MuSyC posters

Select by venue:

Utilizing Green Energy Prediction To Schedule Mixed Batch And Service Jobs In Data Centers

Pub ID: 315	Authors: Baris Aksanli, Jagannathan Venkatesh, Liuyi Zhang, Tajana Simunic Rosing
As brown energy costs grow, renewable energy becomes more widely used. Previous work focused on using immediately available green energy to supplement the non-renewable, or brown energy at the cost of canceling and rescheduling jobs whenever the green energy availability is too low. We design an adaptive data center job scheduler which utilizes short term prediction of solar and wind energy production. This enables us to scale the number of jobs to the expected energy availability, thus reducing the number of cancelled jobs by 2x and improving green energy usage efficiency by 2x over just utilizing the immediately available green energy.

Nov 16, 2011, MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Duty-Cycling Buildings Aggressively: The Next Frontier in HVAC Control

Pub ID:  331 Authors:  Thomas Weng, Bharathan Balaji, Yuvraj Agarwal, Rajesh Gupta

Buildings are known to be the largest consumers of electricity in the United States. While there can be several subsystems that can dominate depending on use modalities of buildings, often the largest electricity consumer is the air-conditioning and ventilation (HVAC) system. Despite this fact, in most buildings the HVAC system is run using fairly primitive control algorithms based on fixed work schedules of people within the buildings causing wasted energy during periods of low occupancy. In this paper we present a novel HVAC control architecture that uses occupancy sensing to guide the operation of a building HVAC system. We show how we can enable aggressive "duty-cycling" of building HVAC systems -- that is, turn them ON or OFF -- to save energy while meeting building performance requirements using inexpensive sensing and control methods. We have deployed our occupancy sensor network across an entire floor of a university building and our data shows several periods of low occupancy with significant opportunities to save energy. Furthermore, by interfacing with the building Energy Management System (EMS) directly and using the real-time occupancy data collected by our occupancy sensor we measure energy savings of up to 13% in the HVAC system by controlling just one floor of our building

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Reducing the Flow Completion Time Tail in Datacenter Networks     [ edit ]

Pub ID:  348 Authors:  David Zats, Tathagata Das, Prashanth Mohan, Randy Katz

Web sites are increasingly backed by complex processing to deliver rich content to users via web pages. Despite the increased complexity, the pages must still be delivered quickly and consistently to meet user’s expectations for interactivity. To achieve this, datacenters typically employ application-level mechanisms to squeeze in as much complex processing as possible while still meeting page delivery deadlines. However, network variability can result in variable packet latency and a long flow completion time tail. This ultimately leads to either reduced page quality to meet deadlines or increased deadline misses. In this work, we evaluate the benefits of a new network congestion management approach for reducing the flow completion time tail. We show that in-network traffic management, multipath data transfers, and traffic differentiation are all essential for reducing the flow completion time tail.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Per-Process Energy Accounting

Pub ID:  316 Authors:  Mian Dong, Lin Zhong

Energy has become an important system resource due to both electricity and thermal concerns. Per-process energy accounting is to obtain the energy contribution by each running process, which is the foundation to energy management & optimization in the operating system (OS). In this work, we transform the problem of per-process energy accounting into the problem of predicting the total energy consumption of the system when any subset of all existing processes are running. We further propose two prediction models to solve the problem.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   JETC: Joint Energy Thermal and Cooling Management for Memory and CPU Subsystems in Servers

Pub ID:  332 Authors:  Raid Ayoub, Rajib Nath, Tajana Simunic Rosing

In this work we propose a joint energy, thermal and cooling management technique (JETC) that significantly lowers per server cooling and energy costs. Our analysis shows that decoupling the optimization of cooling costs and energy consumption of CPU and memory leads to suboptimal solutions due to thermal dependencies between CPU and memory, and non-linearity in cooling costs. This motivates us to develop a holistic solution that integrates the energy, thermal and cooling management to maximize energy savings with negligible performance hit. We develop a comprehensive thermal and cooling model which is used for online optimization in JETC. JETC decisions considers thermal state of CPU $\&$ memory, dependencies between them and fan speed to arrive at energy efficient decisions. It has CPU and memory actuators where it activates them jointly or separately depending on the thermal and cooling state of the system. Memory actuator of JETC reduces the operational energy of memory by cooling aware clustering the pages to a subset of memory modules. CPU actuator saves cooling costs by removing hot spots between and within the sockets, and reduces the effects of thermal coupling with memory. Extensive experimental results we provide show that employing JETC results in 50.7% average energy reduction of memory and cooling subsystems with less than 0.2% performance overhead.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Performance Analysis of Synchronous Models Implementations on Loosely Time-Triggered Architectures     [ edit ]

Pub ID:  317 Authors:  Chung‑Wei Lin, Marco Di Natale, Haibo Zeng, Alberto Sangiovanni‑Vincentelli

