Domain-Specific Profiling - TOOLS 2011

Domain-Specific ProfilingAlexandre Bergel, Oscar Nierstrasz, Lukas Renggli and Jorge Ressia

Proﬁling:Is the activity of analyzing a programexecution.

Traditional Profilers

ProﬁleSource Report

Examples

CPU time

Mondrian

o mpl exityS m C ucasse 2003 yste d D n Lanza a

lit tle impact on the overall execution. T his sampling technique is uall mainstream profilers, such as JProfiler, Your K it, xprof [10], an Message Tally, the standard sampling-based profiler in Pharo Smtually describes the execution in terms of C P U consumption and ieach method of Mondrian:54.8% {11501ms} MOCanvas > > drawOn: 54.8% {11501ms} MORoot(MONode) > > displayOn: 30.9% {6485ms} MONode > > displayOn: | 18.1% {3799ms} MOEdge > > displayOn: ... | 8.4% {1763ms} MOEdge > > displayOn: | | 8.0% {1679ms} MOStraightLineShape > > display:on: | | 2.6% {546ms} FormCanvas > > line:to:width:color: ... 23.4% {4911ms} MOEdge > > displayOn: ... We can observe that the virtual machine spent about 54% othe method displayOn: defined in the class MORoot. A root is thenested node that contains all the nodes of the edges of the visuageneral profiling information says that rendering nodes and edge

Domain-Specific P rofiling 3C P U t i me p rofiling Which is the relationship?Mondrian [9] is an open and agile visualization engine. Mondrian describes avisualization using a graph of (possibly nested) nodes and edges. In June 2010a serious performance issue was raised1 . Tracking down the cause of the poorperformance was not trivial. We first used a standard sample-based profiler. E xecution sampling approximates the time spent in an application’s methodsby periodically stopping a program and recording the current set of methodsunder executions. Such a profiling technique is relatively accurate since it haslit tle impact on the overall execution. T his sampling technique is used by almostall mainstream profilers, such as JProfiler, Your K it, xprof [10], and hprof. Message Tally, the standard sampling-based profiler in Pharo Smalltalk 2 , tex-tually describes the execution in terms of C P U consumption and invocation foreach method of Mondrian:54.8% {11501ms} MOCanvas > > drawOn: 54.8% {11501ms} MORoot(MONode) > > displayOn: 30.9% {6485ms} MONode > > displayOn: ? | 18.1% {3799ms} MOEdge > > displayOn: ... | 8.4% {1763ms} MOEdge > > displayOn: | | 8.0% {1679ms} MOStraightLineShape > > display:on: | | 2.6% {546ms} FormCanvas > > line:to:width:color: ... 23.4% {4911ms} MOEdge > > displayOn: ... We can observe that the virtual machine spent about 54% of its time inthe method displayOn: defined in the class MORoot. A root is the unique non-nested node that contains all the nodes of the edges of the visualization. T hisgeneral profiling information says that rendering nodes and edges consumes agreat share of the C P U time, but it does not help in pinpointing which nodesand edges are responsible for the time spent. Not all graphical elements equallyconsume resources. Traditional execution sampling profilers center their result on the frames ofthe execution stack and completely ignore the identity of the ob ject that receivedthe method call and its arguments. As a consequence, it is hard to track downwhich ob jects cause the slowdown. For the example above, the traditional profilersays that we spent 30.9% in MONode > > displayOn: without saying which nodeswere actually refreshed too often.C overagePetit Parser is a parsing framework combining ideas from scannerless parsing,

Coverage

scg.unibe.ch/research/helvetia/petitparser

Java Grammar 210 Methods100% coverage to build grammar

Java Grammar 210 Methods 100% coverage to build grammarIs each production of the grammar covered?

ProﬁleSource Report Domain

DomainSource Traditional Code Profilers

Domain-SpecificDomain ProfilersSource Traditional Code Profilers

What does it mean?

Specify

Capture

Present

MetaSpy

Instrumenter Profiler

Domain Domain Object Domain Object Domain Object

Instrumentation strategies Profilers MetaInstrumenter Profiler handler model install strategies setUp observeClass:do: tearDown observeClass:selector:do uninstall observePackage:do: observePackagesMatching:do: observeParser:in:do: install Announcement MethodInstrumenter ParserInstrumenter setUp Instrumenter theClass parser tearDownannouncer selector grammar uninstallinstall methdo replacementuninstall doesNotUnderstand: install run:with:in: uninstall install setUp MondrianProfiler OmniBrowserProfiler PetitParserProfiler tearDown setUp setUp setUp uninstall visualize visualize visualize User provided classes

Specify the Domain interestsCapture the runtimeinformationPresent the results

Mondrian Profiler

o mpl exityS m Ccasse 2003 yste Du Lanza,

MondrianProﬁler>>setUpself model root allNodes do: [ :node | self observeObject: node selector: #displayOn: do: [ ... counting ... ] ]

Profiler

PetitParserProﬁler>>setUpself model allParsers do: [ :parser | self observeParser: parser in: self grammar do: [ ... counting ... ] ]

Implementation

Instrumentation

Two optionsHooking to the DomainReflection

Reflection

scg.unibe.ch/research/ bifrost

Organize the Meta-level

ExplicitMeta-objects

Class Instrumentation Meta-objectObject

Class Instrumentation Meta-objectInstrumented Object

Partial Reflection Selective Reifications Unanticipated Runtime IntegrationAdaptation Composition

ProﬁleSource

Domain-specific Information

MetaSpy Domain-Domain Specific ProfilersSource Traditional Code Profilers

http://scg.unibe.ch	154
http://paper.li	2
http://www.slideshare.net	1

Domain-Specific Profiling - TOOLS 2011

by Jorge Ressia

Accessibility

Categories

Upload Details

Usage Rights

3 Embeds 157

Statistics