Huayong Wang

Transportation mode inference from anonymized and aggregated mobile phone call detail records

Transportation mode inference is an important research direction and has many applications. Existing methods are usually based on fine-grained sampling — collecting position data from mobile devices at high frequency. These methods can achieve high accuracy, but also incur cost and complexity in terms of the computational resource and system implementation. Finally, fine-grained sampling is not always available, especially for large-scale deployment. This paper proposes a novel method to infer transportation mode based on coarse-grained call detail records. The method allows estimating the transportation mode share from a given origin to a given destination, looking also at how the share changes over time. The method can achieve acceptable accuracy with trivial cost and complexity. It is suitable for the statistical analysis on transportation modes of a large population. The method can also be used as a complementary tool in situations where fine-grained sampling is unavailable or the balance between accuracy and complexity is critical. A case study using real call detail records data for the city of Boston shows the performance of the proposed method.

From a stream of relational queries to distributed stream processing

Applications from several domains are now being written to process live data originating from hardware and software-based streaming sources. Many of these applications have been written relying solely on database and data warehouse technologies, despite their lack of need for transactional support and ACID properties. In several extreme high-load cases, this approach does not scale to the processing speeds that these applications demand. In this paper we demonstrate an application acceleration approach whereby a regular ODBC-based application is converted into a true streaming application with minimal disruption from a software engineering standpoint. We showcase our approach on three real-world applications. We experimentally demonstrate the substantial performance improvements that can be observed when contrasting the accelerated implementation with the original database-oriented implementation.

A code generation approach for auto-vectorization in the SPADE compiler

We describe an auto-vectorization approach for the Spade stream processing programming language, comprising two ideas. First, we provide support for vectors as a primitive data type. Second, we provide a C++ library with architecture-specific implementations of a large number of pre-vectorized operations as the means to support language extensions. We evaluate our approach with several stream processing operators, contrasting Spades auto-vectorization with the native auto-vectorization provided by the GNU gcc and Intel icc compilers.

Skyeye: an instruction simulator with energy awareness

This paper presents a novel strategy aimed at modeling the instruction energy consumption of ARM microprocessors with dynamic voltage scaling (DVS) support. A novel energy estimation algorithm is designed, which can record the function calls, and generate a detailed energy profile for each function in a specific program. Some of the optimization policies for implementation are also discussed. These optimization policies reduce the workload of the energy estimators for the individual SOC systems. The prototype system, SKYEYE, can automatically detect the voltage/frequency variation activated by DVS system, and adjust the energy estimation model accordingly. The experiment results further prove the effectiveness of the algorithm.

A soft real-time web news classification system with double control loops

This paper proposes a framework for soft real-time text classification system, which use control theory as a scientific underpinning, rather than ad hoc solutions. In order to provide real-time guarantee, two control loops are adopted. The feed forward control loop estimates the suitable number of classifiers according to the current workload, while the feedback control loop provides fine-grained control to the number of classifiers that perform imprecise computation. The soft real-time classification system can accommodate to the change of workload and transitional overload. The theory analysis and experiments result further prove its effectiveness: the variation range of the average response time is kept within ± 3% of the desired value; the computational resource is dynamically reallocated and reclaimed.

Auto-vectorization through code generation for stream processing applications

We describe language- and code generation-based approaches to providing access to architecture-specific vectorization support for high-performance data stream processing applications. We provide an experimental performance evaluation of several stream operators, contrasting our code generation approach with the native auto-vectorization support available in the GNU gcc and Intel icc compilers.

Productivity and performance: improving consumability of hardware transactional memory through a real-world case study

Hardware transactional memory (HTM) is a promising technology to improve the productivity of parallel programming. However, a general agreement has not been reached on the consumability of HTM. User experiences indicate that HTM interface is not straightforward to be adopted by programmers to parallelize existing commercial applications, because of the internal limitation of HTM and the difficulties to identify shared variables hidden in the code. In this paper we demonstrate that, with well-designed encapsulation, HTM can deliver good consumability. Based on the study of a typical commercial application in supply chain simulations - GBSE, we develop a general scheduling engine that encapsulates the HTM interface. With the engine, we can convert the sequential program to multi-threaded model without changing any source code for the simulation logic. The time spent on parallelization is reduced from two months to one week, and the performance is close to the manually tuned counterpart with fine-grained locks.

Apply Feedback Control Theory to Design Soft Real-Time Search Engine System

This paper proposes a design method for soft real-time search engine system, and provides proofs to its correctness and robustness both in control theory and by practical experiments. An analyzable mathematical model is set up to approximately describe the nonlinear and time-varying search engine system. The feedback control theory is adopted to prove the system’s stableness, zero steady state error and zero overshoot. The soft real-time guarantee is satisfied while the feedback system is in stable state. The experiment results further prove the effectiveness of our scheme.