1 Architectural Models for Resource Management in the Grid Rajkumar Buyya Monash University, Australia Steve Chapin, Syracuse University, USA David DiNucci.

1

Architectural Models for Resource Management in the Grid

Rajkumar BuyyaMonash University, Australia

Steve Chapin, Syracuse University, USA

David DiNucciElepar.com, USA

2

Outline

Changes in Computing Landscape Resource Management Issues Architectural Models Hierarchical Resource Management Abstract Owner Market Model Economy Grid Nimrod/G Resource Broker Conclusions

3

Computing Power (HPC) Drivers

Solving grand challenge applications using computer modeling, simulation and analysis

Life SciencesLife Sciences

CAD/CAMCAD/CAM

AerospaceAerospace

Military ApplicationsDigital BiologyDigital Biology Military ApplicationsMilitary Applications

Internet & Ecommerce

4

2100

2100 2100 2100 2100

2100 2100 2100 2100

Desktop SMPs or SuperComputers

LocalCluster

GlobalCluster/Grid

PERFORMANCE

Inter PlanetCluster/Grid ??

•Individual•Group•Department•Campus•State•National•Globe•Inter Planet•Universe

Breaking Administrative Barriers

EnterpriseCluster/Grid

?

Computing Platforms

5

Towards Grid Computing

Unification of geographically distributed resources

6

What is Grid ?

An infrastructure that couples: Computers – PCs, workstations, clusters,

supercomputers, laptops, notebooks, mobile devices, PDA, etc;

Software – e.g., ASPs renting expensive special purpose applications on demand;

Catalogued data and databases – e.g. transparent access to human genome database;

Special devices – e.g., radio telescope – SETI@Home searching for life in galaxy, Austrophysics@Swinburne for pulsars)

People/collaborators.

& offers dependable, consistent, & pervasive access to resources.

7

A Example Grid Infrastructure

8

Sources of Complexity in Grid Resource Management

No single administrative control. No single ownership policy:

Each resource owner has their own policies or scheduling mechanisms;

Users must honour them (particularly external Grid users).

Heterogeneity of resources. Dynamic availability – may appear

and disappear…

9

Sources of Complexity in Grid Resource Management

Unreliable resource – disappear from view!

No uniform cost model - varies from one user’s resource to another and from time of day.

No single access mechanism – Web, custom interfaces, command line…

10

Grid Resource Management Issues

Ack: Globus..

•Authentication (once).

•Specify (code, resources,

etc.).

•Discover resources.

•Negotiate authorization,

acceptable use, Cost, etc.

•Acquire resources.

•Schedule Jobs.

•Initiate computation.

•Steer computation.

•Access remote data-sets.

•Collaborate with results.

•Account for usage.

•Discover resources.

•Negotiate authorisation,

acceptable use, Cost, etc.

•Acquire resources.

•Schedule jobs.

•Initiate computation.

•Steer computation.

Domain 2

Domain 1

11

Architectural Models

MODEL REMARKS SystemsHierarchical It captures model

followed in most contemporary systems.

Globus, Legion, CCS, Apples, NetSolve, Ninf.

Abstract Owner (AO)

Order and delivery model and focuses on long term goals.

Expected to emerge and most peer-2-peer computing systems likely to be based on this.

Market Model It follows economic model for resource discover, sharing, & scheduling.

GRACE, Nimrod/G, JavaMarket, Mariposa.

12

Hierarchical Resource Management

Connection Cloud

UserGlobal

Scheduler

Control Domain

Monitor

Resource

Grid Information Service

Access/Admission Control Agent

Domain Resource Manager or Control Agent

GlobalScheduler

LocalScheduler

- Task

Persistent Job ControlAgent

GlobalSchedule

r

GlobalSchedule

r

GlobalSchedule

r

Deployment Agent

13

Who owns the GRID?

