User Level; 4.4.1.3 Runtime Options; 4.4.1.4 Online Profiling at the System Level; 4.4.2 Methodology of the Experiments; 4.4.3 Results; 5 Conclusions; References