成果報告書詳細
管理番号20110000000807
タイトル*平成22年度中間年報 グリーンネットワーク・システム技術研究開発プロジェクト(グリーンITプロジェクト) エネルギー利用最適化データセンタ基盤技術の研究開発 データサーバの最適構成と進化するアーキテクチャーの開発/ストレージシステム向け省電力技術の開発
公開日2011/8/5
報告書年度2010 - 2010
委託先名富士通株式会社
プロジェクト番号P08017
部署名エネルギー対策推進部
和文要約和文要約等以下本編抜粋:1.研究開発の内容及び成果等 1.1 (2)-3 冗長性除去方式の確立 従来と同等のアクセス性能とストレージ使用量削減30%を実現した冗長性除去技術を確立した。また冗長性除去を複数処理ノード間に分散して行なう技術の試作を完了して、現実のデータセンタ規模の大量データに適用する目処をつけた。 1.1.1 冗長性除去方式の評価 平成21年度に試作したデータ圧縮を含む冗長性除去方式において、実用的なアクセス性能とストレージ使用量削減効果との両立を実現するために、冗長性除去のアルゴリズムの改良とチューニングを実施した。ストレージの使用量削減はファイルサーバに関して、データ圧縮の効果を含めて、43.5%削減を達成した(図1)。
英文要約Title: Development projects on green network system/Development of energy efficient data center technology/Development of an architecture of efficient data servers/Development of energy efficient storage system (FY2008-FY2012) FY2010 Annual Report
1. Evaluation of the primary storage deduplication system
We improved the primary storage deduplication system reported in FY2009. We evaluated its data reduction ability and access performance. The prototype can reduce the amount of data in fileservers down to 57.5%(eliminating 43.5% of data), and its access performance is even faster than raw storages that bypasses deduplication process about random access operations, and slower about sequential access operations. Because random access operations are more popular than sequential access operations in shared storage systems, the prototype satisfies the requirement of this project that claims more than 30% elimination of data and equal access performance to raw storages and must be applicable for the energy efficient data center.
2. Prototyping of the distributed processing function of deduplication
Because the deduplication server requires proportional amount of memory to the amount of storage it provides, it cannot provide datacenter-wide deduplication by itself, even if maximum amount of memory is installed. We developed the distributed processing function in order to scale out the deduplication servers and to provide larger amount of storage by unification of the deduplication servers. The prototype proxy server unifies plural deduplication servers. When a write operation command arrives at the proxy server, it determines the deduplication server according to the value of data, and then it forwards the command to the determined deduplication server. At the same time the proxy server records the correspondence of the block address of the command and ID of the determined deduplication server. When a read operation command arrives at the proxy server, it determines the deduplication server retrieved from the record by block address of the command, and then forwards the command to the determined deduplication server. The proxy server has enough amount of memory to unify up to 1,000 deduplication servers that are large enough to be applicable to the shared storage systems for data centers. We also found the data reduction rate becomes higher in the distributed deduplication system. This is because the larger amount of storage becomes, the larger amount of the duplication data in the storage become.
3. Conferences
We attended following conferences to know the trend in technology of power efficient storage system.
・USENIX Symposium on Operating Systems Design and Implementation (OSDI10)
http://www.usenix.org/events/osdi10/tech/
・USENIX Conference on File and Storage Technologies (FAST11)
http://www.usenix.org/events/fast11/tech/
We also presented our studies in the poster session in FAST11 and received many valuable comments and suggestions.
ダウンロード成果報告書データベース(ユーザ登録必須)から、ダウンロードしてください。

▲トップに戻る