Synchronous languages are used in the most popular software and system modeling environments because of the availability of tools for validation and verification by simulation or model checking. To preserve the validation results, their implementation must be provably correct with respect to the preservation of the semantics properties of interest. An implementation path for synchronous models can easily be defined for time-triggered platforms, at the expense of some inflexibility. A mapping of synchronous models onto a much less restrictive platform, Loosely Time Triggered Architecture (LTTA), that preserves the communication flows has been defined in one previous work. The mapping uses intermediate layers with queues and then backpressure communication channels. Its time performance is analyzed using a virtual clock model with several pessimistic assumptions and limitations, including zero bus delays and no control on schedulability. In this research, we propose the use of the Real-Time Calculus (RTC) as a general model for the analysis of such systems. To apply the analysis to the implementation of synchronous models into LTTA, several additions and fixes to the original RTC theory and analyses are needed, including a fixed equation of the lower bound of performance, an improved computation of the effective service curves, modeling of forks and merges of communication paths, and delay consideration. Experimental results show that it can exactly match the true best/worst performance for 2 processes, and our approaches can indeed achieve a more accurate performance analysis and provide fundamental analysis for other mapping problems and other architectures.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Themis: Energy Management in Virtualized Environments     [ edit ]

Pub ID:  333 Authors:  Liuyi Zhang, Gaurav Dhiman, Vasileios Kontorinis, Tajana Simunic Rosing

Virtualized data centers facilitate higher resource utilization and energy efciency through consolidation. However, mixing services-oriented workloads with throughput (batch) workloads is typically avoided due to complex interactions and widely different quality of service (QoS) requirements. This work demonstrates that consolidating these diverse workloads provides signicant opportunity for improved energy efciency as each tends to stress different aspects of the host hardware. It provides a new unied metric, qMIPS/Watt, which quanties the combined efciency in terms of work done per Joule of the heterogeneous workload combination. It describes a complete VM resource management framework, called Themis, which manages combined services and batch jobs, maximizing energy-efcient throughput of the latter without sacricing the service guarantees of the former. Themis is implemented on a testbed of state of the art server machines, and its resource management policy outperforms prior proposed policies by up to 35% on average in work done per Joule.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Distributed Thermal Management in Heterogeneous System

Pub ID:  318 Authors:  Yen‑Kuan Wu, Sharifi Shervin, Tajana Simunic Rosing

This work addresses thermal management in heterogeneous system where the power states of the CPU can be controlled by the operating system (OS) while OS is not able to control power states of the hardware accelerators. We propose ``Distributed Thermal Management'' (DistriTherm) technique to address the thermal management for heterogeneous system. DistriTherm is a scalable and cooperative distributed thermal management technique which reaches a DVFS decision based on a negotiation protocol among thermally correlated cores to meet performance requirements and ensure temperature improvements. Experimental results show that for our technique can successfully reduce the deadline miss rate by 43% in average compared to localized thermal management techniques while successfully satisfying temperature constraints.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Managing Distributed UPS Energy for Effective Power Capping in Data Centers     [ edit ]

Pub ID:  334 Authors:  Vasileios Kontorinis, Liuyi Zhang, Baris Aksanli, Jack Sampson, Houman Homayoun, Eddie Pettis, Dean Tullsen, Tajana Simunic Rosing

Power over-subscription is a well-known technique to reduce both one-time capital costs and operating costs for modern data centers. However, designing the power infrastructure for a lower power point than the aggregated peak power of all servers requires dynamic techniques to avoid tripping breakers under worst-case power scenarios. This work considers the use of distributed per-server batteries to store energy during low activity periods and use this energy during power spikes. We leverage the distributed nature of the batteries and design algorithms to prolong the duration of their usage. We find that we can reduce data center peak power by 19% and support 23.5% more servers with the same power infrastructure. Our approach reduces data center total cost of ownership per server by 6.2% with zero impact on performance.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   5nW crystal oscillator for ultra-low power     [ edit ]

Pub ID:  319 Authors:  Dongmin Yoon, Dennis Sylvester, David Blaauw

A 5.58nW crystal oscillator circuit is implemented using a DLL to generate pulses driving the crystal. Three voltage domains are internally generated from a single supply to enable small oscillation amplitude with high driver transconductance. Resulting frequency characteristics maintain crystal performance with the lowest power reported for a 32.768kHz crystal oscillator.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Practical Infinite Ontologies for Static Model Analysis

Pub ID:  335 Authors:  Ben Lickly, Edward A. Lee

While many types of knowledge can be encoded in finite ontologies, infinite ontologies can express a larger class of properties. We have found two patterns of infinite ontologies to be broadly useful:

Value-parameterized Concepts: Useful for including value information into the ontologies.
Recursive Ontologies: Useful for including structured information corresponding to structured data types.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   A Transparent μECOG Array for Bidirectional Brain-Machine Interfaces     [ edit ]

