UC Berkeley EECS Technical Reports

FastLane: An Agile Congestion Signaling Mechanism for Improving Datacenter Performance

2013-05-20T07:00:00Z

FastLane: An Agile Congestion Signaling Mechanism for Improving Datacenter Performance

David Zats, Anand Padmanabha Iyer, Randy H. Katz, Ion Stoica and Amin Vahdat

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-113
May 20, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-113.pdf

The drive towards richer, more interactive content places increasingly stringent latency requirements on datacenters. A critical component of meeting these is ensuring that the network responds agilely to congestion, bounding network latency and improving high-percentile flow completion times. We propose a new approach to rapidly detecting and responding to congestion. We introduce FastLane, a congestion signaling mechanism that allows senders to respond more quickly. By delivering signals to senders with high probability and low latency, FastLane allows them to retransmit packets sooner, avoiding resource-wasting timeouts. It also enables senders to make more informed decisions by differentiating between out-of-order delivery and packet loss. We demonstrate through simulation and implementation that FastLane reduces high-percentile flow completion times by over 80% by effectively managing congestion hot-spots. These benefits come at minimal cost - FastLane consumes no more than 2% of bandwidth and 5% of buffers.

Automatic Gesture Recognition and Tracking System for Physiotherapy

2013-05-18T07:00:00Z

Automatic Gesture Recognition and Tracking System for Physiotherapy

Aarthi Ravi

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-112
May 18, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-112.pdf

Physical Therapy is a branch of rehabilitative health that uses specially designed exercises and equipment to help patients regain or improve their physical abilities. Researchers have explored the use of software based rehabilitation tools based on video game consoles like Nintendo Wii Fit and Microsoft Kinect to provide remote rehabilitation facilities to patients. We propose a Kinect based Physical Therapy Application that guides patients to perform exercises and assists therapists to monitor as well as teach the specially designed exercises to patients. The application includes an Automated Exercise Generator for therapists to define exercises, an Automated Gesture Recognition and Tracking System to track and guide the patients while they perform the exercises and a Visual Feedback System for patients to correct the position and movement of their Joints. In this paper, we investigate how to enable continuous tracking of patients for pre-authored physiotherapy exercises. We introduce a state-machine-based approach that tracks a patient's progress and provides continuous feedback indicating whether the patient is doing an exercise correctly or not.

Investigation of Peptoid Thin Films and Their Potential Use in a Biosensor

2013-05-17T07:00:00Z

Investigation of Peptoid Thin Films and Their Potential Use in a Biosensor

Brian Lunt, Gloria Olivier, Felippe Pavinatto, Ronald Zuckermann and Ana Claudia Arias

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-111
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-111.pdf

Advisor: Ana Claudia Arias

Constraints And Techniques For Software Power Management In Production Clusters

2013-05-17T07:00:00Z

Constraints And Techniques For Software Power Management In Production Clusters

Arka Bhattacharya

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-110
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-110.pdf

The proliferation of large clusters supporting online web workloads or large compute-intensive jobs has made cluster power management very important~\cite{Barroso2007}. An analysis of utilization traces of production clusters reveal that a majority of them have a scope for (a) under-provisioning of electrical support infrastructure, leading to savings in {\bf capital expenditure}, and (b) energy savings, leading to savings in {\bf operational expenditure}; both with minimal impact on average job performance. Existing software techniques which tackle either of these problems have seen scant adoption because they do not address key problems and constraints relevant in production clusters. In this thesis, we first investigate possible reductions in cluster power infrastructure provisioning. It is possible that the lower provisioned power level is exceeded due to software behaviors on rare occasions and could cause the entire cluster infrastructure to breach the safety limits. A mechanism to {\it cap} servers to stay within the provisioned budget is needed, and processor frequency scaling based power capping methods are readily available for this purpose. We show that existing methods, when applied across a large number of servers, are not fast enough to operate correctly under rapid power dynamics observed in data centers. We also show that existing methods when applied to an open system (where demand is independent of service rate) can cause cascading failures in the software service hosted, causing the service performance to fall uncontrollably even when power capping is applied for only a small reduction in power consumption. We discuss the causes for both these short-comings and point out techniques that can yield a safe, fast, and stable power capping solution. Next, we address wasteful energy consumption by idle servers in an under-utilized cluster. Despite many clusters having a low average utilization, existing energy management techniques have seen scant adoption because they require modifications to the existing cluster software and network stack, and do not address the reliability concerns that may arise during the course of power-cycling servers in a production cluster. We design, implement and evaluate a defensive energy management system Hypnos, which is unobtrusive, efficient, extensible and gracefully handles possible server software and hardware failures.

