Search: 
 
Listed here are recent talks given at MuSyC venues.
Click here for a list of all MuSyC venues, with their associated talks.


MuSyC talks


Select by venue:  


   Challenges in Energy-Aware Task Scheduling
Pub ID:  356 Authors:  Vivek Sarkar, Lin Zhong
The research vision for the Large Scale Systems theme in MuSyC is to develop a multi-scale cross-layer distributed and hierarchical management scheme, that adapts to dynamically changing workloads. Achieving this vision requires advances in energy management strategies at the software, system, and infrastructure levels. This talk focuses on challenges in energy-aware task scheduling faced by the application runtime system in contributing to these advances. Our ultimate goal is to develop runtime systems that can model and optimize energy-aware scheduling of tasks in both throughput-sensitive jobs and latency-sensitive jobs. Energy modeling at the runtime level focuses on a longstanding challenge of per-process energy accounting in multiprocessing system i.e., how to determine the contribution by a process to the system energy consumption in a time interval when multiple processes are active. We tackle this problem with two novel methodologies. (i) In order to simplify process execution context, we leverage standard coarse-grained interfaces for power/energy and statistical learning to estimate system energy consumption for very short time intervals, i.e., 10 ms. (ii) With simplified process execution context, we exploit and extend technologies from multi-player game theory, in particular, Shapley value, to distribute the system energy consumption during a time interval to processes running in the interval. Achieving a practical and accurate per-process energy accounting provides the foundation for its use in energy-aware task scheduling. At the runtime level, we assume the use of a coordination language to express task decomposition for a process, such as Intel's Concurrent Collections (CnC), and Microsoft's Dryad. Our research has focused on extensions to the CnC language (in collaboration with Intel) to support heterogeneous computing (CPU+GPU+FPGA) and cloud computing (Hadoop) platforms, with energy efficiency as the primary motivation. The CnC model is accompanied by a number of semantic guarantees due to its use of single-assignment data (item) collections. The CnC extensions for heterogeneous computing include support for hybrid scheduling of tasks and data movements across heterogeneous processors. The CnC extensions for Hadoop include the introduction of accumulator collections and reductions steps. One language extension, the ability to create batches of steps, turned out to be useful for both GPU and Hadoop platforms. We conclude by outlining directions for future work in energy modeling and heterogeneous task scheduling to achieve a robust solution for energy-aware task scheduling, and opportunities to be gained by using such a runtime system to bridges between the Applications and OS/VM components in the Large Scale Systems software theme.
Jan 31, 2012,   MuSyC e-Seminar: Challenges in Energy-Aware Task Scheduling

   Trust-based performance improvement in social collaborative filtering and distributed decisions
Pub ID:  341 Authors:  Shanshan Zheng, John Baras
Trust is an important aspect for system design and analysis. It provides useful information for decision makings. We studied the trust-based performance improvement for social collaborative filtering and distribution decisions. Experimental results show that the use of trust can highly improve collaborative filtering performance under malicious settings, and also improve cooperations among users in ad hoc networks.
Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   A Practical Ontology Framework for Static Model Analysis
Pub ID:  309 Authors:  Ben Lickly, Edward A. Lee
In embedded software, there are many reasons to include concepts from the problem domain during design. Not only does doing so make the software more comprehensible to those with domain understanding, it also becomes possible to check that the software conforms to correctness criteria expressed in the domain of interest. Here we present a unified framework that enables users to create ontologies representing arbitrary domains of interest and analyses over those domains. These analyses may then be run against software specifications, encapsulated as models, checking that they are sound with respect to the given ontology. Our approach is general, in that the framework is agnostic to the semantic meaning of the ontologies that it uses and does not privilege the example ontologies that we present here. Where practical use-cases and principled theory exist, we provide for the expression of certain patterns of infinite ontologies. In this paper we present two patterns of infinite ontologies: those containing values, and those containing ontologies recursively. We show how these two patterns map to use cases of unit systems and structured data types, and show how these are applicable to cyber-physical systems examples drawn from automotive and avionic domains. Despite the range of ontologies and analyses that we present here, we see user-built ontologies as a key feature of our approach.
Oct 25, 2011,   MuSyC e-Seminar: A Practical Ontology Framework for Static Model Analysis