Pub ID:  314 Authors:  Peter Ledochowitsch, Elisa Olivero, Tim Blanche, Michel Maharbiz, Jose M. Carmena

We report for the first time the design, fabrication and characterization of an optically transparent electrode array for micro-electrocorticography. We present a 49-channel μECoG array with an electrode pitch of 800 μm and a 16-channel linear μECoG array with an electrode pitch of 200 μm. The backing material was parylene C. Transparent, sputtered indium tin oxide (ITO) was used in conjunction with e-beam evaporated gold to fabricate the electrodes. We provide electrochemical impedance characterization, light transmission data and an in vivo demo for the fabricated devices.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Platform-based Architecture Exploration for Heterogeneous System in Multiple Modeling Environments     [ edit ]

Pub ID:  320 Authors:  Liangpeng Guo, Carlo Caione, Alberto Puggelli, Pierluigi Nuzzo, Alberto Sangiovanni‑Vincentelli

As electronic embedded systems become more and more complex, it’s very challenging to integrate sub-systems on an architecture. One important problem is how to deploy the functions and signals on a distributed architecture. The design space is so large that designers have too many choices that yield different performances. Since the static descriptions are not enough to get an accurate performance estimation under most cases, only simulation can provide the accuracy for architecture exploration. Platform-based Design (PBD) provides a methodology that separates the functional model and architectural model, thus enables the design space exploration by simply changing the mappings. Metropolis and its following versions implement this idea. In this project, we extend this work to allow external modeling tools to make use of the ability of mappings. Ptolemy, SystemC and SystemC-AMS are used as an examples of external tools.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Temporal Logic Planning     [ edit ]

Pub ID:  336 Authors:  Necmiye Ozay, Ufuk Topcu, Petter Nilsson, Richard Murray

Temporal logic planning framework provides a means for automatically synthesizing embedded control software that is provably correct with respect to an expressive subset of linear temporal logic (LTL) specifications. In this poster, we review the basics of this framework and discuss two extensions together with applications in vehicle management systems. First, we consider the problem of designing distributed control protocols that cooperatively achieve high level goals and requirements while dynamically reacting to the changes in the internal system state and external environment. We start with a global specification and decompose it into local ones. These decompositions allow the protocols for each local controller to be separately synthesized and locally implemented while guaranteeing the global specifications to hold. Moreover, they naturally lead to interface specifications between subsystems which facilitate design modularity. The methodology is demonstrated with an application in distributed power allocation in an aircraft. The second problem we consider is hybrid controller synthesis for switched affine systems. Our methodology starts by lifting the problem to a discrete level by constructing a finite transition system that abstracts the behavior of the underlying switched system. At the discrete level, we recast the problem as a two player temporal logic game by treating the environment driven switches as adversaries. The solution strategy for the game (i.e. the discrete plan) is then implemented at the continuous level by solving finite-horizon optimal control problems that establish reachability between discrete states. We demonstrate this idea with an example in aircraft fuel system where the goal is to regulate the fuel levels in multiple tanks during interesting operations like aerial refueling.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Fast Crash Recovery in RAMCloud     [ edit ]

Pub ID:  302 Authors:  Stephen Rumble, John Ousterhout, Diego Ongaro, Ryan Stutsman, Mendel Rosenblum

RAMCloud is a DRAM-based storage system that provides inexpensive durability and availability by recovering quickly after crashes, rather than storing replicas in DRAM. RAMCloud scatters backup data across hundreds or thousands of disks, and it harnesses hundreds of servers in parallel to reconstruct lost data. The system uses a log-structured approach for all its data, in DRAM as well as on disk; this provides high performance both during normal operation and during recovery. RAMCloud employs randomized techniques to manage the system in a scalable and decentralized fashion. In a 60-node cluster, RAMCloud recovers 35 GB of data from a failed server in 1.6 seconds. Our measurements suggest that the approach will scale to recover larger memory sizes (64 GB or more) in less time with larger clusters.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Dynamic Deferral of Workload for Capacity Provisioning in Data Centers     [ edit ]

Pub ID:  321 Authors:  Muhammad Adnan, Yan Ma, Ryo Sugihara, Rajesh Gupta

Recent increase in energy prices has led researchers to find better ways for capacity provisioning in data centers to reduce energy wastage due to the variation in workload. This paper explores the opportunity for cost saving and proposes a novel approach for capacity provisioning under bounded latency requirements for the workload. We investigate how many servers to be kept active and how much workload to be delayed for energy saving while meeting every deadline. We present an offline LP formulation for capacity provisioning by dynamic deferral and give two online algorithms to determine the capacity of the data center and the assignment of workload to servers dynamically. We prove the feasibility of the online algorithms and show that their worst case performance are bounded by a constant factor with respect to the offline formulation. We validate our algorithms on synthetic workload generated from two real HTTP traces and show that they actually perform much better in practice than the worst case, resulting in 20-40% cost-savings.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Duty-Cycling Buildings Aggressively: The Next Frontier in HVAC Control

