📄 http:^^www.cs.wisc.edu^~poosala^tech^qestimate.html
字号:
Date: Mon, 04 Nov 1996 23:50:00 GMTServer: NCSA/1.5Content-type: text/htmlLast-modified: Wed, 07 Feb 1996 19:01:30 GMTContent-length: 3428<HTML><HEAD><title> Research </title></HEAD><H1>Brief description of work</H1><BODY><PRE> Words are flowing out like endless rain into a paper cup, They slither while they pass, they slip away across the universe</PRE><ul><h2> <!WA0><img alt="o" src="http://www.cs.wisc.edu/~poosala/tech/redball.gif"> Query Result Size Estimation</h2>Query optimizers make certain assumptions about the distribution of data inthe database in order to estimate the selectivities of various query operators.For complex (multi-join) queries, errors in these estimates tend to propagateexponentially, possibly resulting in suboptimal plans being chosen. This problemhas become serious even in well-established commerical optimizers. Some form ofhistograms that approximate data distributions has thus become necessary. Wehave identified the optimal histograms for estimating result sizes of querieswith multiple joins and arbitrary selections. These histograms are formed basedon statistics collected from the database, and the resources to do this, mainlycatalog space and collection time, are very limited in a commerical environment.To address this, we have also identified a version of the above histograms thatare near optimal and can be constructed very efficiently. <br><h3> Applications</h3>Query Optimization, Query Design (User Interfaces).</ul><hr><ul><h2> <!WA1><img alt="o" src="http://www.cs.wisc.edu/~poosala/tech/redball.gif"> Query Cost Estimation</h2>An equally critical issue is the estimation of query costs. We are currentlydesigning a unified data structure, which would capture information aboutphysical clustering of data on disk pages in addition to data distributions.The structure is as compact as a histogram and will estimate both sizes andcosts of widely varied operators (e.g., unclustered index scans, or intermediatejoins on RIDs). We are studying the optimality and practicality of thisstructure, hoping that the results will make a variety of estimation algorithmsunnecessary, while providing better estimates of both result sizes and querycosts. <br><h3> Applications</h3> Query Optimization, Query Design (User Interfaces), Physical Database Design, Load Balancing.</ul><hr><ul><h2> <!WA2><img alt="o" src="http://www.cs.wisc.edu/~poosala/tech/redball.gif"> Other Problems </h2> Other problems that we investigated include estimation for parallel join load balancing, result sizes of additional operators, parallel query optimization.<br></ul><hr><ul><h2> <!WA3><img alt="o" src="http://www.cs.wisc.edu/~poosala/tech/redball.gif"> Implementations </h2><ol><li> Designed and implemented efficient algorithms for computing and using histogramsfor size estimation in the DB2-6000 optimizer. <br> <br><li> Designed and implemented the size estimation module for the Precis Query Profiler,used by the Lockheed Martin Missiles and Space Company. <br><br></ol></body><hr><hr><body><h2> <!WA4><A HREF="http://www.cs.wisc.edu/~poosala/tech/pub.html" <i>Publications</i></A></h2><P> If you would like to get a copy of the postscript of any of the publications e-mailed to you, please fill in the form below and click on the button.<h3> Please, include your e-mail address and the document name in the mail. </h5> Thanks.<!WA5><FORM ACTION="http://www.cs.wisc.edu/cgi-bin/mailto/poosala@cs.wisc.edu" METHOD=POST><INPUT TYPE="hidden" NAME="Subj" VALUE="Paper Request"><INPUT TYPE="text" NAME="Msg" SIZE=110,5><P> <INPUT TYPE="submit" VALUE="Mail above note to vishy"><INPUT TYPE="reset" VALUE="Clear text"></FORM></body></html>⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -