Debugging large-scale, data-intensive, distributed applications running in a datacenter (“datacenter applications”) is complex and time-consuming. The key obstacle is non-deterministic failures—hard-to-reproduce program misbehaviors that are immune to traditional cyclic debugging techniques. Datacenter applications are rife with such failures because they operate in highly non-deterministic environments: a typical setup employs thousands of nodes, spread across multiple datacenters, to process terabytes of data per day. In these environments, existing methods for debugging non-deterministic failures are of limited use. They either incur excessive production overheads or don’t scale to multi-node, terabyte-scale processing. To help remedy the situation, we have built a new deterministic replay tool. Our tool, called DCR, enables the reproduction and debugging of non-deterministic failures in production datacenter runs. The key observation behind DCR is that debugging does not always require a precise replica of the original datacenter run. Instead, it often suffices to produce some run that exhibits the original behavior of the control-plane—the most error-prone component of datacenter applications. DCR leverages this observation to relax the determinism guarantees offered by the system, and consequently, to address key requirements of production datacenter applications: lightweight recording of long running programs, causally consistent replay of large-scale clusters, and out-of-the box operation with existing, real world applications running on commodity multiprocessors.

关键词：分布式应用程序；数据中心应用程序；故障排除；数据中心；确定性重放工具（DCR）

查看摘要 索取原文[98页]

Radios available today are designed to be high performance devices. These radios require careful design by experienced and skilled RF IC designers, more expensive RF processes, and large chip areas for RF passives. The resulting cost of these devices is at the dollar level without the off chip components, and the careful design required makes integration of these radios with other circuits (microprocessors, sensors, etc) an expensive proposition. We believe that radios that require limited design skills while still having good performance will enable widespread use of wireless technologies.

关键词：无线电；发射机；振荡器；微处理器；传感器

查看摘要 索取原文[75页]

Benefits of technology scaling and the flexibility of digital circuits favor the digital signal processing in many applications, placing additional burden to the analog-to-digital con- verters (ADCs). This has created a need for energy-efficient ADCs in the GHz sampling frequency and moderate effective resolution range. A dominantly digital nature of successive approximation register (SAR) ADCs makes them a good candidate for an energy-efficient and scalable design, but its sequential operation limits its applicability in the GHz sampling range. Time-interleaving can be used to extend the efficiency of the SAR ADCs to the higher frequencies if the mismatches between the interleaved ADC channels can be handled in an efficient manner. New calibration techniques are proposed for time-interleaved SAR ADCs capable of cor- recting the gain, offset and timing mismatches, as well as the static nonlinearities of individ- ual ADC channels stemming from the capacitor mismatches. The techniques are based on introducing two additional calibration channels that are identical to all other time-interleaved channels and the use of the least mean square algorithm (LMS). The calibration of the chan- nel offset and gain mismatches, as well as the capacitor mismatches, is performed in the background using digital post-processing. The timing mismatches between channels are cor- rected using a mixed-signal feedback, where all calculations are performed in the digital do- main, but the actual timing correction is done in the analog domain by fine-tuning the edges of the sampling clocks. These calibration techniques enable a design of time-interleaved con- verters that use minimum-sized capacitors and operate in the thermal-noise-limited regime for maximum energy and area efficiency. The techniques are demonstrated on a time-interleaved converter that interleaves 24 channels designed in a 65nm CMOS technology. The ADC uses the smallest capacitor value of only 50aF, achieves 50.9dB SNDR at fs = 2.8GHz with the effective-resolution bandwidth higher than the Nyquist frequency, while consuming only 44.6 mW of power.

关键词：技术缩放、数字电路、时间交叉特区、校准技术

查看摘要 索取原文[115页]

We build mobile healthcare applications o help diabetics better monitor their condition, save healthcare cost and improve the quality of ife. Among the solutions we investigated was a prediction algorithm that estimates how a person's lucose levels will change based on speci c food and exercise inputs. Another solution was a recom- endation system that suggests diabetic-friendly food options when a person shops for groceries or ats at a restaurant.

