计算机科学与探索
計算機科學與探索
계산궤과학여탐색
JOURNAL OF FRONTIERS OF COMPUTER SCIENCE & TECHNOLOGY
2015年
2期
182-192
,共11页
王胜%秦小麟%沈尧%李博涵%史文浩
王勝%秦小麟%瀋堯%李博涵%史文浩
왕성%진소린%침요%리박함%사문호
主存索引%持久化%CSB+-树%内存映射%索引头
主存索引%持久化%CSB+-樹%內存映射%索引頭
주존색인%지구화%CSB+-수%내존영사%색인두
main-memory index%durable%CSB+-tree%memory map%index head
现有主存索引方案为实现重用功能仅将更新操作存储到硬盘中,根据操作序列进行索引恢复,实时性和重用性均较差。为进一步提升重用性和实时性,提出了一种可持久化的CSB+-树(cache sensitive B+-tree)索引方案。该方案基于内存映射技术,完整而高效地将索引结构保存到外存中,导入时无需重复创建索引,可节省大量计算资源。针对索引更新过程中出现大量内存碎片问题,采用一种分类内存管理机制进行管理和监视,当内存碎片过多而无法利用时,基于有序键值对进行索引重构以完全消除内存碎片。实验结果表明,所提方案与现有方案相比具有更好的实时性和重用性,同时具有高效的查询处理能力。
現有主存索引方案為實現重用功能僅將更新操作存儲到硬盤中,根據操作序列進行索引恢複,實時性和重用性均較差。為進一步提升重用性和實時性,提齣瞭一種可持久化的CSB+-樹(cache sensitive B+-tree)索引方案。該方案基于內存映射技術,完整而高效地將索引結構保存到外存中,導入時無需重複創建索引,可節省大量計算資源。針對索引更新過程中齣現大量內存碎片問題,採用一種分類內存管理機製進行管理和鑑視,噹內存碎片過多而無法利用時,基于有序鍵值對進行索引重構以完全消除內存碎片。實驗結果錶明,所提方案與現有方案相比具有更好的實時性和重用性,同時具有高效的查詢處理能力。
현유주존색인방안위실현중용공능부장경신조작존저도경반중,근거조작서렬진행색인회복,실시성화중용성균교차。위진일보제승중용성화실시성,제출료일충가지구화적CSB+-수(cache sensitive B+-tree)색인방안。해방안기우내존영사기술,완정이고효지장색인결구보존도외존중,도입시무수중복창건색인,가절성대량계산자원。침대색인경신과정중출현대량내존쇄편문제,채용일충분류내존관리궤제진행관리화감시,당내존쇄편과다이무법이용시,기우유서건치대진행색인중구이완전소제내존쇄편。실험결과표명,소제방안여현유방안상비구유경호적실시성화중용성,동시구유고효적사순처리능력。
To achieve the function of reusing, existing main-memory indexing schemes only store the update opera-tion to the disk and recover the index according to the operating sequence, and are less reusable and real-time. To improve the reusability and efficiency further, this paper proposes a durable CSB+-tree (cache sensitive B+-tree) indexing scheme based on memory mapping technology, which stores the index structure into the disk completely and effi-ciently, and doesn’t need to build the index when loaded. For the problem of memory fragmentation during updating the index, this paper uses a classified memory managing scheme to monitor, and employs a rebuilding mechanism based on the sorted key-value pairs to wipe out the fragmentation when it is too much to reuse. The experimental results show that compared with the existing schemes, the proposed scheme is more real-time and reusable, with effi-cient querying performance.