Pub ID:  337 Authors:  Thomas Weng, Bharathan Balaji, Rajesh Gupta, Yuvraj Agarwal

Buildings are known to be the largest consumers of electricity in the United States. While there can be several subsystems that can dominate depending on use modalities of buildings, often the largest electricity consumer is the air-conditioning and ventilation (HVAC) system. Despite this fact, in most buildings the HVAC system is run using fairly primitive control algorithms based on fixed work schedules of people within the buildings causing wasted energy during periods of low occupancy. In this paper we present a novel HVAC control architecture that uses occupancy sensing to guide the operation of a building HVAC system. We show how we can enable aggressive "duty-cycling" of building HVAC systems -- that is, turn them ON or OFF -- to save energy while meeting building performance requirements using inexpensive sensing and control methods. We have deployed our occupancy sensor network across an entire floor of a university building and our data shows several periods of low occupancy with significant opportunities to save energy. Furthermore, by interfacing with the building Energy Management System (EMS) directly and using the real-time occupancy data collected by our occupancy sensor we measure energy savings of up to 13% in the HVAC system by controlling just one floor of our building.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   QoS-based Dynamic Allocation in Embedded Systems: a Methodology and a Framework

Pub ID:  303 Authors:  Fernando Pianegiani, Alberto Sangiovanni‑Vincentelli

This work focuses on a methodology for the dynamic allocation of embedded resources apt to satisfy requests of service with constraints of Quality of Service (QoS). Given a set of services provided by an embedded system, this methodology suggests how to determine the best match between a requested service and the set of Embedded subSystems (EsSs) able to provide that service. The selection of EsSs is carried out on the basis of the results of an a priori evaluation of the performances of service execution and of the availability of the single embedded resources. Moreover, the proposed methodology defines some guidelines to dynamically schedule and thus to assign the execution of the services to the chosen EsSs. With regard to the selection process this methodology benefits from both the content− and the collaborative−based filtering methods. To evaluate its consistency it has been implemented in a testing framework, named BIOS, that includes a repository of services, of service providers (i.e. of EsSs), of evaluation outcomes of the performances of each EsS in carrying out its own services, of an engine for the dynamic selection of the EsSs able to satisfy the requirements of a given service and of a container of Real−Time Operating Systems (RTOSs) able to efficiently schedule and to assign the execution of the requested services to the chosen EsSs. Considering state−of−the−art services provided by Body Sensor Network (BSN) subsystems, the hybrid solution proposed for the selection process has been tested and it has shown better results in comparison to the content−based filtering technique.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Expected-Utility-Based Sensor Selection For State Estimation

Pub ID:  322 Authors:  David M. Cohen, Sriram Narayanan, Douglas L. Jones

Applications such as long-term environmental monitoring and large-scale surveillance require sensor nodes to run for up to years at a time on a single battery charge. There is often not enough power for sensors to make measurements all of the time. In these cases, one must decide when to run each sensor. To this end, we develop a one-step optimal sensor-scheduling algorithm based on expected-utility maximization. “Utility” is an application-specific measure of the benefit from a given sensor measurement. In dynamic sensing environments that can be modeled by a hidden Markov model, selecting the appropriate combination of sensors at each time instant enables maximization of the expected utility while operating within an energy budget. The Lagrange dual problem is solved to find the minimum expected-utility-to-sensor-cost ratio required to stay within budget. For some budgets, the utility-based algorithm shows more than 300% utility gains over a constant duty-cycle scheme designed to consume the same amount of energy. These benefits are dependent on the energy budget.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Resiliency-Aware Scheduling and Co-Design

Pub ID:  338 Authors:  Jeremy Abramson, Robert Lucas

Instruction-level parallelism limits, shrinking feature sizes and hostile environments create a need for resilient computing. Resiliency-aware scheduling uses Hybrid TMR and intrinsic resiliency to take advantage of inherent properties of a computation to deliver the highest resiliency, lowest latency and smallest area.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   miniTP: a Protocol for the Minimization of the Transmit Power in Wireless Networks

