jGMA: A Wide-Area Messaging and Resource Discovery System

1. jGMA: A Wide-Area Messaging and Resource Discovery System Matthew Grove ICG Student PresentationsJune 2005

2. Outline • Introduction and motivation, • The architecture of jGMA, • Message passing, • Resource discovery, • Current status, • Demo, • Future work.

3. Introduction • jGMA was initially designed to be a “standards-based” messaging layer to bind together the global layer of a project called GridRM, • jGMA requirements: • Provide message passing and resource discovery, • Work locally over a LAN or a wide-area, such as the Internet, • Fast, and have a minimal impact on its hosts, • Scalable, • Able to work through firewalls, • Compliant to the GMA specification.

4. Design Philosophy • Reuse existing software components, if possible, rather than reinvent functionality, • Simple to install, configure and use, • Provide a ‘basic release’ with a choice to extend functionality with a further more sophisticated component.

5. Grid Monitoring Architecture • Registry • Register/Search • Register/Publish • Bind • Producer • Consumer • Events

6. jGMA Implementation • jGMA has the following components: • A Producer and Consumer (shared code-base), • A Mediator which provides WAN access to producers and consumers and hooks for a resource discovery service (which we call the Virtual Registry). • jGMA features: • Written in Java (taking advantage of usual features and libraries), • Small programming interface (API) of 15 calls, • Released ready to run as a 28Kbyte download (you just need Java).

7. jGMA Architecture

8. The two parts of jGMA • Message passing, • Provides communication between Consumers, Producers and Mediators, • Implemented first. • Virtual Registry (resource discovery), • Boot-strapping: Joining the jGMA network with minimal hardwiring, • Communications: Efficient routing of queries between jGMA Virtual Registries, • Caching: Keeping a temporary local copy of some information to reduce the amount of communications between peers.

9. Optimising Message Passing • Benchmarks to measure: 1. Latency (time to send messages), 2. Bandwidth (throughput of messages), 3. Scalability (performance under load). • Comparing with NaradaBrokering • NaradaBrokering has a very similar architecture to the message passing part of jGMA, • With the help of Raz we have measured latency and bandwidth for jGMA and NaradaBrokering. • Results in a paper submitted to the Grid Workshop of Super Computing 2005.

10. Initial Virtual Registry • Investigating whether we can build a jGMA VR service using the mature Internet Relay Chat (IRC) protocol, • The first IRC network was set up in 1988, • IRC servers are connected in a way to attempt to efficiently route messages and provide fault tolerance, • Some networks manage thousands of users. • IRC can provide bootstrapping and communications.

11. IRC VR Schematic

12. Current Status • Message passing layers complete, • The initial implementation of the IRC VR is complete and functional, • Demo applications and benchmarking has thoroughly tested the jGMA software, • Permanent on-line demo: http://dsg.port.ac.uk/projects/jGMA/demos/web/

13. Webcam Browser Demo http://dsg.port.ac.uk/projects/jGMA/demos/web/

14. Future Work • Short term: • Implement a generic caching service for the VR, • Add hooks for debugging, • Longer term: • Provide a updated binary release of jGMA, • Develop an application or library that uses jGMA.

15. Summary • jGMA is functional despite a few missing features, • Current work is focusing on the implementation of the VR, • The design of the jGMA architecture is complete.

16. Links • Project Web page: • http://dsg.port.ac.uk/projects/jGMA/ • The DSG Web page: • http://dsg.port.ac.uk/