Electrical Engineering
      and Computer Sciences

Electrical Engineering and Computer Sciences

COLLEGE OF ENGINEERING

UC Berkeley

Adaptive Execution of Variable-Accuracy Functions

Matthew Michael Denny and Michael Franklin

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2006-28
March 21, 2006

http://www.eecs.berkeley.edu/Pubs/TechRpts/2006/EECS-2006-28.pdf

Many analysis applications require the ability to repeatedly execute sophisticated modeling functions, which can each take minutes or even hours to produce a single answer. Because of this expense, such applications have largely been unable to directly use such models in queries, with either on-demand or continuous query processing technology. Query processors are hindered in their ability to optimize expenseive modeling functions due to the ``black box'' nature of exisiting user-defined function (UDF) interfaces. In this paper, we address the problem of querying over sophisticated models with the development of VAOs (Variable-Accuracy Operators). VAOs use a new function interface that exposes the trade-off between compute time and accuracy that exists in many modeling functions. Using this interface, VAOs adaptively run each function call in a query only to an accuracy needed to answer the query, thus eliminating unneeded work. In this paper, we present the design of VAOs for a set of common query operations. We show the effectiveness of VAOs using a prototype implementation running financial queries over real bond market data.


BibTeX citation:

@techreport{Denny:EECS-2006-28,
    Author = {Denny, Matthew Michael and Franklin, Michael},
    Title = {Adaptive Execution of Variable-Accuracy Functions},
    Institution = {EECS Department, University of California, Berkeley},
    Year = {2006},
    Month = {Mar},
    URL = {http://www.eecs.berkeley.edu/Pubs/TechRpts/2006/EECS-2006-28.html},
    Number = {UCB/EECS-2006-28},
    Abstract = {Many analysis applications require the ability to repeatedly execute sophisticated modeling functions, which can each take minutes or even hours to produce a single answer.  Because of this expense, such applications have largely been unable to directly use such models in queries, with either on-demand or continuous query processing technology.  Query processors are hindered in their ability to optimize expenseive modeling functions due to the ``black box'' nature of exisiting user-defined function (UDF) interfaces. In this paper, we address the problem of querying over sophisticated models with the development of VAOs (Variable-Accuracy Operators).  VAOs use a new function interface that exposes the trade-off between compute time and accuracy that exists in many modeling functions.  Using this interface, VAOs adaptively run each function call in a query only to an accuracy needed to answer the query, thus eliminating unneeded work.  In this paper, we present the design of VAOs for a set of common query operations.  We show the effectiveness of VAOs using a prototype implementation running financial queries over real bond market data.}
}

EndNote citation:

%0 Report
%A Denny, Matthew Michael
%A Franklin, Michael
%T Adaptive Execution of Variable-Accuracy Functions
%I EECS Department, University of California, Berkeley
%D 2006
%8 March 21
%@ UCB/EECS-2006-28
%U http://www.eecs.berkeley.edu/Pubs/TechRpts/2006/EECS-2006-28.html
%F Denny:EECS-2006-28