主催:NII 先端ソフトウェア工学国際研究センター(GRACEセンター)
http://grace-center.jp/
日時:2010年6月16日(水)10:00-12:00
場所:国立情報学研究所(NII) 20階講義室1(2005)(地図)
参加費:無料
お問い合わせ:加藤 弘之 (kato_AT_nii.ac.jp)_AT_を@に書き換えてください。
参加をご希望の方は、セミナー前日までに下記よりご登録をお願いします。
http://grace-center.jp/regist/seminar
==========================================================
FIRST SPEAKER
Kento Emoto (Tokyo University)
=== Title ===
Fusion Optimizations in Parallel Skeleton Libraries –Towards Reliable Parallel Programming
=== Abstract ===
Parallel programming is difficult. However, it is now an essential task for programmers. This situation calls for a methodology for reliable concise parallel programming. To challenge the problem, we have intensively studied “skeletal parallel programming”, in which programmers easily develop parallel programs by composing ready-made parallel program components called skeletons. Moreover, their naively-composed skeleton programs can be optimized via fusion transformations to achieve efficient execution. In this presentation, focusing on the optimization on parallel skeletons, we will introduce a basic theory of our parallel skeletons and their optimizations briefly, and see their implementation in our SkeTo library (C++) and GoG library (Fortress).
SECOND SPEAKER
Koichi Fujikawa (Cirius Technologies)
=== Title ===
Design of a DSL by Ruby for heavy computations over map-reduce clusters
=== Abstract ===
Recently parallel distributed computing for massive data is easier than ever before because of map-reduce technologies, but we still have difficulties to implement the map-reduce code description efficiently as a general level engineer.
“Hadoop papyrus”, which I created as open source software, enables us to describe map-reduce procedure running on Hadoop not Java language but Ruby which is efficient and flexible programming language.
Furthermore it is designed as DSL framework which can describe for a domain specifically.
I will present how map-reduce procedure can be changed into DSL mainly.