Grid computing systems, Globus, AppLes and other things

1
Grid computing systems, Globus, AppLes and other
things

• Paul Card
• November 13, 2002
• 115 Armes 600pm
• pcard_at_ee.umanitoba.ca

2
Grids ?

• Large scale resource sharing
• Heterogeneous resources
• Unstable resource set (dynamic)
• Distributed
• New applications

3
Grid Problem - Ian Foster

• Flexible, secure, coordinated resource sharing
  among dynamic collection of individuals,
  institutions and resources
• virtual organizations

4
Grid Architecture

• Extensible and open
• Grid Technologies
• Protocols
• Services
• APIs
• SDKs

5
Not a Grid ?

• Parallel Computing
• P2P ?
• Artificial Intelligence
• Supercomputing

6
Virtual Organizations

• resource sharing
• Not just networks
• CPU
• Memory
• Disk
• Software
• Data

7
VOs may span many providers

• Storage Providers
• Cycle Providers
• Application Providers
• Network Providers
• Lifetime, Long Short ?
• Secure
• Hetro Homogeneous

8
Grid Cost

• Not only access to greater resource
• Lower Cost
• Higher Utilization
• Extended resource lifetime
• Lower maintenance
• IT Outsourcing

9
Resource Providers

• Heterogeneous
• Set the availability
• Security
• Resource types and parts
• Resource Discovery
• Dissemination

10
Globus

• Defines several protocol layers
• Application
• Collective
• Resource
• Connectivity
• Fabric
• Diagrams stolen from the atonomy of the grid
• Ian Foster

11
(No Transcript)
12
Globus Fabric

• Computational Resource
• Start, Stop, Monitor
• Storage Resources
• Putting, Getting
• Network Resources
• Prioritization, Reservation
• Code Repositories (CVS)
• Catalogs (RDBMS)

13
Globus Fabric

• Link layer
• Toolkit builds on top of existing Internet
  Protocols
• GARA
• General-purpose Architecture for Reservation and
  Allocation
• Implements an slot manager

14
Globus Connectivity

• Single sign on
• Delegation
• Delegate software with users abilities
• Integration with various local security
• Kerberos, UNIX, NT Domains
• User-based trust relationships
• A-gtB -gt C, A-gtC ?

15
Globus Connectivity

• GSI
• Grid Security Infrastructure
• Authentication
• Communication protection
• Authorization
• Transport Layer Security (TLS)

16
Globus Sharing

• Information Protocols
• Structure, State, Configuration, Load
• Management Protocols
• Negotiation
• Toolkit
• GRIP Grid Information Protocol
• GRAM Grid Resource Access Management
• GridFTP
• LDAP

17
Globus Collective

• Directory services
• Resource discovery
• Co-Allocation scheduling brokering
• Resource coordination
• Monitoring and diagnostics
• Failure, attack
• Data replication
• Reliability cost caching
• Grid-enabled programming systems
• Programming model shims

18
Globus Collective

• Workload management
• Problem solving environments
• Software discovery services
• Best implementation, version
• Community authorization
• Community polices
• Community accounting
• Communitiy restrictions
• Collaboratory services
• Groupware

19
(No Transcript)
20
Globus a Grid ?

• Right applications for grids
• Globus provides toolkits
• GRAM, GRIS,
• Just code that makes it easier to build a grid
  like application
• Other means for building Grid Systems or VOs

21
Alternatives

• Java
• J2EE
• Jxtra
• DCOM
• NT Domains
• .NET
• Islands of proprietary solutions

22
Others

• P2P, Seti_at_Home
• The next Internet ?
• Free cycles ?
• Distributed operating system ?
• New programming model ?
• Replace supercomputer ?

23
AppLes

• Application info. Is needed for good scheduling
• Dynamic Information is needed
• Prediction Needed
• A resource is only as good as the application to
  be run

24
AppLes information

• Heterogeneous Application Template
• HAT describes the application characteristics
• Structure, implementation
• User Preferences
• Performance, preferences
• Network Weather Service
• Models

25
AppLes

• Application Level Scheduling
• Resource Selector
• Filters resource combinations
• Planner
• Generates a system independent schedule
• Performance Estimator
• Estimates from candidate set
• Actuator
• Implements best schedule

26
(No Transcript)
27
Algorithm

• 1. Select candidate resource sets S i .
• 2. For each resource set S i do
• Plan a schedule for S i based on strip
  decomposition (the user preference),
• Estimate the cost of the schedule for S i .
• 3. Determine the resource set and schedule with
  the minimum
• predicted execution time (the performance
  measure for this application).
• 4. Actuate the selected schedule.

28
NWS

• Predictive Accuracy
• Accurate estimates in a timely manner
• Non-intrusiveness
• Monitoring should be light
• Execution longevity
• Should be ever present
• Ubiquity
• Should be able to predict for any resource

29
NWS Components

• Persistent State
• Stores and retrieves measurements
• Name Server
• Process / dates to tcp/ip port number address
• Sensor
• Gathers performance measurements from device
• Forecaster
• Does weather predictions

30
Architecture
31
AppLes Globus others

• Are still application or group of application
  specific
• Custom built applications
• Future
• Service Level Agreements
• Standard Applications
• Billing etc
• Everything else