Modern high-performance computer architectures can have tens of millions of processor cores and memory and interprocessor communication subsystems arranged in deep hierarchical architectures. When a parallel program is broken into even larger number of executing threads, processors that are heavily using a given data must be placed close to that data. Further, heavily interacting threads must be kept close to one another in that hierarchy to avoid excessive processor communications that can result in sever latencies and energy consumption, thereby exploiting the underlying locality properties of the executing applications.
Mr. Ahmad Anbar's dissertation, titled "Exploiting Hierarchical Locality for Extreme Scale Architectures" presents an innovative solution based on graph theory for productive extreme-scale high performance computing. The core of his work is to automatically discover and exploit hierarchical locality based on the characteristic of modern programming models, without the need for intervention by application developers. The work has been prototyped on George, a Cray XK7 supercomputer in HPCL and has shown substantial improvements over the state of the art.
After his graduation, Ahmed has jointed Amazon Web Services (AWS), where he is developing solutions for optimizing the builds of Simple Storage Service (S3), the largest-scale object storage service of any cloud infrastructure provider. The research of Ahmad Anbar has resulted in 20+ publications at top journals and conferences such as TACO, PGAS, CCGrid, ICPADS and HPCC.