Talk topeople

Powerappliances

Use GRIDresources

Phone co. Electric co.Abstract

Owner (AO)

I want to:

My interface is:

I arrange serviceand payments with a:(may be many choices)

But resources I “get”may belong to others:

GeneratorsPower linesTransformers

AntennaeCable/fiberSwitches

HPCNetworksInstrumentsPeople

14

Order Pickup

AO is owner or broker User negotiates

with AO through “order window”

That AO may own some resources, and/or it may broker with other AOs for those resources

After negotiation, resources are delivered through “pickup window”

OrderWindow

PickupWindow

PhysicalResource

UserRequests Resources

AO

Order Pickup

ResourceManager

AO1

Manager

DeliverySales

AO2

AO3

15

AO Resources

Resources are objects Classes are

Instrument» Data source, sink,

transform» e.g. programs, people,

files, data collection devices

Channel» Moves data among

instruments Complexes of above

Attributes define sizes, times, connections, etc.

Instrument(File)

Instrument(Program)

Instrument(File)

Instrument(Program)

Channels

Instrument(Telescope)

Instrument(Person)

16

Negotiating with an AO

Make dummy resource(with attributes set to

constants, variables, or “don’t care”)

+ bid + delivery plan+ variable constraints

Resource candidates(values for variables/attributes

+ asking price for each)

Pick one,Try again,Or give up

DeliveryWindow

Resource

Order Window

Assign tasksto resource,use, relinquish

Perhapslater...

17

Many Testbeds ? & who pays ?

GUSTO

Legion TestbedNASA IPG

EcoGrid

18

Testbeds so far -- observations

Who contributed to resources & why ? Volunteers: for fun, challenge, fame, public good

like SETI@Home & distributed.net projects. Collaborators: sharing resources while

developing new technologies of common interest – Globus, Legion, Ecogrid.

How long ? Short duration: GUSTO decommissioned.

What do we need ? Grid Marketplace! Regulates demand and supply, offers incentive

for being players, simple, scalable solution, quasi-deterministic – proven model in real-world.

19

Users in Grid Economy & Strategy Grid Consumers

Execute jobs for solving varying problem size and complexity

Benefit by selecting and aggregating resources wisely Tradeoff timeframe and cost

» Strategy: minimise expenses

Grid Providers Contribute “idle” resource for executing consumer

jobs Benefit by maximizing resource utilisation Tradeoff local requirements & market opportunity

» Strategy: maximise returns on services

20

Building of a Economy Grid “brokerage” system…..

Foundation for the Grid Economy

21

Grid Node N

Grid Architecture for Computational Economy

Grid User

Application

Grid Resource Broker

Grid Service Providers

Grid Explorer

Schedule Advisor

Trade Manager

Job ControlAgent

Deployment Agent

Trade Server

Resource Allocation

ResourceReservation

R1

Misc. services

Information Server(s)

R2 Rm…

Pricing Algorithms

Accounting

Grid Node1

…

Grid Middleware Services

…

…

HealthMonitor

Grid Market Services

JobExec

Info ?

Secure

Trading

QoS

Storage

Sign-on

22

Economic Models for Trading Commodity Market Model Posted Prices Models Bargaining Model Tendering (Contract Net) Model Auction Model

English, first-price sealed-bid, second-price sealded-bid (Vickrey), and Dutch.

Proportional Resource Sharing Model

Shareholder Model Partnership Model

Economy Grid = Globus + GRACE

Applications

MDS

GRAMGlobus Security Interface

Heartbeat MonitorNexus

Local Services

LSF

Condor GRD QBank

PBS

TCP

SolarisIrixLinux

UDP

High-level Services and Tools

DUROC globusrunMPI-G Nimrod/GMPI-IO CC++

GlobusView Grid Status

GASS

GRACE-TS

GARA

GridFabric

GridApps.

GridMiddleware

GridTools

GBankGMD

eCash

JVM

DUROC

Core Services

Science

Engineering Commerce Portals ActiveSheet……

24

A resource broker for managing and steering task farming (parametric sweep) applications on computational Grids based on deadline and computational economy.

