Performance Evaluation Workshop ISCC'09

Michele Stecca

Experiments and Performance Evaluation ofEvent Driven Mashups

Sousse July, 5th 2009

Michele [email protected]

CIPI – University of Padova and Genova (Italy)

2Michele Stecca

Outline

The OPUCE project

Mashup definition

Description of the Mashup execution architectures

Performance tests Test setup Results Conclusions

3Michele Stecca

The OPUCE project

Open Platform for User-centric service Creation and Execution (FP6 IP Nov. ‘06 – Apr ‘09)

User Generated Services Main components of the platform

Service Creation Environment (SCE) Service Lifecycle Manager (SLM) Service Execution Environment (SEE) Portal

4Michele Stecca

Event Driven Mashup Definition

Base Services (Basic blocks) Service Compositions (Mashups) Web Service communications between Base Services

4 Different SEE Implementations

5Michele Stecca

BPEL-based solution

Mashup execution based on a BPEL engine BPEL is not suitable for event driven applications Definition of a “standard” BPEL script structure

BSs Container

Orchestrator on BPEL Engine

BS_1 BS_n

BSs Container

BS_1 BS_n

Wait for Event

Swicth on

Event

Execution Branch

#1

Execution Branch

#n

6Michele Stecca

JSLEE-based solution

Sessions, scalability and asynchronous event management provided by the Mobicents JSLEE Container

JSLEE stands for JAIN Service Logic Execution Environment

BSs Container

Orchestrator on JSLEE container

BS_1 BS_n

BSs Container

BS_1 BS_n

7Michele Stecca

Plain JAVA solution

Stateless Orchestration Ad-hoc Orchestrator lighter than BPEL engine:

only event routing Scalability thanks to Orchestrator replication

Orchestrator Management System

BSs Container

Orchestrator (O_Id=1)

BS_1 BS_n

BSs Container

BS_1 BS_n

Orchestrator (O_Id=m)

8Michele Stecca

Fully Distributed Solution

Stateless Orchestration No centralized Orchestrator: intrinsic

scalability

SEE1.1: fully distributed solution

BaseServiceBaseService

BaseServiceBaseService

BaseService

OrchestrationProxy

BSSCBS

SCBSSCBS

SCBSSC

BS SC = Base Service superclass

OPUCE Service Trigger

or

OPUCE Portal

1. notifyEvent [starting a session]

2. first base service is invoked with event

3. only error and final events are routed back to the OrchestrationProxy

Distributed execution network

9Michele Stecca

Test Setup

traffic generator orchestrator bs „SleepA“

Software test setup

Start SleepA SleepB SleepA SleepB SleepA

End

bs „SleepB“

SleepB

SleepASleepA SleepBT1

T2

Hardware test setup

Traffic generated by SoapUI at different rates (50 – 100 – 200 reqs/sec)

Important: BPEL-based solution crashes as soon as the traffic rate goes beyond 10 reqs/sec

10Michele Stecca

Test Results & Conclusions

Config_1 (50 reqs/sec) Config_2 (100 reqs/sec) Config_3 (200 reqs/sec)

L(ms) T(reqs/sec) L(ms) T(reqs/sec) L(ms) T(reqs/sec)

Plain JAVA 5,70 48,58 5,87 91,06 8,08 161

JSLEE 9,53 48,15 173,91 53,21 226,12 51,85

Distributed 3,34 48,44 5,11 91,03 5,20 166 Which one?

BPEL-solution is not suitable for event driven applications The specific implementation of the JSLEE standard should be improved The Plain JAVA solution shows good performance and complete control The fully distributed solution is better when high performance is required

11Michele Stecca

Questions & Answers