一種面向數(shù)據(jù)倉庫周期性查詢的增量優(yōu)化方法
發(fā)布時間：2018-10-20 09:52
【摘要】：大數(shù)據(jù)蘊含著巨大的價值.分析類查詢是獲取數(shù)據(jù)價值的一種重要手段.為及時把握分析結(jié)果的變化,查詢需要周期性地重復.為此,將不可避免地引入對舊數(shù)據(jù)的重復分析.目前,以重用歷史數(shù)據(jù)的中間結(jié)果、優(yōu)化冗余計算為核心思路的增量分析技術(shù),存在用戶透明性不佳、對歷史結(jié)果存儲位置的選擇不夠智能化等問題,對周期性增量查詢的優(yōu)化效果有限.從兼顧用戶透明性和優(yōu)化收益的角度出發(fā),設(shè)計了一種以語義規(guī)則為指導的增量優(yōu)化方法.該方法擴展了增量描述語法,以查詢操作符的操作語義和輸出語義指導對歷史數(shù)據(jù)存儲、合并位置的選擇,再根據(jù)代價模型和物理查詢?nèi)蝿?wù)的劃分位置對選擇結(jié)果進行調(diào)整,生成優(yōu)化后可以在分布式計算框架(如Map Reduce)周期性調(diào)度執(zhí)行的物理查詢?nèi)蝿?wù).以Apache Hive為基礎(chǔ),實現(xiàn)了上述方法的原型Hive Inc.實驗結(jié)果表明:對于擴展了增量語法描述的TPC-H測試集,Hive Inc相對于優(yōu)化前可以獲得平均2.93倍、最高5.78倍的加速;與經(jīng)典的優(yōu)化技術(shù)Inc MR、Dryad Inc相比,分別可以獲得1.69倍和1.61倍的加速.
[Abstract]:Big data has great value. Analysis class query is an important means to obtain data value. In order to grasp the changes of the analysis results in time, the query needs to be repeated periodically. Therefore, repeated analysis of old data will inevitably be introduced. At present, the incremental analysis technology, which takes the reuse of intermediate results of historical data and the optimization of redundant calculation as the core idea, has some problems such as poor user transparency and lack of intelligence in the selection of storage locations for historical results. The optimization effect of periodic increment query is limited. An incremental optimization method guided by semantic rules is designed from the perspective of considering both user transparency and revenue optimization. The method extends the incremental description syntax and uses the operation semantics and output semantics of the query operator to guide the historical data storage, merge location selection, and then adjust the selection results according to the cost model and the partition position of the physical query task. Physical query tasks that can be scheduled periodically by distributed computing frameworks such as Map Reduce) are generated after optimization. Based on Apache Hive, the prototype Hive Inc. of the above method is implemented. The experimental results show that the, Hive Inc of the TPC-H test set with extended incremental syntax description can achieve an average acceleration of 2.93 times and a maximum of 5.78 times compared with the prior optimization, and 1.69 times and 1.61 times of acceleration compared with the classical optimization technique Inc MR,Dryad Inc, respectively.
【作者單位】： 計算機體系結(jié)構(gòu)國家重點實驗室(中國科學院計算技術(shù)研究所);中國科學院大學;
【基金】：國家高技術(shù)研究發(fā)展計劃(863)(2015AA011505) 國家自然科學基金(61303053,61402445,61402303,61521092)~~
【分類號】：TP311.13
本文編號：2282789
[Abstract]:Big data has great value. Analysis class query is an important means to obtain data value. In order to grasp the changes of the analysis results in time, the query needs to be repeated periodically. Therefore, repeated analysis of old data will inevitably be introduced. At present, the incremental analysis technology, which takes the reuse of intermediate results of historical data and the optimization of redundant calculation as the core idea, has some problems such as poor user transparency and lack of intelligence in the selection of storage locations for historical results. The optimization effect of periodic increment query is limited. An incremental optimization method guided by semantic rules is designed from the perspective of considering both user transparency and revenue optimization. The method extends the incremental description syntax and uses the operation semantics and output semantics of the query operator to guide the historical data storage, merge location selection, and then adjust the selection results according to the cost model and the partition position of the physical query task. Physical query tasks that can be scheduled periodically by distributed computing frameworks such as Map Reduce) are generated after optimization. Based on Apache Hive, the prototype Hive Inc. of the above method is implemented. The experimental results show that the, Hive Inc of the TPC-H test set with extended incremental syntax description can achieve an average acceleration of 2.93 times and a maximum of 5.78 times compared with the prior optimization, and 1.69 times and 1.61 times of acceleration compared with the classical optimization technique Inc MR,Dryad Inc, respectively.
【作者單位】： 計算機體系結(jié)構(gòu)國家重點實驗室(中國科學院計算技術(shù)研究所);中國科學院大學;
【基金】：國家高技術(shù)研究發(fā)展計劃(863)(2015AA011505) 國家自然科學基金(61303053,61402445,61402303,61521092)~~
【分類號】：TP311.13
【相似文獻】
相關(guān)期刊論文 前4條
1 朱華吉;吳華瑞;;空間數(shù)據(jù)庫更新過程中增量信息產(chǎn)生原因分析[J];成都理工大學學報(自然科學版);2007年05期
2 張求喜;;道路線狀要素增量更新[J];公路;2014年04期
3 朱麗云;溫慧敏;;交通路網(wǎng)數(shù)據(jù)自動增量識別與技術(shù)更新[J];交通信息與安全;2009年02期
4 姬存?zhèn)?武芳;鞏現(xiàn)勇;焦洋洋;;居民地要素增量信息表達模型研究[J];武漢大學學報(信息科學版);2013年07期
相關(guān)會議論文 前1條
1 林艷;劉萬增;陳軍;;GIS數(shù)據(jù)庫增量更新的模型研究[A];中國測繪學會九屆四次理事會暨2008年學術(shù)年會論文集[C];2008年
相關(guān)碩士學位論文 前2條
1 李英忠;基于J2EE的企業(yè)增量管理平臺的設(shè)計與實現(xiàn)[D];北京交通大學;2013年
2 孫英杰;基于變化信息文件的增量更新方法研究[D];中南大學;2008年
，本文編號：2282789
本文鏈接：http://sikaile.net/kejilunwen/ruanjiangongchenglunwen/2282789.html
最近更新
教材專著