Key Features A single window to manage & control experiment Resource Discovery Trade for Resources Scheduling Steering & data management

It allows to study the behaviour of some of the output variables against a range of different input scenarios.

What is Nimrod/G ?

25

Nimrod/G Grid Broker Architecture

Grid Middleware

Nimrod/G Client Nimrod/G ClientNimrod/G Client

Grid Information Server(s)

Schedule Advisor

Trading Manager

Nimrod/G Engine

GridStore

Grid Explorer

GE GISTM TS

RM & TS

Grid Dispatcher

RM: Local Resource Manager, TS: Trade Server

Globus,Legion, Condor-g,, Ninf,etc.

G

G

CL

Globus enabled node.Legion enabled node.

CL

Condor enabled node.

RM & TSRM & TS

26

A Nimrod/G Client

A Nimrod/G Client

CostCostDeadlineDeadline

Legion hosts

Globus Hosts

Bezek is in both Globus and Legion Domains

Arlington

Alexandria

Richmond

HamptonNorfolk

Virginia BeachChesapeakePortsmouth

Newport News

Roanoke

Ap p om a toxRive r

Ja m esRive r

Shena nd oa hRive r

Ra p p a ha nnoc kRive r

Potom a cRive r

VIRGINIA77

81

64

64

66

85

27

User Requirements: Deadline/Budget User Requirements: Deadline/Budget

28

Globus+Legion+Condor/G

Australia

Monash Uni.:

Linux cluster

Solaris WS

Nimrod/G

Globus +GRACE_TS

Europe

ZIB/FUB: T3E/Mosix Cardiff: Sun E6500Paderborn: HPCLineLecce: Compaq SCCNR: ClusterCERN: Cluster

Globus +GRACE_TS

Asia/Japan

Tokyo I-Tech.:ETL, Tuskuba

Linux cluster

Globus/LegionGRACE_TS

North America

ANL: SGI/Sun/SP2USC-ISI: SGIUVa: Linux ClusterManitoba: Cluster

Internet

Global Economy Grid

29

Graph 5 - GUSTO Usage for 10 Hour Deadline

0

5

10

15

20

25

30

35

0 2.5 5 7.5 10 12.5 15 17.5 20

Time

No

Pro

cess

es

5 CUs

10 CUs

15 CUs

20 CUs

50 CUs

10 Cost Units

50 Cost Units

20 Cost Units

5 Cost Units

15 Cost Units

Graph 4 - GUSTO Usage for 15 Hour Deadline

0

2

4

6

8

10

12

14

16

18

20

0 2.5 5 7.5 10 12.5 15 17.5 20

Time

Av

era

ge

No

Pro

ce

ss

ors

5 CUs

10 CUs

15 CUs

20 CUs

50 CUs

5 Cos t Units

50 Cos t Units

15 Cos t Units

10 Cos t Units

Graph 3 - GUSTO Usage for 20 Hour Deadline

0

2

4

6

8

10

12

14

16

18

20

0 2.5 5 7.5 10 12.5 15 17.5 20

Time

Ave

rag

e N

o P

roce

sso

rs 5 CUs

10 CUs

15 CUs

20 CUs

50 CUs

5 Cost Units

10 Cost Units

Graph 2 - GUSTO Usage for Ionization Chamber Study

0

10

20

30

40

50

60

70

80

0 2.5 5 7.5 10 12.5 15 17.5 20

Time

Ave

rag

eN

o. P

roce

sso

rs20 Hour deadline15 hour deadline10 hour deadline

30

Conclusions

Proposed three models for Grid resource management architecture Hierarchical, AO, & Market-model

The future systems are likely follow a model that combines all these models.

The future computing (HPC) infrastructure is going to be a Grid of Clusters.

Peer-to-Peer/Grid has already become a darling of venture capitalists.

The impact of Grid on 21st century economy will be the same as electricity on 20th century economy.