Authorship：Lead author   Language：Japanese   Publishing type：Research paper (scientific journal)  

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Institute of Electrical and Electronics Engineers Inc.  

DOI： 10.1109/LDAV.2017.8231856

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Verlag  

DOI： 10.1007/978-3-319-57529-2_57

Scopus

researchmap

Authorship：Lead author   Language：Japanese  

J-GLOBAL

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/BigDataCongress.2016.58

Web of Science

researchmap

DOI： 10.1109/TBDATA.2016.2546301

researchmap

Language：Japanese   Publisher：日本データベース学会  

CiNii Research

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/VAST.2015.7347677

Web of Science

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/BigData.2015.7364108

Web of Science

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/BigData.2015.7363822

Web of Science

researchmap

Authorship：Lead author   Language：Japanese   Publisher：一般社団法人情報処理学会  

We propose a new Monte-Carlo Tree Search (MCTS) algorithm based on Bayesian Approach to improve performance of game-tree search. It consists of two main concepts; (1) using multiple game-tree search with a randomized evaluation function as simulations, (2) treating evaluated values as probability distribution and propagating it through the game-tree using the Bayesian Approach concept. Proposed method is focusing on applying to tactical games such as Shogi, in which MCTS is not currently effective. We apply the method for shogi using a top-level computer player application which is constructed with many domain-specific search techniques. Through large amount of self-play evaluations, we conclude our method can achieve good win ratio against an ordinary game-tree search based player when enough computing resource is available. We also precisely examine performance behaviors of the method, and depict designing directions.

CiNii Research

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/CLOUD.2014.51

Web of Science

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/BigData.2014.7004260

Web of Science

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1007/978-3-319-08608-8_12

Web of Science

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1007/978-3-319-13560-1

DOI： 10.1007/978-3-319-13560-1_81

Web of Science

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Verlag  

DOI： 10.1007/978-3-319-06608-0_44

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：The International Association for Computer and Information Science (ACIS)  

researchmap

Language：Japanese  

ユーザからのサービス需要が時間とともに大きく変動するアプリケーションに対し，オンデマンドに使用リソースを変動させてサービス提供を行えるクラウドコンピューティングと呼ばれる計算環境が広く利用されている．近年では，リソース利用コストの削減に加え，消費電力量の削減が重要な要請になりつつあるが，電源管理において複数のインタフェース規格を使い分けなければならないこと，また計算機，通信機器，記憶装置などの複数のコンポーネントの電源を連携した状態で制御する必要があることなどに，課題があると考えられる．これらの問題点をふまえ，我々は，計算機クラスタ上でユーザプログラムが電源管理を容易に行える統合的インタフェースを提供し，必要なリソースのみ電源を供給することで省電力化を図ることが可能なシステムを構築した．大規模な実クラスタ上で取得されたタスクトレースを利用したシミュレーションを行ったところ，サービスレベルを保ちつつ13%程度の電力量削減が可能であるとの結論を得た．また，電力制御を加えた探索アプリケーションを実クラスタ上で動作させたところ，総消費電力量を14%から23%程度削減できることを確認した．Cloud computing environments that provide on-demand resource allocation features are widely used to serve applications with time-varying demands. On such applications, reducing electric energy consumption is highly required in these days as well as reducing resource consumption. Clusters are often built by several components that are controlled by the several independent power APIs that should be separately managed by application developers. However, these components, such as computing nodes, interconnects, and storages, depend on each others to execute required jobs, and should be controlled in a coordinated way. That makes it difficult to construct energy aware applications. We propose a power controlling interface on cluster environments that provides a simple and unified way to reduce energy consumption through shutting down components that are not required by running jobs. Through a simulation using a large scale task-trace of a real cluster, we show that our approach can reduce energy consumption by around 13%. We also examine an elastic tree-search application on our system, and show that we can reduce energy by between 14% and 23%.