Advisor: David E. Culler

Predicting Student Retention in Massive Open Online Courses using Hidden Markov Models

2013-05-17T07:00:00Z

Predicting Student Retention in Massive Open Online Courses using Hidden Markov Models

Girish Balakrishnan

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-109
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-109.pdf

Massive Open Online Courses (MOOCs) have a high attrition rate: most students who register for a course do not complete it. By examining a student’s history of actions during a course, we can predict whether or not they will drop out in the next week, facilitating interventions to improve retention. We compare predictions resulting from several modeling techniques and several features based on different student behaviors. Our best predictor uses a Hidden Markov Model (HMM) to model sequences of student actions over time, and encodes several continuous features into a single discrete observable state using a simple cross-product method. It yielded an ROC AUC (Receiver Operating Characteristic Area Under the Curve score) of 0.710, considerably better than a random predictor. We also use simpler HMM models to derive information about which student behaviors are most salient in determining student retention.

Advisor: Armando Fox

Speaker Diarization: Current Limitations and New Directions

2013-05-17T07:00:00Z

Speaker Diarization: Current Limitations and New Directions

Mary Tai Knox

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-108
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-108.pdf

Speaker diarization is the problem of determining “who spoke when” in an audio recording when the number and identities of the speakers are unknown. Motivated by applications in automatic speech recognition and audio indexing, speaker diarization has been studied extensively over the past decade, and there are currently a wide variety of approaches – including both top-down and bottom-up unsupervised clustering methods. The contributions of this thesis are to provide a unified analysis of the current state-of-the-art, to understand where and why mistakes occur, and to identify directions for improvements. In the first part of the thesis, we analyze the behavior of six state-of-the-art diarization systems, all evaluated on the National Institute of Standards and Technology (NIST) Rich Transcription 2009 evaluation dataset. While performance is typically assessed in terms of a single number – the diarization error rate (DER) – we further characterize the errors based on speech segment durations and their proximity to speaker change points. It is shown that for all of the systems, performance degrades both as the segment duration decreases and as the proximity to the speaker change point increases. Although short segments are problematic, their overall impact on the DER is small since the majority of scored time occurs in segments greater than 2.5 seconds. By contrast, the amount of time near speaker change points is relatively high, and thus poor performance near these change points contributes significantly to the DER. For example, for the single distant microphone (SDM) and multiple distant microphone (MDM) conditions, over 33% and 40% of the errors occur within 0.5 seconds of a change point for all evaluated systems, respectively. In the next part of the thesis, we focus on the International Computer Science Institute (ICSI) speaker diarization system and explore the effects of various system modifications. This system contains many steps – including speech activity detection, initialization, speaker segmentation, and speaker clustering. Inspired by our previous analysis, we focus on modifications that improve performance near speaker change points. We first implement an alternative to the minimum duration constraint, which sets the shortest amount of speech time before a speaker change can occur. This modification results in a 12% relative improvement of the speaker error rate for the MDM condition, with the largest improvement occurring closest to the speaker change point, and a 3% relative improvement for the SDM condition. Next, we show how the difference between the largest and second largest log-likelihood scores provides valuable information for unsupervised clustering, namely it indicates which regions of the output are likely correct. Lastly, we explore the potential of applying speaker diarization methodologies to other applications. Specifically, we investigate the use of a diarization-based algorithm for the problem of duplication detection, where the goal is to determine whether a given query (e.g., a short audio clip) has been taken from a reference set (e.g., a large collection of copyrighted media). With minimal modifications of the ICSI diarization system, we are able to obtain moderate performance. However, our approach is not competitive with existing approaches designed specifically for the problem of duplication detection, and the extent to which diarization-based approaches are useful for this application remains an open question.

Advisor: Nelson Morgan

Spontaneous Emission Rate Enhancement Using Optical Antennas

2013-05-17T07:00:00Z

Spontaneous Emission Rate Enhancement Using Optical Antennas

Nikhil Kumar

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-107
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-107.pdf

The miniaturization of electronics has relentlessly followed Moore’s law for the past several decades, allowing greater computational power and interconnectivity than ever before. However, limitations on power consumption on chip have put practical limits on speed. This dissertation describes the role that optical antennas can play in reducing power consumption and increasing efficiency and speed for on-chip optical interconnects. High speed optical communication has been dependent on the laser for its narrow linewidth and high modulation bandwidth. It has long outperformed the LED for both practical reasons and its suitable physical characteristics. Lasers however have some downsides when considering short distance communications which may not require narrow linewidths. Typically, they require high powers, take up more space and the rate is inherently limited by gain saturation. LEDs on the other hand are limited by spontaneous emission, a rate that is dependent upon its electromagnetic environment. The use of metallic optical nano-antennas can significantly increase a light emitters coupling to its environment and potentially achieve a rate orders of magnitude faster than stimulated emission. Coupling a light emitter into an efficient nano-optical antenna serves three purposes – 1) a much faster modulation speed can be achieved due to a faster rate of spontaneous emission, 2) the footprint of such a device would be shrunk to the nano-scale, ultimately necessary for large scale integration and 3) the overall efficiency of the emitter can be increased. While the main motivation behind this work is for short distance communications, optical antennas can serve in a host of applications including photodetectors, solar cells, nano-imaging, bio-sensing and data storage. In this thesis we derive the theory behind optical antennas and experimentally show an enhanced spontaneous emission rate of ~12.5x for bar antennas and ~30x for bowtie antennas.