Pub ID:  304 Authors:  Fernando Pianegiani, Alberto Sangiovanni‑Vincentelli

The environmental conditions under which a wireless communication system needs to work not always can be accurately modeled a priori. As a consequence, in many cases the Radio Frequency (RF) output power should be adjusted on the field at run--time in order to guarantee the functionalities and the performances of the network. It is especially true for networks with nonstationary channel, in particular for mobile networks. Today many RF devices allow to set at run--time the RF transmit power to different levels by the help of embedded micro--controllers. In this work we present a protocol for the minimization of the transmit power (miniTP) in wireless networks on the basis of the specifications and the requirements of the considered network system, that is on the basis of its network topology, of its communication protocol and of its expected performances, e.g. in terms of Link Quality (LQ), energy consumption, interference, etc.. miniTP has been designed to work with a large spectrum of wireless communication protocols. Without loss of generality, to provide a more detailed treatment, in the poster we formally describe miniTP for cluster--tree Wireless Sensor Networks (WSNs) that must satisfy given requirements of LQ within each cluster and we analyze its benefits considering the case study of a cluster--tree WSN for the monitoring of servers in large data centers. To simulate and to test miniTP for the considered case study by the use of the TOSSIM 2 simulator, miniTP has been implemented in NesC under TinyOS for the WSN module MicaZ from Crossbow Technology, Inc., which includes the IEEE 802.15.4 RF device CC2420 from Texas Instruments, Inc. that is embedded also in several other WSN nodes. During the test miniTP has set at run--time the minimum transmit power among the ones allowed by the CC2420 transceiver to satisfy the requirements of the case study and in particular to maintain the PARR of the star links within each cluster greater than or equal to a fixed minimum threshold, where the PARR has been defined as the packet reception rate (PRR) for ACK--based transmissions. The results obtained have shown that miniTP has worked as expected under nonstationary conditions, also with mobile nodes.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Control Protocol Synthesis of an Electric Power System

Pub ID:  323 Authors:  Huan Xu, Ufuk Topcu, Richard Murray

In more-electric aircrafts (MEA), aircraft subsystems have shifted from pneumatic and hydraulic sources to engine driven electric ones. This has been challenging for traditional architecture designs which have previously relied on one or two central power sources. New electric power systems (EPS) needs to be able to take on configurations different from those in conventional aircrafts. Computational complexity in MEA becomes an issue for verification of systems and safety certification. Systematic methods for verifying models against their specifications and synthesizing correct-by-construction control protocols are needed to reduce the amount of time in validation and verification. Main design considerations for MEA need to include such things as fault tolerance, timing constraints, and prioritization of sources for each bus. One of the difficulties in designing controllers for these systems comes from the translation of requirements from English-based text to mathematical formulas. We utilize linear temporal logic (LTL) as a formal specification language and extend previous work on synthesis of protocols for embedded controllers to a case study of an EPS on board of an MEA. The following work provides a general framework for the specification and synthesis of these types of systems with a given topology and a typical set of requirements.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Integrated SysML-Modelica Modeling Framework and Trade-off Analysis for VMS     [ edit ]

Pub ID:  339 Authors:  Baobing Wang, Dimitris Spyropoulos, Shah‑An Yang, John Baras, Ion Matei

To be added

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Themis: A Platform for Balanced Data-Intensive Scalable Computing     [ edit ]

Pub ID:  305 Authors:  Alex Rasmussen, Michael Conley, George Porter, Rishi Kapoor, Vinh The Lam, Amin Vahdat

It is increasingly important that data-intensive scalable computing (DISC) systems are balanced, meaning that they utilize their available hardware as efficiently as possible. Balanced DISC systems enable the same workload to be performed much more quickly or on far fewer machines, providing significant cost and energy savings. Unfortunately, it is hard to write DISC systems that are balanced, and harder to make a system that is balanced on one hardware configuration remain balanced when moved to another configuration. We present Themis, a platform for building balanced data-intensive scalable computing systems that eases the process of making DISC systems balanced. Themis provides an extensible logging and monitoring framework as well as a suite of meta-programs designed to automatically recommend configuration adjustments that will make the system more balanced. We are evaluating Themis using ThemisMR, a MapReduce implementation built in the Themis framework that currently holds the 100TB Daytona sort benchmark record.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Behavioral and Parametric Modeling of Boeing 767 Electrical Power System using SysML     [ edit ]

Pub ID:  324 Authors:  John Finn, Di Marco Piergiuseppe , Mohammad Mostafizur Rahman Mozumdar, Alberto Sangiovanni‑Vincentelli

An aircraft’s electric power system is a fundamental component as it provides electrical power to vital aircraft loads such as the landing gear and avionics. Today, electric power system design is largely top-down, which limits the ability to foresee the effects of design decisions made early in the design process. This approach leads to many re-design steps, which increases cost and time to market. The Platform Based Design methodology has been shown to overcome such limitations. Within this methodology, our objective is providing a reactive model of an aircraft power system having functional blocks representing different physical components and controllers using the SysML profile. Specifically, we develop behavioral models and parametric evaluation of the Boeing 767 electric power system. The developed models describe the communication between blocks and analyze the supply and demand of available aircraft power.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Trust-based performance improvement in social collaborative filtering and distributed decisions

Pub ID:  340 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

   Model-Based Design Framework for Wireless Sensor Networks Using SysML, Simulink and Modelica