CiNii Research

researchmap

Language：Japanese   Publisher：日本データベース学会  

CiNii Research

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.2197/ipsjjip.21.17

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1016/j.future.2011.05.026

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：ACM  

DOI： 10.1145/2480362.2480434

Scopus

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/sac/sac2013.html#conf/sac/GrootGYNK13

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

DOI： 10.1109/SCC.2013.116

Scopus

researchmap

Language：English   Publisher：Information and Media Technologies Editorial Board  

It is difficult to fully utilize the parallelism of large-scale computing environments in alpha-beta search. The naive parallel execution of subtrees would result in much less task pruning than may have been possible in sequential execution. This may even degrade total performance. To overcome this difficulty, we propose a two-level task scheduling policy in which all tasks are classified into two priority levels based on the necessity for their results. Low priority level tasks are only executed after all high priority level tasks currently executable have started. When new high priority level tasks are generated, the execution of low priority level tasks is suspended so that high level tasks can be executed. We suggest tasks be classified into the two levels based on the Young Brothers Wait Concept, which is widely used in parallel alpha-beta search. The experimental results revealed that the scheduling policy suppresses the degradation in performance caused by executing tasks whose results are eventually found to be unnecessary. We found the new policy improved performance when task granularity was sufficiently large.

DOI： 10.11185/imt.8.32

CiNii Research

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/SNPD.2013.24

Web of Science

researchmap

Language：Japanese   Publisher：情報処理学会  

Since massive computing resources have been recently available with the progress of information processing, constructing discriminative models with features based on all possible combinations of primitive features has become worth considering. Constructing all possible combinations is easy, and does not require any deep knowledge of the target problem. However, combinatorial explosion results in a huge number of features, which is difficult to be handled efficiently even with massive computational resources. In this paper, we propose a new method to pick out effective features from high dimensional conjunctive features. Our method succeeds in greatly reducing the number of features without the reduction of accuracy in the game of shogi.

CiNii Research

researchmap

Other Link： http://id.nii.ac.jp/1001/00079585/

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

DOI： 10.1109/eScience.2010.43

Scopus

researchmap

Language：Japanese   Publisher：Information Processing Society of Japan (IPSJ)  

For large-scale distributed processing of time-consuming computation with many computers, failures of some of the nodes should not cause failure of the whole computation. For this purpose, several fault-tolerant computation frameworks, such as master-worker and work-stealing methods, have been proposed and actually used. Search problems are generally defined recur-sively: A subproblem is solved using the results of its own child subproblems. Most practical search problems have subproblems shared as children of two or more subproblems. Game tree search is a typical example. With the master-worker framework, many subproblems are likely to be solved repeatedly leading to inefficiency. A search algorithm based on distributed hash table has been developed to eliminate such duplicated computation. Our proposal adds fault-tolerance to the algorithm through recomputation of subproblem results lost with faults. In this paper, a model of search problems with shared subproblems is formalized, and the proposed framework are compared with other methods such as one based on the master-worker framework in search efficiencies of cases with and without faults. The performance on real computers for some practical search problems is also shown.

CiNii Research

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/CIG.2007.368108

Scopus

researchmap

Language：Japanese   Publisher：一般社団法人情報処理学会  

CiNii Research

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/11893318_37

Scopus

researchmap

Language：Japanese   Publisher：一般社団法人情報処理学会  

Recent advances in computer hardware have been making existing computational complexity theory inappropriate for many cases. The random access memory (RAM) model was made unrealistic by large speed gap between the processing units and main memory systems. Distributed computing environments have obsoleted traditional models for parallel computation due to non-negligible diversity in communication delay. In this paper, we propose a new framework for computational complexity, named access complexity, in which the cost lies in data transfer rather than computation itself. The model tries to capture all levels of system hierarchy, from cache systems to globally distributed environments. It models these diverse access costs in a simple and uniform way. We apply the model to analyze some parallel algorithms, to show that the model can analyze well-known algorithms easily. We also show the appropriateness of the model, through comparing the predicted and the measured performance of bitonic sort algorithm.

