Techniques for Shared Resource Management in Systems with Throughput Processors
share
The continued growth of the computational capability of throughput processors has made throughput processors the platform of choice for a wide variety of high performance computing applications. Graphics Processing Units (GPUs) are a prime example of throughput processors that can deliver high performance for applications ranging from typical graphics applications to general-purpose data parallel (GPGPU) applications. However, this success has been accompanied by new performance bottlenecks throughout the memory hierarchy of GPU-based systems. We identify and eliminate performance bottlenecks caused by major sources of interference throughout the memory hierarchy. We introduce changes to the memory hierarchy for systems with GPUs that allow the memory hierarchy to be aware of both CPU and GPU applications' characteristics. We introduce mechanisms to dynamically analyze different applications' characteristics and propose four major changes throughout the memory hierarchy. We propose changes to the cache management and memory scheduling mechanisms to mitigate intra-application interference in GPGPU applications. We propose changes to the memory controller design and its scheduling policy to mitigate inter-application interference in heterogeneous CPU-GPU systems. We redesign the MMU and the memory hierarchy in GPUs to be aware of ddress-translation data in order to mitigate the inter-address-space interference. We introduce a hardware-software cooperative technique that modifies the memory allocation policy to enable large page support in order to further reduce the inter-address-space interference at the shared Translation Lookaside Buffer (TLB). Our evaluations show that the GPU-aware cache and memory management techniques proposed in this dissertation are effective at mitigating the interference caused by GPUs on current and future GPU-based systems.
READ FULL TEXT
