4 research outputs found
Hybrid centralized and distributed scheduling in large shared clusters.
No full textAbstract Datacenter-scale computing for analytics workloads is increasingly common. High operational costs force heterogeneous applications to share cluster resources for achieving economy of scale. Scheduling such large and diverse workloads is inherently hard, and existing approaches tackle this in two alternative ways: 1) centralized solutions offer strict, secure enforcement of scheduling invariants (e.g., fairness, capacity) for heterogeneous applications, 2) distributed solutions offer scalable, efficient scheduling for homogeneous applications. We argue that these solutions are complementary, and advocate a blended approach. Concretely, we propose Mercury, a hybrid resource management framework that supports the full spectrum of scheduling, from centralized to distributed. Mercury exposes a programmatic interface that allows applications to trade-off between scheduling overhead and execution guarantees. Our framework harnesses this flexibility by opportunistically utilizing resources to improve task throughput. Experimental results on production-derived workloads show gains of over 35% in task throughput. These benefits can be translated by appropriate application and framework policies into job throughput or job latency improvements. We have implemented and contributed Mercury as an extension of Apache Hadoop / YARN.
Hybrid centralized and distributed scheduling in large shared clusters.
No full textAbstract Datacenter-scale computing for analytics workloads is increasingly common. High operational costs force heterogeneous applications to share cluster resources for achieving economy of scale. Scheduling such large and diverse workloads is inherently hard, and existing approaches tackle this in two alternative ways: 1) centralized solutions offer strict, secure enforcement of scheduling invariants (e.g., fairness, capacity) for heterogeneous applications, 2) distributed solutions offer scalable, efficient scheduling for homogeneous applications. We argue that these solutions are complementary, and advocate a blended approach. Concretely, we propose Mercury, a hybrid resource management framework that supports the full spectrum of scheduling, from centralized to distributed. Mercury exposes a programmatic interface that allows applications to trade-off between scheduling overhead and execution guarantees. Our framework harnesses this flexibility by opportunistically utilizing resources to improve task throughput. Experimental results on production-derived workloads show gains of over 35% in task throughput. These benefits can be translated by appropriate application and framework policies into job throughput or job latency improvements. We have implemented and contributed 1 Mercury as an extension of Apache Hadoop / YARN
Skyler and Bliss
No full textHong Kong remains the backdrop to the science fiction movies of my youth. The city reminds me of my former training in the financial sector. It is a city in which I could have succeeded in finance, but as far as art goes it is a young city, and I am a young artist. A frustration emerges; much like the mould, the artist also had to develop new skills by killing off his former desires and manipulating technology. My new series entitled HONG KONG surface project shows a new direction in my artistic research in which my technique becomes ever simpler, reducing the traces of pixelation until objects appear almost as they were found and photographed. Skyler and Bliss presents tectonic plates based on satellite images of the Arctic. Working in a hot and humid Hong Kong where mushrooms grow ferociously, a city artificially refrigerated by climate control, this series provides a conceptual image of a imaginary typographic map for survival. (Laurent Segretier