CiNii Research

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

Web of Science

researchmap

Language：Japanese   Publisher：一般社団法人情報処理学会  

Solving large-scale combinatorial optimization problems demands for massive processing power.These problems are often amenable to parallel processing.As programming for parallel processing from scratch requires its own know-how, research has been widely conducted on designing easy-to-use libraries in this area.However, these existing libraries have not been used widely, as they do not allow users' fine tuning specific to problem domains.Domain-specific tuning sometimes reduces the amount of computation more drastically than application of parallel processing.PopKern is a general-purpose parallel combinatorial optimization library package that allows domain-specific fine tuning.PopKern provides typical basic parallel combinatorial optimization algorithms.All problem-domain-specific parts are defined by the user.Problem-domain-independent frameworks of these algorithms are built in the kernel of the library.User-defined parts are plugged in to the kernel.With this design, users have large freedom of applying their expertise in writing user-defined parts, and can enjoy parallel processing without paying any attention to parallel processing details.In this paper, we describes the design and implementation of PopKern.It also shows some evaluation results with huge combinatorial optimization problems.

CiNii Research

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

Web of Science

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese   Publisher：第15回データ工学と情報マネジメントに関するフォーラム(DEIM 2023)  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese   Publisher：第26回ゲームプログラミングワークショップ(GPW2021)  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：English  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese  

researchmap

Language：Japanese   Publisher：Japan Society for Software Science and Technology  

DOI： 10.11309/jssst.33.4_3

CiNii Research

researchmap

Language：Japanese  

researchmap

Language：English  

researchmap

Language：Japanese  

researchmap

Language：Japanese   Publisher：Japan Society for Software Science and Technology  

CiNii Research

researchmap

Language：Japanese  

J-GLOBAL

researchmap

Language：Japanese  

近年、SSDは容量の増加や、価格低下に伴いHDDに変わるストレージ機器として注目を集めている。SSDのアクセス特性としては、ランダムI/Oが高速な点や、並列アクセスによりスループットが上昇する点が挙げられる。大規模データのクエリ処理で多用されるハッシュテーブルを用いた演算は、ランダムI/Oの帯域が十分な環境では、テーブルサイズを小さくすることで、キャッシュ使用量を抑え、なおかつ利用効率が高まり全体の処理時間を短縮できる。本稿では、これを利用し複数問い合わせでキャッシュやI/O帯域を分割して使用することでスループットを向上させる手法について論じる。

CiNii Research

researchmap

Language：Japanese   Publisher：Japan Society for Software Science and Technology  

CiNii Research

researchmap

Language：Japanese  

CiNii Research

researchmap

Language：Japanese   Publisher：情報処理学会  

既存のクラウドシステムに対して社会基盤として要求される高い信頼性や多様性に対する課題を解決するため，複数の異種クラウドシステム間で連携してリソースを監視，解析，制御するインタークラウドのシステムアーキテクチャを提案し，テストベッド評価した結果について報告する．

CiNii Research

researchmap

Language：English  

researchmap

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

The SSDs are expected new storage devices whose transfer rate are thousands faster than that of HDDs. By employing SSDs on behalf of HDDs, performance of data intensive applications is likely to be improved drasti-cally since I/O cost of HDDs is dominant in execution time of data intensive applications. Then, different behavior of data intensive application may be observed in case of SSDs, since SSD's I/O cost is relatively small in total execution time. In this report, we analyze the difference in behaviors of hash join operation on a relational database with both HDD and SSD environment, and clarify that it is necessary to consider the cost model by taking into account characteristics of SSDs

CiNii Research

researchmap

Language：Japanese   Publisher：Information Processing Society of Japan (IPSJ)  

The SSDs are expected new storage devices whose transfer rate are thousands faster than that of HDDs. By employing SSDs on behalf of HDDs, performance of data intensive applications is likely to be improved drastically since I/O cost of HDDs is dominant in execution time of data intensive applications. Then, different behavior of data intensive application may be observed in case of SSDs, since SSD's I/O cost is relatively small in total execution time. In this report, we analyze the difference in behaviors of hash join operation on a relational database with both HDD and SSD environment, and clarify that it is necessary to consider the cost model by taking into account characteristics of SSDs

CiNii Research

researchmap

Language：Japanese   Publisher：Information Processing Society of Japan (IPSJ)  

