Practical systems often need complicated design descriptions because they refer tosituations concerning exceptional cases in addition to normal, regular behaviour. Sinceproperties refer to a lot of exceptional cases, the property formulae to beverified become complicated and hard to understand. Further, the state spacesearched in the model-checking process becomes large, which affects the executionperformance. This paper employs a model-checking technique based on multi-valuedtransition systems so that we can reduce such complexities.

View full abstract

In this paper, we propose a new component extraction technique and for object-oriented programs. Our technique can specify the region that should be packaged into one component, and extract reusable components from existing programs based on the statical analysis of object-oriented class relations. Moreover, using the refactoring technique, our technique can automatically modify classes where the extracted region had been used. The modified classes become the usage example for the new component.

View full abstract

We have proposed a formal model for object-oriented analysis so far. This allows us to verify constructed analysis models and support it by computer.In this paper, we show a verification environment called F-Developer that we have implemented.

View full abstract

SSA form is one of useful intermediate representations for optimizing compilers.But it is necessary to translate out of SSA form.That translation is called "normalization".There are two well-known algorisms.We compare these algorisms.

View full abstract

Program slicing has been used for efficient program debugging activities. In general, slice is computed by analyzing dependency relations between program statements and slice computation algorithms are mainly about static slice and dynamic slice. We proposed DC slice which combined both of techniques.In this paper, we will propose a DC slicing method for Java program using the Java Virtual Machine. This system can compute fine-grained analysis results based on bytecodes by extracting a dependency dynamically during execution of a Java Virtual Machine.

View full abstract

It was difficult to formulize optimizers of compilers by Attribute Grammars, when the language includes unstructured jumps such as gotos. This is because pure AGs do not allow remote attribute accesses and circularity in attribute dependency that arise from goto jumps. To tackle this, we proposed an extension of AGs called circular remote attribute grammars and its evaluators, which eases the above restrictions of pure AGs. In this research, an optimizer for partial redundancy elimination is implemented with our extended AGs. We show that the lazy code motion method, based on solutions of data-frow equations, can be written with AG descriptions in a natural way.

View full abstract

A polymorphic type inference algorithm is proposed by Tofte.However, this algorithm has restriction on polymorphic typevariables. We present a new type inference algorithm to ease this restriction byusing the value graph used to process optimization of the compiler.

View full abstract

We propose a technique to generate efficient code reusing resultsof method lookup . This approach eliminates redundant executionof the method lookup using the partial redundancy elimination(PRE) technique. PRE regards lexically identical expressions asthe candidates of the redundancy. So, we present here how todetect candidates of the redundant lookup based on reduction ofthe type propagation graph. Since redundancy elimination is basedon only equalities among types of receiver objects while methodinlining techniques require the unique type for receiverstatically, our technique can complements the previousworks. Furthermore, our technique does not increase target codeas much as method inlining, so it would play an important rolefor application programs used by embedded systems.

View full abstract

Higher-order rewrite systems are widely used computationalsystems having various application including a model offunctional programming languages, proof checkers and etc.This work proposes a kind of higher-order rewrite system,called binding term rewriting systems. First based on therecent work on mathematical semantics of abstract syntaxwith variable binding, we define a type theory for termsinvolving the precise notion of variable binding. The termsin the type theory are similar to first-order terms ratherthan lambda terms. Then we define rewrite systems on it andshow that termination of this kind of rewrite systems can beshown by using techniques for first-order term rewritingsystems.

View full abstract

This paper presents a method of computing minimal revised specification written in function-free Horn theory. We previously proposed a formalization of minimal revision of logical specification. However, the proposed computational method needed a minimality check of abducibles corresponding with revision in order to get a minimal revised specification. In this paper, we propose a method which directly computes a minimal revised specification without minimality check. The technique is an adaptation of computational method of circumscription which was proposed by the author. We show the correctness of the method, and also show thecompleteness of the method for a restricted class of Horn clauses.

View full abstract

Quasi-quotation, introduced by Quine, is a mechanism to construct aname for an indeterminate expression containing metavariables whosevalues are expressions. To formalize the mechanism, we extend thesimply-typed λ-calculus and propose a typed calculus λq that has the notion of quotation, evaluation, and metavariable.