关键词：血糖水平；糖尿病；移动医疗；医疗器械；移动应用

查看摘要 索取原文[29页]

This thesis presents three new computational tools that bring the strength of hybrid dynamical models and optimal control to applications in Cyber-Physical systems. The first tool is an algorithm that nds the optimal control of a switched hybrid dynamical system under state constraints, the second tool is an algorithm that approximates the trajectories of autonomous hybrid dynamical systems, and the third tool is an algorithm that computes the optimal control of a nonlinear dynamical system using pseudospectral approximations. These results achieve several goals. They extend widely used algorithms to new classes of dynamical systems. They also present novel mathematical techniques that can be applied to develop new, computationally efficient, tools in the context of hybrid dynamical systems. More importantly, they enable the use of control theory in new exciting applications, that because of their number of variables or complexity of their models, cannot be addressed using existing tools.

关键词：物理系统；大规模分布式系统；最优控制配方；混合动力系统；控制理论

查看摘要 索取原文[154页]

A multi-stage compilation framework transforms portions of programs written in a productivity-level language into an efficiency-level language, such as C, with explicit hardware-specific optimizations. It is challenging for compiler programmers to debug errors in the compilation because they must perform complicated end-to-end reasoning, relating the programs across the multiple stages of compilation. To simplify this debugging effort, we present multi-level debugging, a novel combination of error-checking algorithms in a multi-stage compilation environment. Our method particularly aims to model and check sequential and parallel notions of nondeterminism and related bugs introduced by the compilation. Using our method, the programmer can systematically eliminate potential sources of the bug in the compilation process and focus only on the real source. We demonstrated on two real multi-stage compilers the effectiveness of multi-stage debugging in simplifying the diagnosis of manually-injected bugs as well as in an actual bug encountered during compiler development.

关键词：编译器程序员；多级并行化；多等级；程序调整

查看摘要 索取原文[12页]

We examine how analysis tools can be tailored to scaffold novice users into the process of data analysis, encouraging participation and understanding while contributing valuable local insights. Finally, we explore mechanisms for scaling and parallelizing data analysis, even in the absence of a dedicated community or team of analysts. We investigate how individual analysts can crowdsource pieces of social data analysis tasks using paid workers in order to leverage the collective effort of many participants. We demonstrate how large groups of workers can perform cognitively complex tasks like generating and rating hypotheses, and provide tools to help analysts manage the results of this process. These tools and strategies, along with our evaluations of them, highlight the potential of social data analysis in a variety of settings with different kinds of stakeholders. Moreover, our findings suggest leverage points for future social data analysis systems.

关键词：数据分析；定制工具；策略；认知复杂任务

查看摘要 索取原文[140页]

Web applications are generally more exposed to untrusted user content than traditional applications. Thus, web applications face a variety of new and unique threats, especially that of content injection. One method for preventing these types of attacks is web application security policies. These policies specify the behavior or structure of the web application. The goal of this work is twofold. First, we aim to understand how security policies and their systems are currently applied to web applications. Second, we aim to advance the mechanisms used to apply policies to web applications. We focus on the first part through two studies, examining two classes of current web application security policies. We focus on the second part by studying and working towards two new ways of applying policies. These areas will advance the state of the art in understanding and building web application security policies and provide a foundation for future work in securing web applications.

关键词：Web应用程序；安全政策；用户；威胁

查看摘要 索取原文[111页]

Recent results have demonstrated the performance bene ts of communication- avoiding Krylov subspace methods, variants which use blocking strategies to perform O(s) com- putation steps of the algorithm for each communication step. This allows an O(s) reduction in total communication cost, which can lead to signi cant speedups on modern computer architectures. De- spite potential performance bene ts, stability of these variants has been an obstacle to their use in practice.

关键词：Krylov子空间方法；剩余更换；数值稳定；迭代求解；稀疏矩阵

查看摘要 索取原文[24页]

Our main focus here is on demonstrating properties of LBMPC controllers on the quadrotor testbed, with an implementation of LBMPC that runs in real-time on the quadrotor's on-board computer. Robustness to “mis-learning” is one aspect of LBMPC that we demonstrate in an experiment where we deliberately mis-tune a learning algorithm. We also demonstrate the improvement in performance possible when a well-tuned learning algorithm is used, show learning used to update the model in a physically meaningful way, and demonstrate the use of the LBMPC controller in an integrated robotic task requiring speed and precision: we design a controller that enables the quadrotor to catch balls.

关键词：物理；车载模型；控制器；机器人；四旋翼直升机

查看摘要 索取原文[80页]