The SSDs are expected new storage devices whose transfer rate are thousands faster than that of HDDs. By employing SSDs on behalf of HDDs, performance of data intensive applications is likely to be improved drastically since I/O cost of HDDs is dominant in execution time of data intensive applications. Then, different behavior of data intensive application may be observed in case of SSDs, since SSD's I/O cost is relatively small in total execution time. In this report, we analyze the difference in behaviors of hash join operation on a relational database with both HDD and SSD environment, and clarify that it is necessary to consider the cost model by taking into account characteristics of SSDs

CiNii Research

researchmap

Language：Japanese  

SSD は HDD の 100 倍から 1000 倍の転送レートを持つ高速なストレージデバイスとして注目されている。データインテンシブアプリケーションの処理'性能は、ＨＤＤ を用いた環境では入出力コストが支配的なため、SSD により、その処理'性能を大きく向上することが期待できる。一方で、処理コストに占める入出力コストが相対的に小さくなるため、HDD を使用していた際には気づかなかったアプリケーションの挙動が見えてくる。本稿では、関係データベースのハッシュジョインの処理性能を詳細に解析することで、HDD と SSD それぞれを用いた場合の挙動の違いについて考察し、SSD の特性を考慮した処理コストモデルの必要性を明らかにする。The SSDs are expected new storage devices whose transfer rate are thousands faster than that of HDDs. By employing SSDs on behalf of HDDs, performance of data intensive applications is likely to be improved drastically since I/O cost of HDDs is dominant in execution time of data intensive applications. Then, different behavior of data intensive application may be observed in case of SSDs, since SSD's I/O cost is relatively small in total execution time. In this report, we analyze the difference in behaviors of hash join operation on a relational database with both HDD and SSD environment, and clarify that it is necessary to consider the cost model by taking into account characteristics of SSDs

CiNii Research

researchmap

Language：Japanese  

CiNii Research

researchmap

Language：Japanese  

MapReduceをベースとするデータ処理基盤は，アドホックな問い合わせを行う解析基盤として重要性を増している．MapReduce環境では，Map処理時にすべてのレコードをスキャンし，処理を行う．例えば，同じようなレコード選択条件を持つクエリが繰り返し処理された場合でも，毎回，アドホックなクエリとして扱われ，レコードの全スキャンが繰り返される．本稿では，アドホックではあるが同じよう選択条件をもつ問合せ処理をMapReduce環境において効率良く処理するために，クエリの実行時にインデックスを生成，更新していくデータベースの技術であるAdaptive indexingの導入を検討する．Adaptive indexing適用モデルを提案し，基本的なデータ処理に関してシミュレーションによる評価を行った．

CiNii Research

researchmap

Language：Japanese   Publisher：一般社団法人情報処理学会  

クラウドコンピューティングではeコマースやオンラインゲームなどデータベースをシステムの基盤とするようなアプリケーションでの利用が盛んである。それらのアプリケーションに必要なときに必要なだけの計算資源を提供するために、アプリケーションを構成するサーバ群の負荷分散が重要となっている。これらのシステムでは管理面、コスト面での利点から仮想化技術が利用されており、その負荷分散にはVMライブマイグレーションを利用することができる。本論文では複数のVMをクエリの発行元とするオンライントランザクション処理の負荷分散を考える際に、単純にVMライブマイグレーションを利用しただけではむしろ性能低下を招く可能性があり、データベースのキャッシュまで考慮したVMライブマイグレーションの必要性を示す。またデータベースサーバ間でVMがデータベースのどこを参照しているかの情報を共有することでオンライントランザクション処理の負荷分散による性能低下を防ぐ仕組みを検討する。さらにオンライントランザクション処理におけるVMの挙動を実機を用いて詳細に解析し、オンライントランザクション処理の負荷分散においてVMライブマイグレーションを利用した場合の課題について明らかにする。

