Publications by Adwait Jog
2018
Proceedings of the 23rd International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS), Williamsburg, VA, USA, March 2018
Graphics Processing Units (GPUs) exploit large amounts of threadlevel parallelism to provide high instruction throughput and to efficiently hide long-latency stalls. The resulting high throughput, along with continued programmability improvements, have made GPUs an essential computational resource in many domains. Applications from different domains can have vastly different compute and memory demands on the GPU. In a large-scale computing environment, to efficiently accommodate such wide-ranging demands without leaving GPU resources underutilized, multiple applications can share a single GPU, akin to how multiple applications execute concurrently on a CPU. Multi-application concurrency requires several support mechanisms in both hardware and software. One such key mechanism is virtual memory, which manages and protects the address space of each application. However, modern GPUs lack the extensive support for multi-application concurrency available in CPUs, and as a result suffer from high performance overheads when shared by multiple applications, as we demonstrate. We perform a detailed analysis of which multi-application concurrency support limitations hurt GPU performance the most. We find that the poor performance is largely a result of the virtual memory mechanisms employed in modern GPUs. In particular, poor address translation performance is a key obstacle to efficient GPU sharing. State-of-the-art address translation mechanisms, which were designed for single-application execution, experience significant inter-application interference when multiple applications spatially share the GPU. This contention leads to frequent misses in the shared translation lookaside buffer (TLB), where a single miss can induce long-latency stalls for hundreds of threads. As a result, the GPU often cannot schedule enough threads to successfully hide the stalls, which diminishes system throughput and becomes a first-order performance concern. Based on our analysis, we propose MASK, a new GPU framework that provides low-overhead virtual memory support for the concurrent execution of multiple applications. MASK consists of three novel address-translation-aware cache and memory management mechanisms that work together to largely reduce the overhead of address translation: (1) a token-based technique to reduce TLB contention, (2) a bypassing mechanism to improve the effectiveness of cached address translations, and (3) an application-aware memory scheduling scheme to reduce the interference between address translation and data requests. Our evaluations show that MASK restores much of the throughput lost to TLB contention. Relative to a state-of-the-art GPU TLB, MASK improves system throughput by 57.8%, improves IPC throughput by 43.4%, and reduces applicationlevel unfairness by 22.4%. MASK’s system throughput is within 23.2% of an ideal GPU system with no address translation overhead.
@inproceedings{abc,
abstract = {Graphics Processing Units (GPUs) exploit large amounts of threadlevel parallelism to provide high instruction throughput and to efficiently hide long-latency stalls. The resulting high throughput, along with continued programmability improvements, have made GPUs an essential computational resource in many domains. Applications from different domains can have vastly different compute and memory demands on the GPU. In a large-scale computing environment, to efficiently accommodate such wide-ranging demands without leaving GPU resources underutilized, multiple applications can share a single GPU, akin to how multiple applications execute concurrently on a CPU. Multi-application concurrency requires several support mechanisms in both hardware and software. One such key mechanism is virtual memory, which manages and protects the address space of each application. However, modern GPUs lack the extensive support for multi-application concurrency available in CPUs, and as a result suffer from high performance overheads when shared by multiple applications, as we demonstrate. We perform a detailed analysis of which multi-application concurrency support limitations hurt GPU performance the most. We find that the poor performance is largely a result of the virtual memory mechanisms employed in modern GPUs. In particular, poor address translation performance is a key obstacle to efficient GPU sharing. State-of-the-art address translation mechanisms, which were designed for single-application execution, experience significant inter-application interference when multiple applications spatially share the GPU. This contention leads to frequent misses in the shared translation lookaside buffer (TLB), where a single miss can induce long-latency stalls for hundreds of threads. As a result, the GPU often cannot schedule enough threads to successfully hide the stalls, which diminishes system throughput and becomes a first-order performance concern. Based on our analysis, we propose MASK, a new GPU framework that provides low-overhead virtual memory support for the concurrent execution of multiple applications. MASK consists of three novel address-translation-aware cache and memory management mechanisms that work together to largely reduce the overhead of address translation: (1) a token-based technique to reduce TLB contention, (2) a bypassing mechanism to improve the effectiveness of cached address translations, and (3) an application-aware memory scheduling scheme to reduce the interference between address translation and data requests. Our evaluations show that MASK restores much of the throughput lost to TLB contention. Relative to a state-of-the-art GPU TLB, MASK improves system throughput by 57.8\%, improves IPC throughput by 43.4\%, and reduces applicationlevel unfairness by 22.4\%. MASK{\textquoteright}s system throughput is within 23.2\% of an ideal GPU system with no address translation overhead.},
author = {Rachata Ausavarungnirun and Vance Miller and Joshua Landgraf and Saugata Ghose and Jayneel Gandhi and Adwait Jog and Christopher Rossbach and Onur Mutlu},
booktitle = {Proceedings of the 23rd International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS)},
title = {MASK: Redesigning the GPU Memory Hierarchy to Support Multi-Application Concurrency},
venue = {Williamsburg, VA, USA},
year = {2018}
}
2016
49th Annual IEEE/ACM International Symposium on Microarchitecture, MICRO 2016, Taipei, Taiwan, October 2016
@inproceedings{abc,
author = {Nandita Vijaykumar and Kevin Hsieh and Gennady Pekhimenko and Samira Manabi Khan and Ashish Shrestha and Saugata Ghose and Adwait Jog and Phillip B. Gibbons and Onur Mutlu},
booktitle = {49th Annual IEEE/ACM International Symposium on Microarchitecture, MICRO 2016, Taipei, Taiwan},
title = {Zorua: A holistic approach to resource virtualization in GPUs.},
url = {http://dx.doi.org/10.1109/MICRO.2016.7783718},
year = {2016}
}
Proceedings of the 2016 International Conference on Parallel Architectures and Compilation, PACT 2016, Haifa, Israel, September 2016
@inproceedings{abc,
author = {Ashutosh Pattnaik and Xulong Tang and Adwait Jog and Onur Kayiran and Asit K. Mishra and Mahmut T. Kandemir and Onur Mutlu and Chita R. Das},
booktitle = {Proceedings of the 2016 International Conference on Parallel Architectures and Compilation, PACT 2016, Haifa, Israel},
title = {Scheduling Techniques for GPU Architectures with Processing-In-Memory Capabilities.},
url = {http://doi.acm.org/10.1145/2967938.2967940},
year = {2016}
}
Proceedings of the 2016 International Conference on Parallel Architectures and Compilation, PACT 2016, Haifa, Israel, September 2016
@inproceedings{abc,
author = {Onur Kayiran and Adwait Jog and Ashutosh Pattnaik and Rachata Ausavarungnirun and Xulong Tang and Mahmut T. Kandemir and Gabriel H. Loh and Onur Mutlu and Chita R. Das},
booktitle = {Proceedings of the 2016 International Conference on Parallel Architectures and Compilation, PACT 2016, Haifa, Israel},
title = {{\^I}{\textonequarter}C-States: Fine-grained GPU Datapath Power Management.},
url = {http://doi.acm.org/10.1145/2967938.2967941},
year = {2016}
}
Proceedings of the 2016 ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Science, Antibes Juan-Les-Pins, France, June 2016
@inproceedings{abc,
author = {Adwait Jog and Onur Kayiran and Ashutosh Pattnaik and Mahmut T. Kandemir and Onur Mutlu and Ravishankar Iyer and Chita R. Das},
booktitle = {Proceedings of the 2016 ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Science, Antibes Juan-Les-Pins, France},
title = {Exploiting Core Criticality for Enhanced GPU Performance.},
url = {http://doi.acm.org/10.1145/2896377.2901468},
year = {2016}
}
CoRR, January 2016
Modern Graphics Processing Units (GPUs) are well provisioned to support the concurrent execution of thousands of threads. Unfortunately, different bottlenecks during execution and heterogeneous application requirements create imbalances in utilization of resources in the cores. For example, when a GPU is bottlenecked by the available off-chip memory bandwidth, its computational resources are often overwhelmingly idle, waiting for data from memory to arrive.
This work describes the Core-Assisted Bottleneck Acceleration (CABA) framework that employs idle on-chip resources to alleviate different bottlenecks in GPU execution. CABA provides flexible mechanisms to automatically generate "assist warps" that execute on GPU cores to perform specific tasks that can improve GPU performance and efficiency.
