The Grid

Background and Architecture

1. Keys to success for IT technologies

Infrastructure Open Standards

Infrastructure

without, no one can use the technology financed

by governments, very firstby industry, after interest increases

Open Standards

fast evolution efficiency will be increased more and more integrated

2. How Infrastructure is built

railroads, telephones and power development was very complicated started in small regional areas connected to a bigger network to be successful:

acceptance by the users support by the governments

financial promotive

3. Scientific demands

computational vs. observational = 1 to 10 remote access to

data instrumentation

data and computation intensive powerful management of resources

Problems of experimental science

only few resources worldwide research must be done on site

The Grid may allow corporative work between scientists team spread all over the world

4. Business Impact

large corporations are global in extent

The Grid may link suppliers, manufactures and

customers unite a company into a single collaborative

team

Infrastructure for the masses

only accepted widely if itbecomes transparent to the userdoesn‘t need much knowledge is highly reliable

The growth of technology

development phase technology itself is importantmainly experts

mass adoptionapplications, reliability and availabilitycontrol returns to experts

sinks into background

5. History

ARPANET started in the early 1970s experimental network developed important protocols

TCP/IPnotion packet switching

Evolution (1)

1997, GT2usability and interoperabilitysolutions for authentication and resource discovery and accessprotocols, APIs and servicesGT2 „standards“

not formular not for public review

Evolution (2)

2002, OGSAextended GT2 concepts and technologiesservice-oriented architecturesWeb servicesprovides framework

6. Concepts (1)

Analogies to Peer-to-Peer file sharing sharing in terms of The Grid

direct access to computers software data sensors all other resources

6. Concepts (2)

sharing under a certain set of rulesmechanisms for

accounting payment (if needed)

in money in access to user‘s local resources

Concepts (3)

achieving various QoS decomposing of integrated infrastructure

into fragmanted systems different resources

shared under certain circumstances pool of resources

members can use under a certain set of rules

7. Architecture

seperated into different layers providing different levels of abstraction

lowest level first step for resources into the Grid

core protocol establishing secure connection between Grid

members shared access to local resources base for many different applications

Fabric Layer (1)

provides local resources shared over the Grid computational power storage access to sensors

translating local protocols to Grid protocols components in this layer will act as proxy objects

Fabric Layer (2)

Component provides access to one kind of resources implements resource specific operations general operations for concurrent access show higher-level protocols the resources‘

structure state capabilities

Connectivity and Resource Layer

narrow neck (hourglass model) based on many maybe different fabric layer

technologies base for many very highspread technologies small set of core abstractions and protocols local resources connected to those how ask for

them

Communication protocols

include transportroutingnaming

defined by the ISO/OSI model TCP/IP protocol stack

Security (Connectivity Layer)

one base functionality of this layer secure exchange of data identity verifying

users resources

implementations should base on existing standards support single sign-on

Resource Access (Resource Layer)

enabling user to interact with remote resources defines protocols for

secure negotiation initiation monitoring control accounting payment

information protocols managing protocols

Collective Layer

protocols and services to provide interactions across collections of resources

in many cases built inside the application for examples

weather forecast programNetsolve/GridSolve 2.0

Application Layer

comprises the user applications applications constructed by calling upon

services of any layer may introduce different layers

8. Implementations

Globus Toolkit Version 2 (GT2) Open Grid Service Architecture

(OGSA)

GT2 (1)

first standardized implementation Grid protocols at higher levels assumes suitable software on fabric elements

CPU scheduling file system management sytem monitoring

some components for discovering information about common resource types

GT2 (2)

connectivity layer defined by GSI protocolssingle sign-on authenticationcommunication protectionrestricted delegation of rights

GT2 (3)

implements GRAM protocol provides secure, reliable creation and

management of remote computation uses

„gate-keeper“ to initiate „job manager“ to manage

„GRAM reporter“ for local computations

GT2 (4)

Monitoring and Discovery Service (MDS-2)discovering and accessing

configuration status information

data modelresource-level protocolsconfigurable local registrycollective registry

OGSA (1)

standardization of core GT protocols use essential Grid functions in different

settings service orientated uniform treatment of all network entries

OGSA (2)

Grid service implements

standard interfaces behaviors conventions

services are defined by the OGSI

Resources

Need some sort of refundmentFinancialother

P2P-Networks: eMule

Refundment by Priority Modifier * Waitingtime => Queue Rank

Credit System in eMule

Ratio1 = 2*Up / Down Ratio2 = SQRT(Up+2) Modifier = Min{Ratio1,Ratio2} 1<=Modifier<=10 If Up < 1 MB => Modifier = 1 If Down = 0 => Modifier = 10

OurGrid

CPU-Sharing Round based Problems:

Free Riders ID Changers

rA(B) = v(B,A)−v(A,B)

rA(B): reputation of B relative to A v(B,A): Value of favours B done to A v(A,B): Value of favours A done to B rho: probability of consumer in turn f: probability of freerider epsilon: probability of a freerider getting a

resource

rA(B) = v(B,A)−v(A,B)

f=0,5

rA(B) = v(B,A)−v(A,B)

rho=0,5

rA(B) = v(B,A)−v(A,B)

rho=0,5

rA(B) = max{0, v(B,A) - v(A,B)}

rho=0,5

rA(B) = max{0, v(B,A) - v(A,B) + log(v(B,A))}

rho=0,5

Thank You