CiNii Research

researchmap

Language：Japanese   Publisher：一般社団法人情報処理学会  

クラウドコンピューティングではeコマースやオンラインゲームなどデータベースをシステムの基盤とするようなアプリケーションでの利用が盛んである。それらのアプリケーションに必要なときに必要なだけの計算資源を提供するために、アプリケーションを構成するサーバ群の負荷分散が重要となっている。これらのシステムでは管理面、コスト面での利点から仮想化技術が利用されており、その負荷分散にはVMライブマイグレーションを利用することができる。本論文では複数のVMをクエリの発行元とするオンライントランザクション処理の負荷分散を考える際に、単純にVMライブマイグレーションを利用しただけではむしろ性能低下を招く可能性があり、データベースのキャッシュまで考慮したVMライブマイグレーションの必要性を示す。またデータベースサーバ間でVMがデータベースのどこを参照しているかの情報を共有することでオンライントランザクション処理の負荷分散による性能低下を防ぐ仕組みを検討する。さらにオンライントランザクション処理におけるVMの挙動を実機を用いて詳細に解析し、オンライントランザクション処理の負荷分散においてVMライブマイグレーションを利用した場合の課題について明らかにする。

CiNii Research

researchmap

Language：Japanese  

researchmap

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

Cloud Computing is widely used for applications based on databases such as e-commerce or online game. Load distribution is important aspect for such applications to provide computational resources on demand. In these system, virtualization is used from the point of view of managing cost and monetary cost, and VM live migration can be used for load distribution. In this paper, we point out that just using VM live migration in the system where multple VMs issuing queries for load distribution can cause performance degradation, and VM live migration considering databases' caches is necessary. Then we consider the load distribution system prevents performance degradation by utilizing database cache information. Furthermore, we analyze VM behavior during OLTP, and clarify the problem of OLTP load distribution using VM live migration.

CiNii Research

researchmap

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

MapReduce-based processing infrastructure is getting more important as a data analysis system,which executes ad-hoc queries, in addition to regular batch queries. In MapReduce environment, each map task scans all records and processes them. This all-scan access is repeated if similar queries is repeatedly issued for example the queries which have the range condition on the same column. In this paper, in order to process efficiently ad-hoc but with-similar-range-condition query in Map Reduce environment, we will consider the introduction of adaptive indexing techniques into Map Reduce environment. Adaptive indexing is a DBMS technique index creation and update during the query execution. Every query, index is updated so that subsequent queries will take shorter execution time by using this improved index. We will discuss the challenge around the adaptation of the adaptive indexing into MapReduce environment and propose some model.

CiNii Research

researchmap

Language：Japanese  

MapReduce をベースとするデータ処理基盤は，定型的かつ定期的な問い合わせとともに，アドホックな問い合わせを行う解析基盤として重要性を増している． MapReduce 環境では， Map 処理時にすべてのレコードをスキャンし，処理を行う．例えば，同じようなレコード選択条件を持つクエリが繰り返し処理された場合でも，毎回，アドホックなクエリとして扱われ，レコードの全スキャンが繰り返される．本稿では，アドホックではあるが同じよう選択条件をもつ問合せ処理を MapReduce 環境において効率良く処理するために，データベースの技術である Adaptive indexing の導入を検討する． Adaptive indexing とはクエリの実行時にインデックスを生成，更新していく手法であり，異なるレンジ検索が繰り返し発行されるごとに，インデックスを更新し，次のクエリにてそのインデクスを利用することで，クエリの処理時間が速くなる． Adaptive indexing 適用に関する MapReduce 環境における課題を検討し，適用モデルを提案する．MapReduce-based processing infrastructure is getting more important as a data analysis system,which executes ad-hoc queries, in addition to regular batch queries. In MapReduce environment, each map task scans all records and processes them. This all-scan access is repeated if similar queries is repeatedly issued for example the queries which have the range condition on the same column. In this paper, in order to process efficiently ad-hoc but with-similar-range-condition query in Map Reduce environment, we will consider the introduction of adaptive indexing techniques into Map Reduce environment. Adaptive indexing is a DBMS technique index creation and update during the query execution. Every query, index is updated so that subsequent queries will take shorter execution time by using this improved index. We will discuss the challenge around the adaptation of the adaptive indexing into MapReduce environment and propose some model.

CiNii Research

researchmap

Language：Japanese  

クラウドコンピューティングでは ｅ コマースやオンラインゲームなどデータベースをシステムの基盤とするようなアプリケーションでの利用が盛んである。それらのアプリケーションに必要なときに必要なだけの計算資源を提供するために、アプリケーションを構成するサーバ群の負荷分散が重要となっている。これらのシステムでは管理面、コスト面での利点から仮想化技術が利用されており、その負荷分散には VM ライブマイグレーションを利用することができる。本論文では複数の VM をクエリの発行元とするオンライントランザクション処理の負荷分散を考える際に、単純に VM ライブマイグレーションを利用しただけではむしろ性能低下を招く可能性があり、データベースのキャッシュまで考慮した VM ライブマイグレーションの必要性を示す。またデータベースサーバ間で VM がデータベースのどこを参照しているかの情報を共有することでオンライントランザクション処理の負荷分散による性能低下を防ぐ仕組みを検討する。さらにオンライントランザクション処理における VM の挙動を実機を用いて詳細に解析し、オンライントランザクション処理の負荷分散において VM ライブマイグレーションを利用した場合の課題について明らかにする。Cloud Computing is widely used for applications based on databases such as e-commerce or online game. Load distribution is important aspect for such applications to provide computational resources on demand. In these system, virtualization is used from the point of view of managing cost and monetary cost, and VM live migration can be used for load distribution. In this paper, we point out that just using VM live migration in the system where multple VMs issuing queries for load distribution can cause performance degradation, and VM live migration considering databases' caches is necessary. Then we consider the load distribution system prevents performance degradation by utilizing database cache information. Furthermore, we analyze VM behavior during OLTP, and clarify the problem of OLTP load distribution using VM live migration.

CiNii Research

researchmap

Language：English  

CiNii Research

researchmap

Other Link： http://id.nii.ac.jp/1001/00087051/

Language：Japanese  

CiNii Research

researchmap

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

CiNii Research

researchmap

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

Virtualization is used as a key technology for Cloud Computing to manage computational resources. VM live migration, which is one of the element technology of virtualization, enables cloud providers to consolidate computational resources or distribute loads. As digital data such as web contents or sensing data have been rapidly increasing, it becomes imperative to process massive data efficiently in clients' applications. However the behavior of VM live migration in which massive I/O operations are required haven't precisely analyzed yet. In this report, we analyze the behavior of VM live migration in which I/O intensive applications runs with large I/O access in detail using real machines, and then classify the importance of VM live migration performance with heavy I/O applications.

CiNii Research

researchmap

Language：Japanese   Publisher：人工知能学会  

CiNii Research

researchmap

Language：Japanese  

CiNii Research

researchmap

Language：Japanese   Publisher：情報処理学会  

CiNii Research

researchmap

Language：Japanese  

DOI： 10.7210/jrsj.28.1082

CiNii Research

researchmap

Language：Japanese  

With abundant computing resources that have become available with recent information processing apparatus, expressing game positions with features based on all the possible com- binations of primitive features has become worth considering. Constructing the set of all possible combinations is easy without deep knowledge of the subject, however, combinatorial explosion results in a huge set of features, which can hardly be handled efficiently even with massive computational resources. In this paper, we propose a new method to skim off the most effective features from the set of high dimensional conjunctive features. Applying this method to the game of shogi, the evaluation function constructed shows better accuracy thanthose obtained using conventional methods.

CiNii Research

researchmap

Language：Japanese  

従来の並列ゲーム木探索手法の多くは、プロセッサが数十という環境を想定しているため、計算量を抑制することに配慮するあまり並列度が低下し、多数のプロセッサを有効利用できない。本稿では、数百以上のプロセッサからなる環境において大きな速度向上を実現するために、必要なタスクの実行を妨げないようにスケジューリングしつつ、必要性が明らかでないタスクも投機実行することを提案する。提案手法を実装して評価したところ、タスクの粒度と優先度の設定が適切であれば、大きな速度向上が得られることがわかった。また、逐次探索プログラムとの対戦でも提案手法の優位性を示すことができた。Most of the conventional parallel game tree search methods try to prevent increased total computation in environments with dezens of processors, resulting in low parallelism, which cannot utilize a large number of processors effectively. In this paper, we propose a method to realize large speed-up in environments with hundreds of processors by executing speculative tasks, which may revealed to be unnecessary afterwords. These speculative tasks are controlled so as not to disturb mandatory tasks. Evaluation through implementing the proposed method shows high speed-up with appropriate granularity and priority settings. It also shows better performance than program with conventional method.