CABA enables the use of idle computational units and pipelines to alleviate the memory bandwidth bottleneck, e.g., by using assist warps to perform data compression to transfer less data from memory. Conversely, the same framework can be employed to handle cases where the GPU is bottlenecked by the available computational units, in which case the memory pipelines are idle and can be used by CABA to speed up computation, e.g., by performing memoization using assist warps.
We provide a comprehensive design and evaluation of CABA to perform effective and flexible data compression in the GPU memory hierarchy to alleviate the memory bandwidth bottleneck. Our extensive evaluations show that CABA, when used to implement data compression, provides an average performance improvement of 41.7% (as high as 2.6X) across a variety of memory-bandwidth-sensitive GPGPU applications.
@article{abc,
abstract = {Modern Graphics Processing Units (GPUs) are well provisioned to support the concurrent execution of thousands of threads. Unfortunately, different bottlenecks during execution and heterogeneous application requirements create imbalances in utilization of resources in the cores. For example, when a GPU is bottlenecked by the available off-chip memory bandwidth, its computational resources are often overwhelmingly idle, waiting for data from memory to arrive.
This work describes the Core-Assisted Bottleneck Acceleration (CABA) framework that employs idle on-chip resources to alleviate different bottlenecks in GPU execution. CABA provides flexible mechanisms to automatically generate "assist warps" that execute on GPU cores to perform specific tasks that can improve GPU performance and efficiency.
CABA enables the use of idle computational units and pipelines to alleviate the memory bandwidth bottleneck, e.g., by using assist warps to perform data compression to transfer less data from memory. Conversely, the same framework can be employed to handle cases where the GPU is bottlenecked by the available computational units, in which case the memory pipelines are idle and can be used by CABA to speed up computation, e.g., by performing memoization using assist warps.
We provide a comprehensive design and evaluation of CABA to perform effective and flexible data compression in the GPU memory hierarchy to alleviate the memory bandwidth bottleneck. Our extensive evaluations show that CABA, when used to implement data compression, provides an average performance improvement of 41.7\% (as high as 2.6X) across a variety of memory-bandwidth-sensitive GPGPU applications.},
author = {Nandita Vijaykumar and Gennady Pekhimenko and Adwait Jog and Saugata Ghose and Abhishek Bhowmick and Rachata Ausavarungnirun and Chita R. Das and Mahmut T. Kandemir and Todd C. Mowry and Onur Mutlu},
journal = {CoRR},
title = {A Framework for Accelerating Bottlenecks in GPU Execution with Assist Warps.},
url = {http://arxiv.org/abs/1602.01348},
year = {2016}
}
2015
Proceedings of the 42nd Annual International Symposium on Computer Architecture, Portland, OR, USA, June 2015
Modern Graphics Processing Units (GPUs) are well provisioned to support the concurrent execution of thousands of threads. Unfortunately, diUerent bottlenecks during execution and heterogeneous application requirements create imbalances in utilization of resources in the
cores. For example, when a GPU is bottlenecked by the available on-chip memory bandwidth, its computational resources are often overwhelmingly idle, waiting for data from memory to arrive. This paper introduces the Core-Assisted Bottleneck Acceleration (CABA) framework that employs idle on-chip resources to alleviate different bottlenecks in GPU execution. CABA provides flexible mechanisms to automatically generate “assist warps” that execute on GPU cores to perform specific tasks that can improve GPU performance and
efficiency. CABA enables the use of idle computational units and pipelines to alleviate the memory bandwidth bottleneck, e.g., by using assist warps to perform data compression to transfer less data from memory. Conversely, the same framework can be employed to handle cases where the GPU is bottlenecked by the available computational units, in which case the memory pipelines are idle and can be used by CABA to speed up computation, e.g., by performing memoization using assist warps. We provide a comprehensive design and evaluation of CABA to perform effective and flexible data compression in the GPU memory hierarchy to alleviate the memory bandwidth bottleneck. Our extensive evaluations show that CABA, when used to implement data compression, provides an average performance improvement of 41.7% (as high as 2.6X) across a variety of memory-bandwidth-sensitive GPGPU applications.
