EMF-IncQuery: Blazing-fast reaction time

even for very large diagrams

Ákos Horváth, Ábel Hegedüs, Zoltán Ujhelyi, István RáthIncQuery Labs Ltd.

Gábor Bergmann, Csaba Debreceni, Dániel VarróBudapest University of Technology and Economics

Outline

Query-based view models• Overview• Evaluation

Motivation and backgroundSirius and queries

Sirius integration• Overview• Evaluation

Conclusion• Conclusion

Main Contributorso Ádám Lengyelo Ábel Hegedüso Zoltán Ujhelyio István Rátho Ákos Horváth

Introduction

3

??I don’t need all that information

Can I define a simplified model?

Can I define a viewpoint

to visualize it?

Maintanence• Incrementally• Immediately

Maintenance:• Incrementally• Immediately

Viewpoint:• Different view of

underlying model• Abstraction hides

complexity

4

Motivating scenario: CONCERTOEU-ECSEL project (started in 2013)• General MDD toolchain for safety-critical systems

o Analysis and code generation for component based systemso UML based modelingo Non-modifiable PSM

Simplified Example

FAM_PilotControl : Function

FAM_Navigation : Function

FAM_FMS : Function

FAM_EMS : Function

nav2ems:InformationLink

provider

consumer

subFunctions

provider

consumer

PilotControl

SubS1

Navigation

FMS

SubS2

EMS

tag: func

tag: func

tag: func

tag: func

EMS: Engine Management System FMS: Flight Management SystemInPort/OutPort

nav2fms:InformationLink

SimulinkConcerto Component models

(UML + profiles)

Id

Id Other SubSystem without tagFunction SubSystem with "func" tag)

Port Blocks

id:Function id:InformationLink

View

More details on Matlab Simulink and Eclipse integration: https://github.com/FTSRG/massif

5

Background: SiriusSirius• Custom concrete syntax for visualization

o Tree, table, graph, etc.• Provides viewpoint definition over EMF models• Abstraction can be defined using interpreted

expressionso MTL – Acceleo Model-to-Text languageo OCL- Object Constraint languageo AQL (recommended as of 3.1) – Acceleo Query

Language• Supports several viewpoints over

the same abstract syntaxMore details on Sirius: https://eclipse.org/sirius/

Background: EMF-IncQueryEMF-IncQuery• Incremental model query engine• Own query language = IQPL

o declarativeo graph pattern based

6

Query Model

Updated results Result deltas

Evaluator

Model change

Efficient change propagation

Always up-to-date results without model re-traversal

Track changes of your model in terms of queries

More details on EMF-IncQuery integration: http://www.eclipse.org/incquery/

IncQuery as a query language in Sirius

8

VSM Render

Overview: IQPL as query language for Sirius

Sirius provides API to provide custom expression interpreter• org.eclipse.sirius.common.expressionInterpreter• org.eclipse.sirius.common.proposalProvider

4. UI updates

EMF Model

B. ChangeNotificationsA. Model

Modification

Live QueriesLive

Queries

2. Get queries

3. Query results

1. UI refresh

DEMOSame model query written in • MTL• AQL• IncQuery

Using IncQuery in Sirius

9

Master:AppTypeSM1: State Machine

S1: State

S2: State

sendSignal()

:HostInstance

:Master

Slave:AppTypeSM1: State Machine

SA: State

SB: State

receiveSignal()

:HostInstance

:Slave

communicates

allocatedTo allocatedTo

instanceOf

instanceOf

10

Evaluation: Interpreted expression

MTL AQL EIQ MTL AQL EIQ MTL AQL EIQSmall Medium Large

0

5000

10000

15000

20000

25000

30000

35000First Execution

Recalculation

Model SizeExec

utio

n Ti

me

[Ms]

Models EObjects EReferences EAttributes Diagram nodes Diagram edgesSmall 3550 34222 9471 12 17Medium 6994 124708 22129 17 13Large 63580 1233581 457230 167 6154

