AXP: Adaptive eXecution Platform for Services Grids Prof. Dr. Andreas Polze Peter Tröger Operating Systems and Middleware Group

AXP: Adaptive eXecution Platformfor Services Grids

Prof. Dr. Andreas PolzePeter Tröger

Operating Systems and Middleware Group

Who We Are

Hasso Plattner Institute @ UP

■ Privately funded institute

■ Bachelor and Master in IT Systems Engineering

■ ~50 people for ~400 students

Operating Systems and Middleware group

■ Prof. Dr. Andreas Polze + 6 Ph.D.’s + 2 Post-Docs

■ “Extending the reach of middleware”

■ System predictability in wide-area distributed computing

■ Dependable control systems with small devices

BB-Grid Workshop 11.12.2006

Teaching

Architecture of COTS operating systems

■ Windows 2000, Mac OS X, BSD Unix, Solaris

Architecture of component-based, distributed systems

■ Development of distributed applications with middleware platforms

■ CORBA, J2EE, COM+, .NET, Web-Services

Operating systems for embedded and mobile devices

■ Windows CE, .NET Compact Framework

■ Real-time operating systems (LynxOS, VxWorks, QNX)

Paradigms, design patterns and implementation strategies

■ Predictable behaviour for component-based, distributed and embedded systems

■ Performance, fault tolerance and timeliness


Research

Extension of middleware for predictable systems

■ Paradigms, design patterns, implementation strategies

■ Timeliness, security, fault tolerance

Distributed systems

■ Predictable behaviour in heterogeneous distributed systems

■ Legacy integration and vertical monitoring in SOA frameworks

■ Standardization and service orientation in grid environments

Embedded systems

■ Analytical redundancy and online replacement

■ Dynamic (re-)configuration of component systems

■ Aspect-oriented programming in C# and .NET


Asp

ect

Weavin

g

AdaptiveReconfiguration

Extending The Reach Of Middleware

Coord

inatio

nLa

ng

uages

Dynam

icPla

cem

ent

Grid ServiceProvisioning

Adaptive eXecution Platform

Micro

soft

WR

K

PredictableResources

Distributed Control Lab Adaptive Services Grid

Dynam

icU

pd

ate

s

Em

bedded

Mid

dle

ware

Sta

ndard

s

Reso

urce

Partitio

nin

g

Embedded Devices Distributed Systems

ProgrammingModels


Registry

S1

S3

Service Composition

S3S1

Services Grid Infrastructure

Reply

App

AdaptiveProcess

Management

Register

Service Provider

Request

App

Service Requester

S1 S2 S3

Integration of• Internal Services• External Legacy Systems• External Web / Grid Services

S2S2S2

Instance Monitoring & SLA Negotiation

Scalability,SLA fulfillment

S2

*S1*


ASG Architecture

ASG Architecture

Service Infrastructure

Monitoring

Coordination

Deployment

CORBA

.NET Remoting

COM

SOAP

RMI

REST

Dynamic Service Composition

SOAP

Adaptive Process Management

Enactment

Negotiation Profiling

SLA Management

Facade

SOAP

Service Consumer

SOAP SOAP

Semantic Service Discovery

Ext

erna

l Ser

vice

s

Service IntegrationTool


Stateful Service Concept

Stateless Service Stateful Service Stateless Service Stateful Service

Stateful Resource

Operation Call

WS Theory ASG SI

Runtime Env Runtime Env Runtime Env Runtime Env

WS Practice WSRF Grid

OperationCall

State Query

Resource Usage

State Query

OperationCall

State Query


AXP Architecture


AXP for DCL

Coordination Layer Request Processing

ServiceDeployment

ServiceFactory

JobSubmission

Result Analysis

LifecycleExperiment Management

Experiment Discovery

Logical Service Instance (Experiment Job)

Lego Experim. Service

DCL Frontend Web Services

ServiceInstance

Lego Compile Service HDL Experim.

ServiceServiceInstance

HDL Simulation Service


AXP Service Lifecycle

ServiceDeployment

ServiceInstantiation

ServicePlacement

ServiceUpdate

ServiceDestruction

ServiceUsage

ServiceMonitoring

ServiceUndeployment

Client

Client

Client

AdminAdmin


Dynamic Placement

Coordination layer decides about placement of services on execution resources■ AXP stack ensures data consistency for concurrent activities from clients■ Demands central state data handling■ One client-visible logical instance per instantiation■ Multiple physical instances on differing execution hosts, dynamic routing

Apply dynamic resource allocation strategies■ Theoretical foundations from Capacity Planning research (Q-RAM, LogP)

Service Container A

Service Container B

Service Container C

svc1.1 svc2.1

svc1.2

ServiceRequest Router

svc1.3

svc3.1


Unified Monitoring Data Model

Request packageenters platform

(source: WSQM)

Service reachable, but broken

(source: Laprie)

Time for EJB / Handler processing(source: JSR-77)

Finished requests / uptime(source: WSQM)

Service not reachable(source: WSLA)

Service

Resource


AXP & DaimlerChrysler Telematic

ASG C5Mobile Service Provider

Renesas SH7780

QNX Neutrino

MOST BUS

GSM/ GPRS

Module

CAN BUS

JSME

Commu-nication Module

External Service

Proxy Service

Atomic Service

Atomic Service

OSGi

C-5 Coordination Layer

SOAP

SOAPProtocol

Stack


Grid Aspects of AXP

Dynamic Placement■ On-demand allocation of grid resources as execution host■ Submit stand-alone service executable (servlets) or container

Application of standards■ Re-use of WSRF work in service environment ■ DRMAA API specification work■ J2EE-compliant, portable implementation

Scheduling■ Resource allocation strategies■ Theoretical foundations (capacity planning, Q-RAM, LogP)■ Resource partitioning on grid nodes■ Re-use of grid prediction mechanisms (NWS, meta-schedulers)


ASG Testbed in BB-Grid

• BB-Grid Head Node

• Dual-Xeon; 400GB RAID5; daily backup

• Dual-Xeon, 2GB RAM, 250GB HDD, Debian Linux

• Java, Condor

• 4-Way UltraSparc2,16GB RAM, 140GB HDD

• Solaris 10, Java, Condor

• Dual-Itanium, 1GB RAM, 12GB HDD

• Debian Linux, Java, Condor


Cooperation

EU projects

■ Adaptive Services Grid

■ Leonardo Da Vinci

Deutsche Post IT-Solutions

■ AOP

■ Embedded systems

Microsoft / Microsoft Research

■ Micro.NET

■ Windows Research Kernel

■ Curriculum Research Kit

Bachelor projects

■ DaimlerChrysler Research

■ Siemens AG

■ Software AG


Backup


The Distributed Control Lab

Visual Studio Integration


Asp

ect

Weavin

g

AdaptiveReconfiguration

The Big Picture

Coord

inatio

nLa

ng

uages

Dynam

icPla

cem

ent

Grid ServiceProvisioning

Adaptive eXecution Platform

Micro

soft

WR

K

Realtim

e.N

et

PredictableResources

Distributed Control Lab Adaptive Services Grid

Dynam

icU

pd

ate

s

Micro

.Net

Lego.N

et

Mig

ratio

n

Mon

itorin

gD

ata

Mod

el

SO

C

Reso

urce

Partitio

nin

g

Serv

iceC

om

positio

n

Sem

an

ticW

eb

SLA

Man

ag

em

ent

Documents

AXP: Adaptive eXecution Platform for Services Grids Prof. Dr. Andreas Polze Peter Tröger Operating Systems and Middleware Group