View full abstract

The random access memory (RAM) model, widely used in computational complexity theories,has become quite unrealistic due to deepening memory hierarchy in modern computers.In addition, parallel computers , clusters, and globally distributed computing environments have become popular, where communication delays between nodes and their differences can not be neglected. In spite of this, general and useful computational complexity models thatcan handle different network topologies and differences of communication delays do not yet exist.We propose the “Computing Continuum Model” -- a computation complexity model thatcan represent the memory hierarchy and network communication in the same abstraction.In this paper, we shows an abstract machine on our model.

View full abstract

Although there have been major improvements in Java performance from its beginning, it is still behind that of Fortran for numerical computing. One of the major reasons for this poor performance is due to the difficulty of optimizing floating-point computations with the strict accuracy requirements of Java. In particular, the inability to use fused multiply-add (FMA) instructions and reassociations has a large impact on Java performance. In this paper, we propose a new technique called floating-point (FP) speculation. In FP speculation, two (speculative and verification) versions of the same floating-point computation run in parallel. After the completion of the verificationcomputation, the compensation code runs if the results are not identical. With FP speculation, we can optimize floating-point computations as though they were written in Fortran while preserving the strict accuracy requirements of Java.

View full abstract

In this paper, we discuss an implementation of an optimizer using Static Single Assignment (SSA) form on compiler infrastructure COINS. Three methods of building SSA form and two methods of translating out of SSA form are implemented on this optimizer. Further, we have been implementing some basic optimizations such as conditional constant propagation, common subexpression elimination, dead code elimination and so on.As the advantage of our SSA optimizer, new methods and optimization based on SSA form can be easily implemented on this optimizer as infrastructure.

View full abstract

Merging continuously processed instructions in a super-instruction isan efficient technique for improving the performance of interpreters.The article proposes an algorithm which merges instructions on thelevel of operational semantics. By combining the algorithm with theinterpreter generator which we proposed previously, we can generateefficient interpreters automatically which are specialized intoapplication programs. Preliminary results show that our schemeeffectively improves the execution speed of application programs to 210% of originally interpretation, that is 16% better than the resultof previously proposed scheme.

View full abstract

In this paper we present a resolution system for knowledge bases (TBox and ABox) in DL (Description Logic) with clausal sets. In order to achieve reasoning upon such knowledge bases consisting of TBox- and ABox-statements and clauses, we consider an extension of the DL resolution proposed by Areces. This extended resolution is obtained by incorporating inference rules for the composition of ABox-statements and clauses and for TBox-statements, and unification for first-order terms.

View full abstract

Descriptive set theory is used to compare the expressive power of sublogics of CTL*. One application is a simple proof of the fact that no equivalent CTL formula exist for the CTL* formula EGF(p).The proof consists of showing that in a suitable semantics CTL formulas correspond to Borel sets while [[EGF(p)]] is analytic but not Borel.

View full abstract

Kozen and Smith show existence of the free Kleene algebra with testsgenerated by a pair (B, Σ) of finite sets. They give an explicitconstruction of Kleene algebra with tests P_B,Σ,and prove that a subalgebra of P_B,Σ is the freelygenerated by (B, Σ). We show the existence of free algebrawithout assuming the finiteness condition. Moreover we provide aconstruction of Kleene algebra with tests Q_B,Σ bymeans of standard constructions such as coproducts and adjunctions andshow that Q_B,Σ is isomorphic to P_B,Σ, whenever P_B,Σ is defined, i.e., B is finite.

View full abstract

We propose LMNtal, a simple concurrent language model based onthe rewriting of nested graphs with links represented as logical variables.LMNtal aims at unifying various computational models based on multisetrewriting and serving as the basis of general-purpose programming languagesthat cover the spectrum ranging from wide-area to nanoscale computation.LMNtal also facilitates the programming of dynamic data structures.

View full abstract

The problem whether a given equation is an inductive theorem onrecursive data structures like natural numbers or lists is undecidablein general. In this paper, we propose a general framework forcharacterizing decidable classes of inductive equational theorems. Byapplying the rewriting induction technique, we present a simpledecision procedure for inductive theorems. We show that it is decidablewhether a given simple equation is an inductive theorem of a confluent,terminating and sufficient complete term rewriting system.

