For future HPC systems which embody tens of thousands of processors, it will be unfeasible to construct a fully-connected packet switching network with high bisection bandwidth due to cost and power consumption, etc. We propose a low cost hybrid network which exploits an EPS network whose bandwidths of upstream links are narrow and an OCS network. Moreover, we propose a forwarding based communication methodology where optical pathways are used as shortcuts of EPS upstream links for MPI applications. From results of simulation, we confirmed that combination of our network with our communication methodology achieved better performance than that of full bisection EPS only network with only small number of optical pathways in case of communication pattern which has locality. We also confirmed that in case of all-to-all communication, which does not have locality, the combination achieved almost the same performance as that of EPS only network whose upstream link bandwidth is twice larger than that of EPS network in our network.
Many Web applications provide information in table formats that are useful for data comparison (e.g. comparison of hotel plans for hotel search applications). Using mashup technologies, one can add niche criterion to meet various user's requests, but large number of columns will decrease the visibility of the table and may incur many needless requests to Web services. Interactive selection of columns and filtering facilities may be helpful, but the implementation of these facilities on bare AJAX framework will require careful treatments of asynchronous requests. This paper presents a framework for Web application development that provides easy creation of mashups and AJAX widgets for data browsing. One can create mashups only by preparing configure file to attain mashup data model that works in demand-driven manners for interactive access via widgets.
We develop StrutsLint which is a coding checker for Web application based on Struts framework. StrutsLint parses JSP, Java and configuration files of Struts, and generates control and data dependency graph among languages. StrutsLint checks consistency of target application by evaluating consistency of constraint and so on. Through presenting StrutsLint functions, this paper describes about consistency check for Struts application as a component-ware.
Multivariate regression models have been commonly used to estimate the software development effort to assist project planning and/or management. Since project data sets for model construction often contain missing values, we need to build a complete data set that has no missing values either by using imputation methods or by removing projects and metrics having missing values (removing method). However, while there are several ways to build the complete data set, it is unclear which method is the most suitable for the project data set. In this paper, using project data of 706 cases (47% missing value rate) collected from several companies, we applied four imputation methods (mean imputation, pair-wise deletion, k-nn method and applied CF method) and the removing method to build regression models. Then, using project data of 143 cases (having no missing values), we evaluated the estimation performance of models after applying each imputation and removing method. The result showed that the similarity-based imputation methods (k-nn method and applied CF method) showed the best performance.
Enterprises must flexibly change their business processes in response to changes in their business environment to survive in the intensely competitive society in these days. To secure this flexibility, it is essential to share domain knowledge why they are using present business processes and competitiveness of their own business process. In this paper, we propose a new method to compose a business process by stepwise application of goal decomposition from formal description of a business goal, and we demonstrate that business processes composed by the method can be verified to satisfy given business requirements. This method is applicable to composition of planning business processes whose formalization have been thought as a difficult task. We also show that we can formalize business domain knowledge, and that business goal decomposition can be patternized.
We propose a web annotation system that has functionalities of stickies on web pages and bidirectional links between the stickies. The stickies allow for important parts of a web page which contains large amounts of data to be highlighted. We have implemented a positioning method based on the Document Object Model (DOM), as well as a new method for placing stickies which depends on web contents related to the information referenced by the stickies. By using web agents, the system automatically generates bidirectional links between stickies referencing similar information and subsequently categorizes them. Such stickies and links can be used as user preferences, and have the potential to become a much better alternative to bookmarks and tags. The preferences proposed here can be applied to develop a new search engine and a recommendation system.
Program dependence graphs (PDGs) represent data and control dependences between statements in a program. PDGs should have formal semantics which is equivalent to that of programs. Although a few formal semantics of PDGs have been introduced, they could only treat structured programs. In this paper, we present an operational semantics of PDGs, corresponding programs of which have complicated control flow. Then, we define a class of deterministic PDGs and prove that programs whose PDGs are deterministic PDGs have the same meaning as the corresponding PDGs.
This paper presents a method to compile the Standard ML module language into a flattened intermediate language. An innovative point on this approach lies in viewing a functor as a code template with place holders representing the functor arguments. Each functor application fills these place holders in the code template with the actual functor argument and generates a fresh structure. After this compilation, module language constructs are all eliminated. This method allows us to compile the full set of Standard ML language into a typed intermediate language that contains no special mechanism for module, and provides a simpler model for separate compilation. The proposed compilation method has been successfully implemented in the SML# compiler, which demonstrates the feasibility of the method. This paper also reports the details of our implementation.
Statechart is a powerful modeling language that is taken in UML. We propose a verification algorithm by abstraction and refinement of hierarchical structure of statechart. The algorithm performs reachability analysis on an abstract model. Subsequently, if the counterexample exists, the algorithm refines an abstract model using it. Moreover, the algorithm checks the validity of counterexample and performs reachability analysis again when it is a spurious counterexample. The algorithm repeats these procedures automatically until a solution is provided. We have implemented a prototype of our verification algorithm and confirmed that there is an effect to downsize state space by experiments.