1. Globus and PlanetLab Resource Management Solutions Compared M. Ripeanu, M. Bowman, J. Chase, I. Foster, M. Milenkovic Presented by Dionysis Logothetis

2. Outline Background Starting assumptions Solutions Globus and PlanetLab cooperation

3. Globus & PlanetLab Both address similar problems  Discover, monitor and allocate resources  Securely  In a coordinated way Why do we want to compare?  Some pieces might be transferred  Some pieces might be complementary Why is it tricky to compare?  Both are active projects  Projects are complementary  Globus is a software toolkit PlanetLab is a deployment

4. Grids & Globus Grids  Resource sharing  Dynamic multi-institutional VOs  Coordinated problem solving Globus  Resource discovery, management, monitoring, data movement, security  Uniform API at the library level  Independent of underlying resource management mechanism  Standards-based (OGSA, Web Services, WSRF)

5. PlanetLab Infrastructure testbed Best suited for services that need dispersed points of presence Experimental and production use Designed to run on dedicated hosts Uses virtualization  Low level system abstraction  The user sees a distributed set of virtual containers  Unix-style API Higher value services are built on top Currently: 580 nodes on 275 sites

6. Similar problems, different assumptions User communities Application characteristics Resource characteristics Resource ownership

7. User communities Globus  User diversity  Rich functionality to meet application requirements  Standards-based  GSI framework Tools Protocols APIs PlanetLab  CS Researchers  Minimal functionality Duplicated user effort Competing services  Limited functionality Example: Security

8. Applications Grid applications  Compute intensive  Storage requirements  Network bandwidth  A “necessary evil” PlanetLab services  Network intensive  An objective Resource Distribution

9. Resources Globus  Wide resource heterogeneity  All major OSs supported  Java-based implementations  even larger set of possible environments PlanetLab  Testbed for new services  No legacy hardware or software  Assumes/exploits homogeneity  Supports Intel-based Linux machines

10. Resource ownership Globus  Resource owner controls the site PlanetLab  Limited control for the resource owner  Homogeneity is required  OS and architecture are mandated  PlanetLab admin has root access on nodes  PlanetLab admin has access to a remote power button

11. Different assumptions, different solutions Local resource management Federated resource sharing

12. Local resource management Globus  Unified interface for local resource management  Underlying mechanisms may vary  Main abstractions: Service Job PlanetLab  Low level management functionality  Same for all individual resources  Main abstraction: Virtual Machine

13. Federated resource sharing Global view of the resources Basic concept: delegation  Resource usage delegation Delegate the right to consume resources of a site Delegate to an application or a broker  Identity delegation Delegate one’s identity to another to act on his behalf

14. Resource usage delegation Globus  WS-Agreement Agreement represeantation Re-negotiation protocols Agreement state monitoring  Enforcement mechanism is undefined  User binds an agreement to a job or service PlanetLab  Resource capability 160 bit opaque value Represents access to resources (CPU, disk, network, memory, etc.)  Passed between services  Services can add more info (e.g. authorization, authentication)

15. Identity delegation Globus  GSI  X.509 proxy certificates Time limited Delegation of subset of privileges No third-party needed Dynamic delegation Delegation to dynamic entities PlanetLab  No mechanism  Services can implement their own Service A Service B

16. Global resource allocation and scheduling Globus  Exploits identity delegation PlanetLab  Exploits resource usage delegation Node Managers Brokers Job submission & identity delegation Job submission Forward Capability Provide capability Job submission … … … … … …

17. Globus and PlanetLab cooperation PlanetLab is a platform hosting Globus Example: Data grids  Geographically distributed data resources  Globus provides the services for Data management, access and analysis Security  PlanetLab provides low level services to optimize transfers (e.g. BANANAS) Example: Dynamic Runtime Environments  Globus is used to initiate a virtual container on a PlanetLab node

18. Recommendations Globus  Support for resource usage rights delegation  Community contribution integration PlanetLab  Service interoperability  Support for identity delegation

19. Discussion Is it really useful to run Globus over PlanetLab? How would users benefit from the Globus and PlanetLab cooperation? Can the two systems benefit from each other?

20. End