Pub ID:  306 Authors:  Baobing Wang, John Baras

Existing ad hoc system design methods for Wireless Sensor Networks (WSNs) suffer from lack of reusability, trade-off analysis and design space exploration methods and tools. In addition, the interactions between the continuous-time physical environments and WSNs have not been well studied. In this paper, we propose a model-based systems design (MBSD) framework for WSNs, which is a systematic methodology applying systems engineering principles to support system requirements, design, analysis and verification/validation processes. Firstly, we describe a hierarchy of model libraries to model various behaviors and structures of WSNs, including physical environments, physical platforms, communication and computation components, system services and applications. Based on the MBSD framework, we introduce a system design flow to compose both continuous-time and event-triggered-time modules to develop applications with support for trade-off analysis, design space exploration and interactive simulations. Next, the main modules for physical platforms, the Media Access Control (MAC) layer, wireless channels and physical environments are described in detail, and are modeled in the Systems Modeling Language (SysML), Simulink and Modelica. Finally, we use a building thermal control system as the case study to demonstrate the composability, reusability and flexibility of the proposed MBSD framework.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Receding Horizon Control Subject to Energy Hierarchical Constraints

Pub ID:  325 Authors:  Eduardo Arvelo, Nuno Martins

We are interested in developing receding horizon control (RHC) strategies that work well with systems that have a hierarchical energy allocation policy and may exhibit integral-type constraints. We have developed a RHC algorithm based on contractive constraints and varying horizon lengths that are well suited for problems of this type. These ideas are demonstrated using an aircraft environmental control subsystem. Furthermore, we are investigating randomized-based approaches to solving RHC problems. This method can be implemented on small platforms (such as miniature robots) that have computational and power constraints.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Efficient Network Topologies and Protocols for Communication Networks in Collaborative Control     [ edit ]

Pub ID:  342 Authors:  Hua Chen, Pedram Hovareshti, John Baras

Network topologies and protocols are key elements of the communication networks for collaborative control application. Often a large group of micro-agents collaborate in a distributed fashion to perform a mission. We proposed a three-tier organization of collaborative control networks consisting of connectivity, communication and action graphs. We proposed an efficient network formation design based on networks motifs. We proposed a cross-layer MAC protocol based on opportunistic scheduling and motion planning.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Closed-Loop Decoder Adaptation (CLDA) Algorithms for Brain-Machine Interfaces     [ edit ]

Pub ID:  307 Authors:  Jose M. Carmena, Siddharth Dangi, Amy Orsborn

Brain-Machine Interfaces (BMIs), which aim to restore motor function to patients with disabilities, are ideal platform demonstrators for small scale systems. They allow us to explore, analyze, and implement advanced closed-loop learning systems. BMIs use a decoding algorithm – or “decoder” – to translate the user’s neural activity into control signals for a computer cursor, prosthetic device, or other actuator. Closed-Loop Decoder Adaptation (CLDA) refers to the process of updating the decoder “online” (while the subject is using the BMI.) CLDA is meant to improve the decoder to make it more accurately represent the true underlying mapping between the user’s neural activity and their intended movements. We explore various CLDA algorithms and present experimental results demonstrating their effect on BMI performance.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Development of Building Automation and Control (BAC) Systems

Pub ID:  326 Authors:  Mehdi Maasoumy, Yang Yang, Alberto Sangiovanni‑Vincentelli

In this work, we propose methodologies for designing Building Automation and Control (BAC) systems, including methodology for building simulation models, methodology for existing physical buildings, and a design flow for BAC systems. A case study of heterogeneous room temperature control system is conducted to show the effectiveness of the method.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Physical Layer Security

Pub ID:  343 Authors:  Tuan (Johnny) Ta, John Baras

Security has traditionally been done almost entirely on software. Reality shows that hackers are becoming more and more sophisticated. Research and industry are moving toward a new approach: physical layer and hardware security. These techniques make use of the peculiarities of waveform, RF, hardware to create unshakable fingerprints for authentication purpose. We propose a technique in which an intelligently designed tag is embedded to the physical waveform to authenticate the transmitter. Using such technique, we propose a 2-stage authentication process in a remote authentication framework. The first stage involves authenticating the user to the device. Here we introduce a robust technique to verify the integrity of fingerprint scanners which combats the vulnerability of fingerprints for being public and easy to capture. The second stage utilizes the embedded tag to authenticate the device to the network. We show 2 applications of our physical layer security techniques as case studies.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   TAP: Token-Based Adaptive Power Gating     [ edit ]

Pub ID:  308 Authors:  Richard Strong, Tajana Simunic Rosing, SeokHyeong Kang, Andrew Kahng

In mobile systems, the problems of short battery life and increased temperature are exacerbated by wasted leakage power. Leakage power waste can be reduced by power-gating a core while it is stalled waiting for a resource. In this work, we propose and model memory miss power gating (MMPG), a low-overhead technique to enable power-gating of an active core when it stalls during a long memory access. We describe a programmable two-stage power gating switch design that can vary a core's wakeup delay while maintaining voltage noise limits and leakage power savings. We also model the processor power distribution network and the effect of memory miss power gating on neighboring cores. Last, we apply our power gating technique to actual benchmarks, and examine energy savings and overheads from power gating stalled cores during long memory accesses. Our analyses show the potential for over 38% energy savings given 'perfect' power gating on memory misses; we achieve energy savings exceeding 20% for a practical, counter-based implementation.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Towards an Ultra Low-Energy Computation with Asynchronous Circuits