   Utility Maximization for Ultra-Energy-Efficient Systems     [ edit ]   
Pub ID:  296 Authors:  Douglas L. Jones, Le Long, David M. Cohen, David Jun
Emerging applications such as brain implants, personal health monitors, and advanced wireless sensing nodes require small, autonomous systems with much higher capability and simultaneously much lower system energy consumption. We turn to nature for inspiration: Most animals' energy expenditure is not constant but is associated with expected UTILITY; they rest much of the time when little is to be gained, and occasionally expend great bursts of energy to escape life-threatening danger or to capture prey, when the payoff is very high. We propose a similar philosophy, "expected utility maximization," for low-power systems. The end-to-end system performance defined in user-meaningful terms is the utility metric. Scalable system components capable of efficient operation across various power/performance tradeoffs are jointly and dynamically managed to maximize the expected utility payoff of the system over time within the limited total energy budget. We apply this framework to a wide variety of problems and show considerable benefits over conventional system designs such as constant-duty-cycle operation.
Sep 27, 2011,   Utility Maximization for Ultra-Energy-Efficient Systems

   Temperature and Cooling Management in Computing Systems
Pub ID:  289 Authors:  Raid Ayoub, Tajana Simunic Rosing
Temperature and cooling are critical aspects of design in todays and future computing systems. High temperature has a significant impact on reliability, performance, leakage power and cooling energy costs. State of the art temperature management techniques come with performance overhead and do not optimize for cooling energy costs. Energy management techniques usually focus on optimizing the computing energy without considering the impact on temperature or cooling system. In general, managing temperature, cooling and energy separately leads to suboptimal solutions. In this thesis we introduce a new hierarchical approach that manages the temperature, cooling and energy problems jointly and with low overhead. Our approach addresses microarchitecture, core, socket and system levels. At the microarchitectural level we achieve temperature and energy optimizations by eliminating the redundant writes to the register file at minimal per-formance overhead. The experimental results show that our technique is able to achieve on average 22% energy savings in register file with 4 degree C reduction in temperature. We next introduce a novel core level proactive thermal management technique that intelligently allocates jobs across cores of a single CPU socket to create a better thermal balance across the chip. We introduce a novel temperature predictor that is based on the band limited property of the temperature frequency spectrum where the prediction coefficients can be identified accurately at design time. Our results show that applying our algorithm considerably reduces the average system temperature, hottest core temperature, and improves performance by 6 degree C, 8 degree C and 72% respectively. At the CPU socket level, we propose a new algorithm which schedules the workload between sockets to minimize cooling energy by creating a better balance in temperature between the sockets. The reported results show that combining the socket level with the core level optimizations can result in cooling energy savings of 80% on average at performance overhead of less than 1%. Finally, we describe a combined temperature, cooling and energy management approach that significantly lowers the cooling energy costs of the system as well as the operational energy of memory. We introduce a comprehensive thermal and cooling model which is used for online decisions. This technique clusters the memory accesses to subset of memory modules in tandem with balancing the temperature between and within the CPU sockets. The experimental results show that our method delivers an average cooling and memory energy savings of up to 70% compared to the state of the art techniques at performance overhead of less than 1%.
Aug 29, 2011,   Temperature and Cooling Management in Computing Systems

   The Swarm at the Edge of the Cloud - A New Face of Wireless
Pub ID:  293 Author:  Jan Rabaey
Mobile devices such as laptops, netbooks, tablets, smart phones and game consoles have become our de facto interface to the vast amount of information delivery and processing capabilities of the cloud. The move to mobility has been enabled by the dual forces of ubiquitous wireless connectivity combined with the increasing energy efficiency offered by Moore's law. Yet, a major component of the mobile remains largely untapped: the capability to interact with the world immediately around us. A third layer of information acquisition and processing devices - commonly called the sensory swarm - is emerging, enabled by even more pervasive wireless networking and the introduction of novel ultra-low power technologies. This gives rise to the true emergence of concepts such as cyber-physical and bio-cyber systems, immersive computing, and augmented reality. The functionality of the swarm arises from connections of devices, leading to a convergence between Moore’s and Metcalfe’s laws, in which scaling refers not any longer to the number of transistors per chip, but rather to the number of interconnected devices. Enabling this fascinating paradigm – which represents true wireless ubiquity – still requires major breakthroughs on a number of fronts. Providing the always-connected abstraction and the reliability needed for many of the intended applications requires a careful balancing of resources that are in high demand: spectrum and energy. This paper analyzes those challenges, and proposes some disruptive solutions that engage the complete stack – from circuit to system.
Aug 29, 2011,   Temperature and Cooling Management in Computing Systems