Advisor: Eli Yablonovitch

Pebble Games and Complexity

2013-05-17T07:00:00Z

Pebble Games and Complexity

Siu Man Chan

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-106
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-106.pdf

We study the connection between pebble games and complexity. First, we derive complexity results using pebble games. It is shown that three pebble games used for studying computational complexity are equivalent: namely, the two-person pebble game of Dymond–Tompa, the two-person pebble game of Raz–McKenzie, and the one-person reversible pebble game of Bennett have the same pebble costs over any directed acyclic graph. The three pebble games have been used for studying parallel complexity and for proving lower bounds under restricted settings, and we show one more such lower bound on circuit-depth. Second, the pebble costs are applied to proof complexity. Concerning a family of unsatisfiable CNFs called pebbling contradictions, the pebble cost in any of the pebble games controls the scaling of some parameters of resolution refutations. Namely, the pebble cost controls the minimum depth of resolution refutations and the minimum size of tree-like resolution refutations. Finally, we study the space complexity of computing the pebble costs and of computing the minimum depth of resolution refutations. It is PSPACE-complete to compute the pebble cost in any of the three pebble games, and to compute the minimum depth of resolution refutations.

Advisor: Elchanan Mossel

User Interface: 3D Feedback

2013-05-17T07:00:00Z

User Interface: 3D Feedback

Seyed Hassan Elahi

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-105
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-105.pdf

In this project we examine the possibilities of bringing the gesture recognition to Windows Phone and Surface tablet devices by integrating Kinect to the aforementioned devices. We will demonstrate this by designing a physiotherapy application. In physiotherapy exercises, it is very important and critical to the patient to do the exercises as accurately as possible. So our physiotherapy application should have the ability to give feedback to the users in order for them to do the exercises correctly. For this purpose, we decided to render 3D arrows that show users how to adjust incorrectly positioned limbs. One way to do this is to use 3D animation on top of the skeleton image from the Kinect, so the user will know how and how much he or she should change the angle of the wrong joint to correct the position. This paper presents an approach for modeling shapes in an open source 3D application, Blender, and subsequently importing and animating those shapes in an interactive application created in the Microsoft XNA framework.

Advisor: Bernhard Boser

The Internals of GameTime: Implementation and Evaluation of a Timing Analyzer for Embedded Software

2013-05-17T07:00:00Z

The Internals of GameTime: Implementation and Evaluation of a Timing Analyzer for Embedded Software

Jonathan Prakash Kotker

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-104
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-104.pdf

Timing analysis is central to the design and implementation of cyber-physical systems. This thesis presents GameTime, a timing analysis toolkit that is based on a combination of game-theoretic online learning and systematic testing using satisfiability modulo theories (SMT) solvers. GameTime can be used to tackle a range of problems related to program timing, including estimating the worst-case execution time, predicting the distribution of execution times, and detecting timing-related anomalies. This thesis describes the details of the implementation of GameTime. The notion of basis paths is used to handle the exponentially many paths in a program. The issues that arise during the translation of statements in C programs to the equivalent clauses in SMT queries are presented, and the techniques used by GameTime to address these issues are elaborated through examples. Finally, experimental results demonstrate the speed of GameTime analysis and the accuracy of the predictions made by GameTime, with a relative error margin of less than 5% on most of the benchmarks measured.

Advisor: Sanjit A. Seshia

User-Guided Inverse 3D Modeling

2013-05-17T07:00:00Z

User-Guided Inverse 3D Modeling

James Andrews

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-103
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-103.pdf