Pub ID:  327 Authors:  Tsung‑Te Liu, Jan Rabaey

Emerging biomedical applications would benefit from the availability of digital processors with substantially increased energy-efficiency. One approach to realize ultra-low-energy processors is to scale the VDD aggressively to below the transistor threshold, yet the major increase in delay variability under PVT variations combined with the dominance of leakage power makes robust subthreshold computations and further voltage scaling extremely challenging. We developed an asynchronous self-timed approach that allows for an adaptive adjustment to latency variations and support for an inherent leakage minimization for both static and dynamic variations, leading to a robust and energy-efficient sub-threshold computation architecture. The 65nm CMOS asynchronous neural signal processor presented in this poster demonstrates a 2× reduction in power consumption and a better statistical characteristic than the synchronous design. The proposed approach is essential to enable CMOS circuit to fully benefit from the continued technology scaling and realize ultra-low-voltage operations, without incurring the leakage and variability issues associated with the conventional synchronous implementation.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   A Case Study: Energy-scalable Detection on the MSP430     [ edit ]

Pub ID:  344 Authors:  David Jun, Long Le, Douglas L. Jones

Detection of natural signals represent an energy-demanding class of applications that cannot be developed using traditional microcontroller design principles. We demonstrate how scalable algorithms and dynamic voltage and frequency scaling can result in up to 96% improvement in energy consumption on a TI MSP430 processor.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Fabrication of True 3D Integrated Antenna

Pub ID:  310 Authors:  Peter Gadfort, Paul Franzon

In order to create a 1 mm³ sensor which can be implanted and powered by near field inductive power harvesting, a true 3D integration packaging, with vertically attached die, will be needed in order to build a robust power delivery system. The addition of orthogonal power coils have a negligible effect on the individual coil performances, while being able to provide power if the sensor moves after implantation. Here we present a method for fabricating a true three-dimensional implantable integrated sensor at the 3 mm³ and 5 mm³ scales.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   An Introspective Framework for Soft Error Resilience     [ edit ]

Pub ID:  328 Authors:  Saurabh Hukerikar, Robert Lucas, Pedro Diniz, Jeff Draper

Soft errors in memory or logic are an increasingly common occurrence. Traditional hardware techniques focus on error correction to offer programmers the illusion of an error-free execution environment. Hardware pairing and triple-modular redundancy (TMR) can be used to detect and correct soft errors in logic, but are wasteful in terms of space and power. At the memory level, parity and SEC/DED (Single Error Correction Double Error Detection) codes can be used to correct them. Yet, multi-bit errors today are not corrected and are likely to result in catastrophic system failure. As error rates increase, techniques such as check-pointing will become less effective, as dwindling MTBF shortens the check-pointing interval.

In this work, we explore a software approach for tolerating multi-bit errors in memory by combining user-provided knowledge with system introspection. An introspective runtime framework observes the address space to determine the contextual significance of the corrupted data and examines trends of the execution and hardware resources used to transparently reassign resources so as to avoid execution and memory resources that consistently exhibit failures. At the application level, rather than enduring system failure, we offer the programmer an application-level interface to specify where, and which errors can be tolerated. Programmers also specify what corrective actions, if any, to undertake in the case of the unrecoverable data failure. We also leverage compiler- based program analysis coupled with user directives to reduce the overhead of the implementation of the remedies, such as reducing the volume of data saved/restored with check-pointing.

Ultimately, with the synergistic combination of programmer knowledge and compiler and runtime introspection we aim to substantially increase the dependability of the applications that will be running on large scale computing platforms by the end of this decade.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Energy Efficiency for Large-Scale MapReduceWorkloads with Significant Interactive Analysis     [ edit ]

Pub ID:  345 Authors:  Yanpei Chen, Sara Alspaugh, Dhruba Bothakur, Randy Katz

MapReduce workloads have evolved to include increasing amounts of time-sensitive, interactive data analysis; we refer to such workloads as MapReduce with Interactive Analysis (MIA). Such workloads run on large clusters, whose size and cost make energy efficiency a critical concern. Prior work on MapReduce energy efficiency have not yet considered this workload class. Increasing hardware utilization helps improve efficiency, but is challenging to achieve for MIA workloads. These concerns lead us to develop BEEMR, an energy efficient MapReduce workload manager motivated by empirical analysis of real-life MIA traces at Facebook. The key insight is that although MIA clusters host huge data volumes, the interactive jobs operate on a small fraction of the data, and thus can be served by a small pool of dedicated machines; the less time-sensitive jobs can run on the rest of the cluster in a batch fashion. BEEMR achieves 40-50% energy savings under tight design constraints, and represents a first step towards improving energy efficiency for an increasingly important class of datacenter workloads.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   A Modular 1mm^3 Die-Stacked Sensing Platform     [ edit ]

