Shi Dianxi

Research and Implementation of Design Pattern-Oriented Model Transformation

Design pattern describes a repeatedly presenting issue during software designing, as well the solution to it. Applying design pattern enables developers to reuse it to solve a specified designing issue. Applying and maintaining design pattern in models is a duplicating work. In order to automatically apply design patterns in models, we propose a design pattern oriented model transformation approach. In this paper, we designed a set of atom mappings and pattern mappings, as well the combination of them, the model transformation code template. Such a template can generate XSLT based model transformation code, and enable automatically applying design pattern in software models. A case study is provided to demonstrate our approach.

Bytecode-based software monitoring and trusted evolution programming framework

Point to reliability of distributed software in open complex environment, this paper proposes a bytecode-based software monitoring and trusted evolution programming framework, which injects monitoring ability and evolution ability into the software in construction and running phase. The framework is comprised of three module, business-logic and monitor-requirement description module (BMDM), monitoring capability infusing module (MCIM), and online evolution module (OEM), BMDM provides monitoring requirement expression language MRL+, MCIM compiles monitoring requirements monitoring requirement into aspect code, insert aspect code into source system by bytecode-based AOP technology, generate the target system with monitoring ability, and send monitoring information to OEM by event. OEM compares the state event with predefined evolution rules, when the predefined rules are triggered, perform the evolution actions, achieve real-time systems monitoring and correction. Finally, actual cases are analyzed to show that the framework can construct reliability software with monitoring ability and evolution ability efficiently.

Extend the Service and Component in Meta-level for Pervasive Computing

Powerful and cheap computing devices are flooding the world, which arises the primary challenge for pervasive computing: how to represent devices and their functions without distracting user attention. Most existing pervasive computing platforms use traditional components and services as the answer. However, both of them have limitations. Existing services in the form of Interface are small in granularity, lacking of semantics and distracting to invoke remotely. Existing components are lacking of a unified model for representing dependencies and a mechanism for dealing with these dependencies, especially the runtime dependency. In this paper, we extend the Service and Component in meta-level for pervasive computing. Firstly, we extend them in meta-model, making the Service as the aggregation of Interface. Secondly, we entitle semantics to services by applying ontology and provide a new service invoking mechanism. Thirdly, we wrap Component by the enhanced Service to decouple the compile-time and runtime dependencies; and introduce a novel group interaction mechanism for multi-service interaction. Finally, a prototype platform which applies enhanced Service and Component is presented; and two demos are used to show how the extended Service and Component work and benefit.

An Early-Delivery Alogrithm for Consensus-based Atomic Broadcast

Although a large number of consensus and atomic broadcast algorithms based on failure detectors have been published during the years, they are still criticized for poor performance in terms of communication latency. To reduce the latency, some algorithms based on optimistic assumptions have been proposed. When the optimistic assumption holds, these algorithms deliver messages more quickly than other ones. However, if the assumption does not hold, the efficiency of these algorithms deteriorates. In this paper, we propose an early-delivery atomic broadcast algorithm which exploits the optimistic assumption in a way that never leads to high latency. This is achieved by attaching information to all sending messages and collecting the information from communication history records. The simulation experiments show that our algorithm is a good trade-off between conservative ones and optimistic ones