CiNii Research

researchmap

Language：Japanese   Publisher：The Robotics Society of Japan  

DOI： 10.7210/jrsj.28.241

CiNii Research

researchmap

Language：Japanese   Publisher：The Robotics Society of Japan  

DOI： 10.7210/jrsj.28.1082

CiNii Research

researchmap

Language：Japanese  

CiNii Research

researchmap

Language：Japanese   Publisher：Information Processing Society of Japan (IPSJ)  

The acquisition of the synonym is necessary to a IR (information retrival) system and summary production. When understanding synonym by IR system, it is possible to search using synonym. And, when understanding the abbreviation which is one kind of synonym by summary production, it is possible to generate a shorter sentence. I suggest technique to figure out an abbreviation which is kind of the synonym automatically. Since human generates an abbreviation, feeling of human is important for it. There, mora and syllable is used for the feature of CRF to take a human sence in, an abbreviation is presumed from input original language.

CiNii Research

researchmap

Language：Japanese   Publisher：Information Processing Society of Japan (IPSJ)  

While evolutionary algorithms (EA) and reinforcement learning (RL) are widely used to adjust game stage evaluation parameters, there are few examples to combine these two methods. In this study, we propose a new method combining genetic algorithms (GA) and TD(λ) learning on the basis of hybrid GA. We applied the method to the adjustment of Othello stage evaluation parameters and decided optimal parameters of balancing GA with TD (λ) learning by experiments. A player with parameters tuned with the proposed method showed winning rates of between 54.3% to 69.5% against players tuned with published methods.

CiNii Research

researchmap

Language：Japanese  

CiNii Research

researchmap

Language：Japanese  

CiNii Research

researchmap

Language：Japanese   Publisher：Information Processing Society of Japan (IPSJ)  

Accuracy of evaluation functions is one of the critical factors in computer game players. Evaluation functions are usually constructed manually as a weighted linear combination of evaluation features that characterize game positions. Selecting evaluation features and tuning their weights require deep knowledge of the game. In this paper, we propose a new fast and scalable method to automatically generate game position features based on game records to be used in evaluation functions. Our method treats two-class problems which are common in many types of games. Evaluation features are built as conjunctions of the simplest features representing positions. We select these features based on two measures : frequency and conditional mutual information. To evaluate the proposed method, we applied it to 200, 000 Othello positions. The proposed selection method is found to be effective, because the Naive Bayesian classifier using automatically generated features is more accurate than other classifiers using simple features. We also show that this generation method can treat large problems by dividing them into small problems and can be parallelized easily.

CiNii Research

researchmap

Other Link： http://id.nii.ac.jp/1001/00009784/

Language：Japanese   Publisher：日本ソフトウェア科学会  

CiNii Research

researchmap

Language：Japanese   Publisher：Information Processing Society of Japan (IPSJ)  

We describe InTrigger, a large-scale distributed computing environment. Our goal is to install over 1000 cores spread out across 20 to 30 sites, and we have already installed 514 cores spread out across 6 sites. We explain how we use technologies such as PXE boot and IPMI in order to manage such a multi-site distributed computing environment while making flexible configuration changes. We also introduce a few of the tools that we provide to make it easier to use distributed environments. Finally, we show that some experiments can only be performed with an environment like InTrigger, and describe what we have learned by performing one such experiment.

CiNii Research

researchmap

Language：Japanese   Publisher：人工知能学会  

CiNii Research

researchmap

Language：Japanese  

CiNii Research

researchmap

Language：Japanese  

CiNii Research

researchmap

Language：Japanese  

大規模な分散環境で長時間の計算を行う際には耐故障性が必要である．これまでにも耐故障性を有する大規模な分散計算システムが実用化されている．しかしそれらは基本的にサーバ・クライアント方式に基づくもので，効率的に解ける問題は互いに依存関係のない部分問題に分割可能なものに限られる．そうではない問題の例として組合せ最適化問題やゲーム木探索といった探索問題がある．これらの問題を解く際にはタスクをより簡単な小さなタスクに分割することが再帰的に行われるが，その際に異なる親タスクが共通の子タスクを派生することがある．効率的な探索のためには計算ノードどうしが通信して共通部分を発見し，その結果を再利用することが必要になってくる．サーバ・クライアント方式では部分問題間のこのような依存性を扱うことは難しい．本発表では共通部分問題の結果を再利用し，かつ耐故障および計算途中のノードの参加・脱退が可能な枠組みとして分散ハッシュ表を用いた手法を提案する．本手法では故障時に問題を再実行することで耐故障を実現するが，その際にすでに求められている部分問題の解を効率的に再利用して故障からの迅速な復旧を試みている．本手法を実装し，ルービックキューブをアプリケーションとして63台のPCクラスタで実験を行ったところ48倍の台数効果を得ることができた．また，故意に故障を生じさせる実験を行い，本手法の耐故障性について検証を行った．Fault-tolerance is necessary for widely distributed, long-running, parallel computation. The server-client framework has been commonly used to solve many practical search problems demanding a large amount of computing resources. This framework, however, only can treat embarrassingly parallel problems that can be easily divided into mutually independent subproblems. Problems such as combinatorial optimization and game tree search can also be divided into smaller subproblems recursively, but different parent problems may generate common child problems. For efficiency, these shared child problems should be computed only once reusing the already computed results. The server-client framework cannot treat such dependencies. In this paper, we propose a framework of parallel computation for such problems using a distributed hash table. Fault tolerance is achieved by simply redoing the root problem. All the results of already solved subproblems are reused and thus the recovery process is quite efficient. We implemented this framework and evaluated it through applying it to A* search of Rubik's cube problem. We obtained 48 times speedup using 63 PCs of a cluster. We also tried to verify the fault tolerance of this framework through experiments of inserting artificial faults to some nodes.

CiNii Research

researchmap

Language：Japanese   Publisher：Information Processing Society of Japan (IPSJ)  

This paper presents a case study to operate the Grid testbed for the Grid Challenge in SACSIS2006. The Grid Challenge is a programming competition on a Grid testbed, which is organized by multiple computing resources installed in universities and laboratories. In the last competition, the Grid testbed with more than 1200 CPUs was operated. The paper shows hardware/software specifications of the Grid testbed, and reports experience of the operation, which includes accounting, job management, and troubleshooting.

CiNii Research

researchmap

Language：Japanese   Publisher：Information Processing Society of Japan (IPSJ)  

Checking mates in Shogi and Go have been studied for a long time as a field of computer game players. Great success has been achieved in Shogi, but there are still many challenges in Go. We apply a static evaluation function "Bouzy's 5/21 algorithm" which can estimate a state of position to Depth-First Proof-Number(Df-pn)+ search which is an advanced approach of AND/OR tree search have achived great successes in checking mates of Shogi and confirms an effect in Tsumego. This thesis presents the result of using this approach to a onccyc problem as a field of Tsumego.

CiNii Research

researchmap

Language：Japanese  

J-GLOBAL

researchmap

Language：Japanese  

CiNii Research

researchmap

Language：Japanese  

CiNii Research

researchmap

J-GLOBAL

researchmap

Language：Japanese  

CiNii Research

researchmap

Publisher：Japan Society for Software Science and Technology  

単一計算機内のメモリ階層から計算機間のネットワーク遅延の差異までを統一的に記述できる計算量モデル「計算連続体モデル」を提案している．このモデルでは，計算実体はメモリ上のあらゆる場所に存在し，場所を移動させながら計算を行なう．計算コストは，メモリアクセスに要するコストで表され，計算を行なっている実行場所からアクセス対象データが格納されている場所までの距離に基づいて表現される．これまでに，このモデルの検証に用いる仮想機械を開発しているが，低水準な命令セットであるため，大規模なプログラミングは効率上困難である．そこで，より高水準な記述のできる機械語の設計を行なった．設計した言語は，C言語をベースとし，実行場所を指定する構文とマルチスレッド並列処理の機能を有する．本稿では，この計算連続体モデルに基づく仮想機械用の高水準機械語の設計と実装について述べる．

DOI： 10.11309/jssstconference.2003.0.84.0

CiNii Research

researchmap

J-GLOBAL

researchmap

Language：Japanese  

CiNii Research

researchmap

Language：Japanese  

CiNii Research

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap