Validated Cost Models for Sensor Network Queries-2009

8/3/2019 Validated Cost Models for Sensor Network Queries-2009

http://slidepdf.com/reader/full/validated-cost-models-for-sensor-network-queries-2009 1/21



Catalog

• Motivation

• Problem

•

QUERY LANGUAGE AND ALGEBRA• PHYSICAL OPERATORS

• COST MODELS

• VALIDATION STRATEGY

• EXPERIMENTAL RESULTS

• CONCLUSIONS



Motivation

• Generating a good execution plan for adeclarative query has long been a centralproblem in data management research



Problem

• Starting with the seminal work in System R,there has always been widespread recognitionof the important role that cost estimation models(CEMs) have in query optimization.

• Modern query optimization uses estimationmodels to inform decisions as to which concretealgorithms and auxiliary data structures to use in

evaluating algebraic operations – particularly, inherently expensive ones, such as join

and group-bys



INTRODUCTION

• Parallel and distributed query execution

•

The extension of query technology to datastreams has once more highlighted therelevance of CEMs



QUERY LANGUAGE AND ALGEBRA

• SNEEql is a declarative query language forWSNs in- spired by expressive stream querylanguages, particularly CQL.

• In SNEEql, an evaluation episode is determinedby the acquisition of data

– i.e. setting an acquisition rate sets the rate at whichevaluation episodes occur



QUERY LANGUAGE AND ALGEBRA - Cont.

• In SNEEql, the only structured type is tuple. Theprimitive collection types in SNEEql are :

– window

– stream



PHYSICAL OPERATORS

• The operations in the SNEEql physical algebracan be expressed as concrete algorithms at alevel in which it becomes possible to structurally

derive CEMs for memory, time and energy forthem.

•The methodology with the SP ACQUIRE andTRANSMIT physical operators



PHYSICAL OPERATORS - Cont.

• The methodology is as follows:

– (1) We define, in pseudocode, the processing logic ofan operator that gets executed at an evaluation

episode – (2) We declare, in the pseudocode, the state kept by

the operator to support the processing logic in (1)

– (3) We fairly directly derive from (2) a CEM for

memory – (4) We derive from the algorithmic structure of the

operator a CEM for duration in the classical way



PHYSICAL OPERATORS - Cont.1

– (5) We derive from the CEM for duration obtainedin (4) a CEM for energy by multiplying each addendin the for- mer by the corresponding unit cost in

energy



COST MODELS

• The style of pseudocode used in PHYSICALOPERATORS can be built upon with a view to derivingmemory, duration and energy CEMs for the

corresponding operator.

• Since the pseudocode declares the state kept insupport of its processing logic, deriving a CEM for

memory is tantamount to writing a summation in whicheach addend is the result of applying a primitive likesizeof to each scalar variable and summing the scalarelements in collection variables



• The derivation of a CEM for duration is similarlystraightforward. The only additional concernsare:

– (a) to focus on steps which use processing moreintensively

– (b) to formulate the expressions that act asmultiplicands on the processing blocks that are

iterated over and that quantify the number of passesin the iteration



VALIDATION STRATEGY

• The accuracy of CEMs is a determinant factor inthe quality of the decisions taken by an optimizer



EXPERIMENTAL RESULTS

• The following table show that the validationresults for the memory, duration and energyCEMs for SP_ACQUIRE vary as the number of

sensors and the tuple size vary



CONCLUSIONS

• It describes a methodological approach to thederivation of memory, time and energy CEMs forQEPs intended for execution over WSNs.

• We have shown how the energy CEM is fairlydirectly derivable from the duration CEM alone.Finally, we have shown that the CEMs derived

by our methodology are valid

Validated Cost Models for Sensor Network Queries-2009

Documents

Transcript of Validated Cost Models for Sensor Network Queries-2009