View full abstract

Context-sensitive rewriting is a restriction of term rewriting used to model evaluation strategies infunctional programming and in programming languages like OBJ. For example, under certain conditionstermination of an OBJ program is equivalent to innermost termination of the corresponding contextsensitiverewrite system. To prove termination of context-sensitive rewriting, several methods havebeen proposed in the literature which transform context-sensitive rewrite systems into ordinary rewritesystems such that termination of the transformed ordinary system implies termination of the originalcontext-sensitive system. None of these transformations is very satisfactory when it comes to provinginnermost termination. We present a simple transformation which is both sound and complete forinnermost termination.

View full abstract

The UML statechart diagrams is widely used modeling language for describing behavioural aspect of software systems. Its rich notation of structurization makes the description brief but difficult to be well-formed and consistent with the describer's notation. We present a procedure to check on well-formedness of UML statechart diagrams. The procedure checks the existence of conflicting transitions, inappropriate history indicators and complete transitions while constructing model graphs of UML statechart diagrams for LTL model checking. These checks contribute to removal of both ambiguous and undefined descriptions of behaviour. After the construction of the well-formed model, the procedure executes model checking with specifications written in LTL for the consistency between the behaviour of the system and the describer's intention.

View full abstract

To make an animation that adaptively responds to user inputs or animationcircumstance, we sometime have to control the animation by programs.To write such programs requires detailed knowledge of 3D-CG programming with careful object manipulations. Our goal is to provide a programming environmentfor 3D-CG animation that features (1) a animation API based on a sophisticated concurrent model, and (2) a GUI programming environment that can automaticallygenerate program codes from the corresponding GUI operations.

View full abstract

We present a geometric constraint solving method for interactive 3D applications. Its novel feature is to handle scene graphs by processing coordinate transformations in geometric constraint satisfaction. It enables programmers to use 3D geometric constraints for various purposes such as geometric layout, constrained dragging, and inverse kinematics. We demonstrate its usefulness by presenting a sample interactive 3D application using Chorus3D, a constraint solver which implements our method.

View full abstract

This paper describes the methods of easy-to-understand programs visualization in 3D Visual Programming system “3D-PP&rdquo. 3D-PP visualize GHC program with virtual 3D space and 3D objects.To enable programers to detect arguments and input and output mode of arguments, we introduce “Argument objects&rdquo. To suppress dispersion of objects, we introduce a structuring method to enable a 3D object to contain a group of objects, called “Hierarchical program structure”. We implemented “Argument objects” and “Hierarchical program structure” in 3D-PP.

View full abstract

We designed a visual browser for log browsing, adopting the Hyperbolic Tree and Hyperbolic Cell to visualize log data. The Hyperbolic Tree provides a whole view of log data, showing quantitative change in the log data by focusing on a specified part. The Hyperbolic Cell expresses the time change of log data. This enables real-time investigation and comparison between log attributes. Our visual browser allows us to browse effectively, helping to check users' behavior within the network and to find traces of intrusion.

View full abstract

ClassFactory is a development support system that allows multiple programmers distributed over the network to participate in the development of a single software system. ClassFactory has a built-in synchronization mechanism to avoid conflicts among activities of different programmers to a single source file, by locking the file. However, the file-based locking mechanism does not allow multiple programmers to work on a single file simultaneously even if they work on different parts of the file, which will result in low efficiency and parallelism of development activities. This paper will propose a smaller grain locking mechanism that will achieve high parallelism of development activities, and verify its validity for actual opensource development projects.

View full abstract

Some of software are distributed in binary format.There is no need to compile software distributed in binary format.However, the software is not customizable flexiblysince it is impossible to modify compile-time option of the software.To make it possible, we propose a new method where hint information for customization is embedded in binary format.When the software is installed, we can customize the software by using the embedded information.We also implement the method.

View full abstract

Aspect-oriented languages support clear separation of code which crosscuts two or more classes. We believe it is possible to combine independently developed aspects safely, if each aspect is designed according to certain rules. As first step for finding out and verifying such rules, we describe how to verify rules for safely composable mixins. The notion of "Design by Contract" and "behavioral subtyping" are used for description and verification of the rules.

View full abstract

We report middleware named UnIt System(Universal Interaction System) that offers flexible user interface for controlling home appliances and applications in Ubiquitous Computing. If you use UnIt System in home computing, you can control home appliances in a uniform way, and change control devices dynamically by context information.

View full abstract

In recent years, a computer comes to be carried in household-electric-appliances apparatus, and those apparatus is made useful [connecting with a network]. Therefore, the middleware supporting communication between household-electric-appliances apparatus exist. However, two or more middleware exists and the protocol and architecture differ from each other for every middleware. Therefore, it is difficult to perform mutual communication between the apparatus connected to other middleware networks. In this paper, the new middleware which realizes communication between heterogeneous middleware is proposed by preparing the middle class.

View full abstract

Distributed object computing (DOC) middlewares are widely used forconstructing large-scale distributed systems. Although it provides manyfunctions, there are still issues needing to be resolved. It is not easyto develop systems with DOC middleware because developers must learn anddevelop both application specific logics and DOC middleware specificlogics. This paper proposes a framework called the Model-DrivenDistributed System (MDDS). In this framework, developers can focus oncreating the application model using UML and writing applicationspecific code, without being concerned with DOC middleware specific APIsand logics. This paper describes (1) the functions MDDS provides, (2)the architecture that MDDS-based application uses.

View full abstract

This paper presents a compound document framework, whichcan build a document from visible software components and adapt themto the needs of their users. Since each component is implemented as amobile agent, a document can travel over a network as a whole, withall its embedded components. The framework also provides severalvalue-added mechanisms for visually manipulating components embeddedin a compound document and for sharing a window on the screen amongthe components. It enables a document to itinerate among remotecomputers according to its content. It also can offer components forsupporting various operations for documents, for example, forwardingand duplication. This paper briefly describes this framework and itsprototype implementation.

View full abstract

Virtual Private Networks(VPN) is technique to draw virtual line between two hosts.When we build VPN, user authentication is perfomed. But, afterwe built VPN, no authentication is performed. Therfore, all user who loged on that host can use this VPN freely. If one user built VPN from host that is used plural users at the same time, other users can use that VPN and access to other host. This situation is unsafe. To build VPN which can use only specific user, we propose Personal VPN, which introduces the notion of users into virtual private network. We realize Personal VPN by limitting users who can use VPN.

View full abstract

A large scale network game frame work is proposed forPS2Linux. Our proposed psx-linda method for old PlayStationis not suitable for large scale network games, becauseit requires many acknowledges and resulted communicationconcentration on the linda server. We use Suci librarywhich can control acknowledge in detail for the frame work. We also introduce one to one corespondent agentwith PS2 network client.

View full abstract

In this papaer, we propose a protocol and interfaces for sharing distributed resources such as a data, events and agents at a local network consists of embedded devices such as a sensor. This system is composed of resource searching by IP multicasting, name resolving and key update protocol. And it has the authentication and encryption functions for distributed resource management.

View full abstract

Recently, much attention has been paid to research on real-world navigation systems that can recognize a user's physical situation and present situation-dependent contents to guide him or her. Although some real-world navigation systems have already been implemented, they have not been widely adopted, because they assume the use of a computer network system and are difficult to use for both providers and users. To solve this problem, we propose a small magical device based on an existing device called a PHS. If providers and users use magical devices instead of computer network systems, they can use the real-world navigation systems easily and at low cost.

View full abstract

Trees are useful data types, but developing efficient parallel algorithms manipulating trees is known to be difficult, because of their irregular and imbalance structure. This paper proposes an efficient implementation of parallel tree skeletons on distributed systems. We adopt the m-bridge technique to efficiently partition trees into connected components, and partially evaluate each components in parallel even when some components may contain unknown information. The experimental result indicates that our implementation is practically useful even for imbalanced trees.

View full abstract

Several kinds of Object Calculus were proposed to model and analyze object oriented computation.None of Object Calculus includes the exception handling feature which is very useful and incorporated in many languages.This paper extends Ob_<:1,which is the simplest of Object Calculus,to Ob_<:1by introducing the exception handling feature.We present the operational semantics and type system for Ob_<:1 and then prove the subject reduction theorem for Ob_<:1.

