ダウンロード数: 264

このアイテムのファイル:
ファイル 記述 サイズフォーマット 
ipsjjip.20.128.pdf1.52 MBAdobe PDF見る/開く
タイトル: Parallel Graph Traversals using Work-Stealing Frameworks for Many-core Platforms
著者: Yasugi, Masahiro
Hiraishi, Tasuku  KAKEN_id  orcid https://orcid.org/0000-0003-4285-893X (unconfirmed)
Umatani, Seiji  KAKEN_id
Yuasa, Taiichi
著者名の別形: 八杉, 昌宏
平石, 拓
馬谷, 誠二
湯淺, 太一
キーワード: work stealing
graph traversals
spanning tree
many-core
発行日: Jan-2012
出版者: Information Processing Society of Japan
誌名: Journal of Information Processing
巻: 20
号: 1
開始ページ: 128
終了ページ: 139
抄録: Parallel programming/execution frameworks for many/multi-core platforms should support as many applications as possible. In general, work-stealing frameworks provide efficient load balancing even for irregular parallel applications. Unfortunately, naïve parallel programs which traverse graph-based data structures (e.g., for constructing spanning trees) cause stack overflow or unacceptable load imbalance. In this study, we develop parallel programs to perform probabilistically balanced divide-and-conquer graph traversals. We propose a programming technique for accumulating overflowed calls for the next iteration of repeated parallel stages. In an emerging backtracking-based work-stealing framework called “Tascell, ” which features on-demand concurrency, we propose a programming technique for long-term exclusive use of workspaces, leading to a similar technique also in the Cilk framework.
著作権等: © 2012 by the Information Processing Society of Japan
URI: http://hdl.handle.net/2433/168288
DOI(出版社版): 10.2197/ipsjjip.20.128
出現コレクション:学術雑誌掲載論文等

アイテムの詳細レコードを表示する

Export to RefWorks


出力フォーマット 


このリポジトリに保管されているアイテムはすべて著作権により保護されています。