Haichuan Wang

Haichuan Wang

Auto-generated profile, based on publicly available information

Haichuan Wang

School of Computer Science and Technology, University of Science and Technology of China, Hefei, Anhui, China

Research Focus & Metrics

Academic Career Length

12Years (2015-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
17 December 2020 in IEEE Transactions on Mobile Computing
Verified
Journal Article
Closed access
MCMin Chen
Min Chen
HWHaichuan Wang
Haichuan Wang
ZMZeyu Meng
Zeyu Meng
HXHongli Xu
Hongli Xu
YXYang Xu
Yang Xu
HHe Huang
He Huang
Under the paradigm of edge computing, the enormous data generated at the network edge can be processed locally. To make full utilization of these widely distributed data, we focus on an edge computing system that conducts distributed machine learning using gradient-descent based approaches. To ensure the system’s performance, there are two major challenges: how to collect data from multiple data source nodes for training jobs and how to allocate the limited resources on each edge server among these jobs. In this paper, we jointly consider the two challenges for distributed training (without service requirement), aiming to maximize the system throughput while ensuring the system’s quality of service (QoS). Specifically, we formulate the joint problem as a mixed-integer non-linear program, which is NP-hard, and propose an efficient approximation algorithm. Furthermore, we take service placement into consideration for diverse training jobs and propose an approximation algorithm. We also analyze that our proposed algorithm can achieve the constant bipartite approximation under many practical situations. We build a test-bed to evaluate the effectiveness of our proposed algorithm in a practical scenario. Extensive simulation results and testing results show that the proposed algorithms can improve the system throughput 56-69 percent compared with the conventional algorithms.

23 October 2015 in ACM SIGPLAN Notices
Verified
Journal Article
Open access
PDF file
HWHaichuan Wang
Haichuan Wang
DPDavid Padua
David Padua
PWPeng Wu
Peng Wu
R is a popular dynamic language designed for statistical computing. Despite R's huge user base, the inefficiency in R's language implementation becomes a major pain-point in everyday use as well as an obstacle to apply R to solve large scale analytics problems. The two most common approaches to improve the performance of dynamic languages are: implementing more efficient interpretation strategies and extending the interpreter with Just-In-Time (JIT) compiler. However, both approaches require significant changes to the interpreter, and complicate the adoption by development teams as a result. This paper presents a new approach to improve execution efficiency of R programs by vectorizing the widely used Apply class of operations. Apply accepts two parameters: a function and a collection of input data elements. The standard implementation of Apply iteratively invokes the input function with each element in the data collection. Our approach combines data transformation and function vectorization to convert the looping-over-data execution of the standard Apply into a single invocation of a vectorized function that contains a sequence of vector operations over the input data. This conversion can significantly speed-up the execution of Apply operations in R by reducing the number of interpretation steps. We implemented the vectorization transformation as an R package. To enable the optimization, all that is needed is to invoke the package, and the user can use a normal R interpreter without any changes. The evaluation shows that the proposed method delivers significant performance improvements for a collection of data analysis algorithm benchmarks. This is achieved without any native code generation and using only a single-thread of execution.