View full abstract

Return barrier is a mechanism for real-time garbage collection, whichdevides the process of root marking into small chunks and thus reduce the suspension time of the main program. In case we implement garbage collection equipped with retun barrier on aparallel system with multiple mutators and a collector, a mutual exclusion is necessary between a mutator and the collector when themutator processes return from a function call.In this paper, we propose an implementation method for parallel garbage collection equipped with return barrier with low overhead of mutual exclusion.

View full abstract

Active Software have the ability to know the state of computation and to change itself dynamically for holding availability. In this paper, we construct programs with active functions which have activation condition and function body to express the expected state and corresponding processing. By separating activation condition and function body, we could get more flexibility and possibility to change programs dynamically. We propose some styles of dynamic updating and linking of active software, and show the effect of them through an example. Then, as a problem in future, we discuss the notation of dynamic updating based on pi-calculus presentation about dynamic linking.

View full abstract

We propose a runtime security policy enforcement mechanism. In this mechanism, a possibly insecure mobile object is modified so that its execution can be supervised by the security metaobject generated from a policy definition. In this paper, we present our policy description language Polaris/J and some examples of security policies written in the language. Polaris/J is based on a finite process algebra enhanced with the use of explicit date dependency (data flow) information.

View full abstract

Following recent growing popularity of XML as data representation more and more researches have been devoted to better functional frameworks for supporting XML processing. Of that, XDuce employs regular expressions as types and this enables programmers to manipulate XML-based data flexibly. This flexibility, however, brings difficulty to function fusion, which is an optimization technique to eliminate intermediate data structures. This paper makes the first attempt to automatically fuse functions with regular patterns.

View full abstract

We propose a type system for the Java bytecode language, provethe type soundness, and develop a type inference algorithm.In contrast to the existing proposals, our type system yields atyped term calculus similar to type systems of lambda calculi. This enables us to transfer existing techniques and results oftype theory to a JVM-style bytecode language.We show that a subroutine can be considered as a polymorphic function,and that there is ML-style type inference algorithm.

View full abstract

[in Japanese]

View full abstract

Copying garbage collection, which was developed by Cheney,copies objects in breadth-first order. Therefore, in most cases,the locality of references is reduced, and the number of cache/TLB misses and page faults increases. Depth-first methods have beenproposed, but the improvement of the locality is limited.In this research, we propose to improve the locality by hierarchicallygrouping objects. We also propose a method which groups as many objectsas possible with a bounded work space.

View full abstract

In this paper, we consider a Snapshot algorithm which uses UDP-based communication library “Suci”. Suci controls a flow and congestion from user level. Various Snapshot algorithm is proposed, but they have a reliability of communication on transport layer. We show a reduction of a network traffic by controlling flow from user level.

View full abstract

It is desirable that users of an intranet use fair shares of the bandwidth of the link to the Internet. This paper presents a simple packet dropping mechanism to be used in the gateway between the intranet and the Internet. This mechanism drops packets for a certain short period of time that destined to a host in the intranet which has used excess incoming bandwidth. Usage of bandwidth for each host is estimated by counting incoming packets.We implemented this mechanism in a PC router using ALTQ and did experiments in a small network to evaluate the performance of the implementation. The experimental result showed this mechanism gives fair shares to the intranet hosts.

View full abstract

We present TS-I/O(Time Stamp - I/O), which is a real-time system for recording/playing analog sound. Other real-time OSs are not suitable for real-time sound operaion and are complex. Therefore we developed TS-I/O to get time stamp at I/O interruption. Applications can know that they could record/play sound data correctly and can get better quality with our system.

View full abstract

We propose a method for extracting pairs of synonyms from search logs.We introduce a relation “link” over words, based on their co-occurrences in each query.Pairs of synonyms are extracted by evaluating their link strength. An experiment is performed to verify effectiveness of this methodwith the log data of “Catwalk”. Catwalk is a full-textretrieval system in use for the archives on a linux-users ML. We analyzed about 70,000 queries logged from February 2002 to June 2002.

View full abstract