Pub ID:  311 Authors:  Yoonmyung Lee, Gyouho Kim, Suyoung Bang, Yejoong Kim, Inhee Lee, Dennis Sylvester, David Blaauw

An 1.0 mm³ general-purpose heterogeneous sensor node platform with a stackable multi-layer structure is proposed. To support modular structure, ultra-low power I²C interface for inter-layer communication is implemented. The system has an ultra-low power optical wakeup receiver, GOC (Global Optical Communication), which allows for re-programming or synchronization. It also includes an ultra-low power PMU (Power Management Unit) with mulit-modal harvesting and BOD (Brown-Out Detector) to prevent processor malfunctions and battery damage. Image and temperature sensors are implemented, but the modularity of the system allows end users to easily replace or add layers to incorporate specific circuits in appropriate technologies as needed.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   OS-level power minimization under tight performance constraints in general purpose systems

Pub ID:  329 Authors:  Raid Ayoub, Yanqin Jin, Tajana Simunic Rosing

State-of-art OS-level power managers such as the on-demand governor in Linux do not consider performance constraint nor application characteristics. However people usually want to achieve power reduction while meeting SLAs. In order to tackle this problem, we propose a new DVFS algorithm for enterprise systems that elevates performance as a first order control parameter and manages frequency and voltage as a function of performance requirements. We implement our algorithm on real Intel Westmere platform in Linux and demonstrate its ability to reduce the standard deviation from target performance by more than 90% over state of the art policies while reducing average power by 17%. Experiment results show that our approach can capture the dynamic fluctuation of characteristics of different workloads or workload combinations and make wise decisions in the management of CPU frequencies.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Low-Power Scalable Platform for Electrocorticography     [ edit ]

Pub ID:  346 Authors:  Wen Li, Rikky Muller, Peter Ledochowitsch, Jan Rabaey, Hanh‑Phuc Le, Toni Bjorninen

This poster shows a first prototype for a scalable and fully implantable wireless platform for electrocorticography (ECoG), an electrophysiological technique where electrical potentials are recorded from the surface of the cerebral cortex. An implantable system requires neural recording electronics integrated with wireless power coupling and wireless data transmission.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Checking for Circular Dependencies in Distributed Stream Programs

Pub ID:  312 Authors:  Dai Bui, Edward A. Lee

This work presents a cyclic dependency analysis for stream-based programs. Specifically, we focus on the cyclo-static dataflow (CSDF) programming model with control messages through teleport messaging as implemented in the StreamIt framework. Unlike existing cyclic dependency analyses, we allow overlapped teleport messages. An overlapped teleport message is one that traverses actors that themselves transmit teleport messages, which can complicate the stream graph topology with teleport messages. Therefore, the challenge in this work is to decide whether such stream graphs are feasible in the presence of such complex teleport messages. Our analysis addresses this challenge by first ensuring that the stream graph with teleport messages is feasible, and then computing an execution schedule for the CSDF graph in the presence of complex overlapped teleport messaging constraints. Consequently, our analysis accepts a larger class of CSDF stream graphs with complex teleport messaging topologies for execution.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Managing Plug-Loads for Demand Response within Buildings

Pub ID:  330 Authors:  Yuvraj Agarwal, Thomas Weng, Bharathan Balaji, Rajesh Gupta

Demand Response is the ability of a building to to reduce loads due to requests from the grid or extremely high prices. Current technologies only target HVAC and lighting for the reductions in energy demands. We have developed the technology to enable energy reduction on commodity plug loads. Furthermore, we use the occupancy and device type information to enable the best possible policy for the building managers to implement.

Our Synergy Energy Meter measures plug load device power consumption and uses ZigBee for data transmission. It also has the ability to actuate, store device priority and device type. Our Energy Auditor Server uses the accumulated information and provides the appropriate mechanisms for the building managers to implement the most optimal policies.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

   Connectivity Brokerage: Enabling Seamless Cooperation in Wireless Networks

Pub ID:  347 Authors:  Arash Parsa, Jan Rabaey

The explosive growth in the density of wirelessly connected devices and their traffic load is creating interference and gradually leading to a severe spectrum shortage. Approaches to address this challenge include dynamic spectrum allocation (cognitive radio) and pro-active interference mitigation strategies requiring cooperation and collaboration between heterogeneous networking technologies. We have introduced the Connectivity Brokerage framework as a technology agnostic mediation layer that enables information exchange, cooperation and collaboration among heterogeneous wireless technologies and applications. A managed spectrum scenario is being used as the application driver for the CB framework, enabling researchers to develop key techniques for more advanced spectrum utilization models.

Nov 16, 2011,   MuSyC/GSRC Joint Review, Clark Kerr Campus, Berkeley, CA

You are not logged in