@inproceedings{abc,
abstract = {Modern Graphics Processing Units (GPUs) are well provisioned to support the concurrent execution of thousands of threads. Unfortunately, diUerent bottlenecks during execution and heterogeneous application requirements create imbalances in utilization of resources in the
cores. For example, when a GPU is bottlenecked by the available on-chip memory bandwidth, its computational resources are often overwhelmingly idle, waiting for data from memory to arrive. This paper introduces the Core-Assisted Bottleneck Acceleration (CABA) framework that employs idle on-chip resources to alleviate different bottlenecks in GPU execution. CABA provides flexible mechanisms to automatically generate {\textquotedblleft}assist warps{\textquotedblright} that execute on GPU cores to perform specific tasks that can improve GPU performance and
efficiency. CABA enables the use of idle computational units and pipelines to alleviate the memory bandwidth bottleneck, e.g., by using assist warps to perform data compression to transfer less data from memory. Conversely, the same framework can be employed to handle cases where the GPU is bottlenecked by the available computational units, in which case the memory pipelines are idle and can be used by CABA to speed up computation, e.g., by performing memoization using assist warps. We provide a comprehensive design and evaluation of CABA to perform effective and flexible data compression in the GPU memory hierarchy to alleviate the memory bandwidth bottleneck. Our extensive evaluations show that CABA, when used to implement data compression, provides an average performance improvement of 41.7\% (as high as 2.6X) across a variety of memory-bandwidth-sensitive GPGPU applications.},
author = {Nandita Vijaykumar and Gennady Pekhimenko and Adwait Jog and Abhishek Bhowmick and Rachata Ausavarungnirun and Chita R. Das and Mahmut T. Kandemir and Todd C. Mowry and Onur Mutlu},
booktitle = {Proceedings of the 42nd Annual International Symposium on Computer Architecture},
title = {A case for core-assisted bottleneck acceleration in GPUs: enabling flexible data compression with assist warps.},
url = {http://doi.acm.org/10.1145/2749469.2750399},
venue = {Portland, OR, USA},
year = {2015}
}
2014
47th Annual IEEE/ACM International Symposium on Microarchitecture, MICRO 2014, Cambridge, United Kingdom, December 2014
@inproceedings{abc,
author = {Onur Kayiran and Nachiappan Chidambaram Nachiappan and Adwait Jog and Rachata Ausavarungnirun and Mahmut T. Kandemir and Gabriel H. Loh and Onur Mutlu and Chita R. Das},
booktitle = {47th Annual IEEE/ACM International Symposium on Microarchitecture, MICRO 2014, Cambridge, United Kingdom},
title = {Managing GPU Concurrency in Heterogeneous Architectures.},
url = {http://dx.doi.org/10.1109/MICRO.2014.62},
year = {2014}
}
2013
The 40th Annual International Symposium on Computer Architecture, ISCA'13, Tel-Aviv, Israel, June 2013
@inproceedings{abc,
author = {Adwait Jog and Onur Kayiran and Asit K. Mishra and Mahmut T. Kandemir and Onur Mutlu and Ravishankar Iyer and Chita R. Das},
booktitle = {The 40th Annual International Symposium on Computer Architecture, ISCA{\textquoteright}13, Tel-Aviv, Israel},
title = {Orchestrated scheduling and prefetching for GPGPUs.},
url = {http://doi.acm.org/10.1145/2485922.2485951},
year = {2013}
}
Architectural Support for Programming Languages and Operating Systems, ASPLOS '13, Houston, TX, March 2013
@inproceedings{abc,
author = {Adwait Jog and Onur Kayiran and Nachiappan Chidambaram Nachiappan and Asit K. Mishra and Mahmut T. Kandemir and Onur Mutlu and Ravishankar Iyer and Chita R. Das},
booktitle = {Architectural Support for Programming Languages and Operating Systems, ASPLOS {\textquoteright}13, Houston, TX},
title = {OWL: cooperative thread array aware scheduling techniques for improving GPGPU performance.},
url = {http://doi.acm.org/10.1145/2451116.2451158},
year = {2013}
}