Profiler was used to isolate query execution timeAQL • provides good

performance• Low memory

profileIncQuery• Recalculations

< 50 ms• Requires up to

2x memoryo Large ~1.2 Gb

Query-Based Resources

EVM

VIEW MODELS AND SIRIUS

DerivedModel

EMF Model

A. Model Modification

B. ChangeNotifications

Live QueriesLive

Transformation

Live QueriesLive

TransformationDerivedModel

C. Delta updates

C. Delta updates

D. UI refresh

D. UI refreshB. ChangeNotifications

More details on EVM: http://wiki.eclipse.org/EMFIncQuery/DeveloperDocumentation/EventDrivenVM

VSM + Render

More matches can appear at the same time• Ordered execution schema (priority for rules)Internal traceability for created objects• Explicit definitionConfiguration model hides underlying EVM rule definitions• Predefined set of manipulation rules availableIn summary: One way incremental synchronization arbitrary transformation

Execution of motivating example

11

FAM_PilotControl : Function

FAM_Navigation : Function

FAM_FMS : Function

subFunctions

consumer

PilotControl

SubS1

Navigation

FMS

tag: func

tag: func

tag: func

Simulink CCM

subFunctions

Query results Traceability

functionf_1

f_2

f_3

Trace

Trace

Trace

a 2 3appear create add

Query results

functionIdentifierf_1 i_1

f_2 i_2

f_3 i_3

subFunctionf_1 i_1

f_2 i_2

b appear 4 set

Updating derived modelsInitial setup of derivation rules• EClassifiers, EStructuralFeatures

Query result deltas• Delta = (Found, Lost, Updated)

Based on EMF-IncQuery Event-Driven Virtual machineIntegration architecture

14

Source model

Live transformatio

nrules

Query engine

IncQuery- EVM

Derived model

Change notifications

Match set delta

ApplicationModel

manipulationConfiguratio

n Model manipulation

1

23

4

SiriusUI

update5

DEMO

15

Standard Sirius domain• Family representation• Incremental synchronization

o On-the-fly(Concerto EMF-UML2)• Viewpoint for simplified representation

Using IncQuery in Sirius

16

First timeexecution

9+16 81+160 201+400 401+800 601+1200 801+1600 1201+24000

5000

10000

15000

20000

25000

30000

35000

Transformation

Size of target (view) model (elements + references)Runt

ime

[Ms]

Models EObjects EReferences Diagram nodes Diagram edges9+16 38 89 9 1681+160 371 890 81 160201+400 926 2225 201 400401+800 1851 4450 401 800601+1200 2776 6675 601 1200801+1600 3701 8900 801 16001201+2400 5551 13350 1201 2400 10+

derivation rules

Transformation ~33% of overall runtime

Memory consumption ~3.5x of original model

Evaluation - Concerto

17

Evaluation - ConcertoIncrementalrecalculation

0

200

400

600

800

1000

1200

Deletion Deletion + Sirius

Size of target (view) model (elements + references)

Runt

ime

[ms]

Models EObjects EReferences Diagram nodes Diagram edges9+16 38 89 9 1681+160 371 890 81 160201+400 926 2225 201 400401+800 1851 4450 401 800601+1200 2776 6675 601 1200801+1600 3701 8900 801 16001201+2400 5551 13350 1201 2400 Transformation

re-execution < 50 ms

Refreshing < 1 s

Refreshing is also faster

Conclusions

19

ConclusionsProof-of-concept version is available• IncQuery does well as expected in incremental recalculations• Incremental refreshing is not (yet) available

o Started working on the issues with Obeo • Usage requires deep knowledge of both Sirius and IncQuery• Common base with new EMF-IncQuery Viewers

Experience• Concerto EU-ECSEL project

o Works with Papyrus and EMF-UML2!• Aimed application scenario

o Viewpoint definition directly for EMF models (one-to-one mapping)

o Online synchronization

20

Final pointsThe examples and more details are available form• https://github.com/FTSRG/iq-sirius-integration• Contributors:

o Main: BME-FTSRG, IncQuery Labs Ltd.o Supporting projects: Concerto (EU-Artemis)o (Hopefully) future partner: Obeo

• Presentation will be available on slideshare

Your contributions (feedback, forum posts, ideas, patches) are very welcome!• To what direction should we enhance this approach?