This thesis introduces and explores the idea of ``user-guided inverse 3D modeling'' -- an interactive approach to shape construction and redesign that extracts well-structured, parameterized, procedural descriptions from unstructured, hierarchically flat input data, such as point clouds, boundary representation meshes, or even multiple pictorial views of a given inspirational prototype. This approach combines traditional ``forward'' 3D modeling tools with a system of user-guided extraction modules and optimization routines. With a few cursor strokes users can express their preferences regarding the type of modeling primitives to be used in a particular area of the given prototype to be approximated, and they can also select the degree of parameterization associated with each modeling routine. The results are then pliable, structured descriptions that are well suited to implement the particular design modifications intended by the user. We present research on the components that make up an inverse 3D modeling system. Our research focuses have been (1) fitting of modeling primitives to data; (2) fitting higher-level structure (CSG expressions, symmetry, and hierarchy) to data; (3) minimal user interactions that guide these fitting processes; and finally (4) shape editing using the fitted structure. We also explore what is necessary to ultimately arrive at a clean, exportable result. A key research question guiding the design of our system has been how bringing the user into the loop affects the problem: i.e., where can the user's inputs simplify, accelerate, or otherwise improve the fitting process? What simple inputs can the user provide to unambiguously extract a desired primitive fit? And how can we let the user immediately begin using their fitted primitives to edit the shape? Answering these questions leads us to many new research findings, including: (1) We identify simple, lightweight inputs that allow the system to extract primitives with a great deal of user control, and we present detailed analysis of how these inputs can be ambiguous and what additional inputs may be needed to exactly capture the user's desired primitive fit. (2) We show multiple improvements to the state of the art in fast, effective fitting methods for sweeps and quadric surfaces. These improvements include identifying and fixing a major problem with previously-standard methods for kinematic surface fitting that led to grossly wrong results for data with small noise, and a new insight into the nature of direct, algebraic fitting that leads us to more accurate type-specific quadric fitting. (3) We show how the user can guide a boundary-to-CSG reverse-engineering process: Specifically, we identify cases where the best result of such a process depends on the user's intent, and we show how simple user interactions can let users specify their intents. (4) In addition to the diverse problems faced by individual structure-fitting and shape editing modules, the combination of these modules also presents new research opportunities; we identify new interactions between these diverse modules, including transformations between primitive types and hierarchies that can be imposed across modules.

Advisor: Carlo H. Séquin

Autonomous Navigation and Collision Avoidance Robot

2013-05-17T07:00:00Z

Autonomous Navigation and Collision Avoidance Robot

Pengqi Cheng

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-102
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-102.pdf

In the last 50 years, information technology has automated many jobs like word processing, mathematical calculation and data retrieval. However, as driving a car is much more difficult than such simple jobs, it has always been a dream to do so autonomously with a computer. With the development of related technologies, people have begun significant work on making this dream a reality. For automakers, it is more profitable to invest in a mature market -- semi-automatic vehicles. However, this project aims to develop a fully automated driving system that is profitable in the market. To accomplish this, this project uses a single personal computer instead of computer clusters as in experimental projects like Stanford's Junior. This opens up the possibility of migrating the software into high-end smartphones due to their gradually increasing performance. In this case, we simplified existing algorithms and still keep the basic functionality, but we still have to make sure that our system is safe enough for passengers. In addition, this project only uses cheap sensors. For instance, the video cameras are low-end webcams and cameras in smartphones rather than high-end professional cameras. Thus, the definition of our sensors is lower and the responding time may be longer. So we have to use multiple sensors to reach redundancy and robustness. This paper discusses a subset of the larger autonomous vehicle project. In particular, it includes the GPS driver module migration and Orca Robotics middleware integration.

Advisor: Pieter Abbeel

Design a Natural User Interface for Gesture Recognition Application

2013-05-17T07:00:00Z

Design a Natural User Interface for Gesture Recognition Application

Zhaochen Liu

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-101
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-101.pdf

Gesture recognition allows users to interact with their electronic devices in a manner that mimics their interaction with the physical world, thus making devices easier to use for all users. Interacting with the software, using gesture and body motion is definitely a cool ideal. However, designing effective interactions is a challenge on a gesture-controlled application. We partnered with the Microsoft Windows Phone Team to develop a Kinect physiotherapy application. This application enables the patients to perform recovery exercises remotely at home. It checks the patient’s body movements against doctors’ instructions and provides real-time instructions to the patients. In order to validate the market value for such an application and further enhance this prototype, we conducted market studies and user tests. In particular, I was in charge of the development of the user interface in this project. Also, I developed guidelines for designing the user interaction of gesture-controlled applications.

Advisor: Björn Hartmann

Communication-Avoiding Parallel Recursive Algorithms for Matrix Multiplication

2013-05-17T07:00:00Z

Communication-Avoiding Parallel Recursive Algorithms for Matrix Multiplication

Benjamin Lipshitz

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-100
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-100.pdf

Advisor: James Demmel

Design and Analysis of Digitally Modulated Transmitters for Efficiency Enhancement

2013-05-17T07:00:00Z

Design and Analysis of Digitally Modulated Transmitters for Efficiency Enhancement

Lu Ye

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2013-99
May 17, 2013

http://www.eecs.berkeley.edu/Pubs/TechRpts/2013/EECS-2013-99.pdf

Advisor: Ali Niknejad