Grid Publisher Development
Introduction
The goal of this wiki page is to plan but also track the work on Grid Publisher foreseen by the Monitoring Working Group.
Plan
Phase I
Goal: Set up a simple
ActiveMQ based environment and test it using
GridFTP log data:
- take historical logs from CERN production services to replay or generate simulated data transfer events
- adapt log parser (Python) to publish to ActiveMQ broker
- set up a single instance of broker under Tomcat application server
- develop a simple message summarizer (Java) that consumes GridFTP log messsages in ActiveMQ, does simple summarisation (for example number of events per minute) and publishes the results back to ActiveMQ
- put an additional consumer that simply reads the messages
- test the capabilities of the system in following dimentions:
- number of messages per second
- number of parallel consumers/producers
- number of topics
- complexity of JMS selector
Details:
- machines for test installation provided by James (lxb6117, lxb6118)
- an existing Python implementation of STOMP protocol pointed out by James, to be evaluated/extended by Wojtek
Phase II
Goal: Evaluate possible broker network configurations, set up and test the most promising one(s), test durable subscriptions
- enable persistent messaging on broker running at lxb6117 (MySQL database)
- provide high availability and fault tolerance by Master-Slave broker configurations (Shared Filesystem based, JDBC based)
- test networks of brokers and brokers' discovery methods
- test client-side dynamic failover
Phase III
Goal: Develop security mechanisms by using digital signatures and encryption on the message level with Grid certificates (proxies).
Details to come
Progress tracking
2007-07-19 - planning meeting
- Participants: James Casey, Ian Neilson, Wojciech Czech, Piotr Nyczyk
- Goals achieved: Initial planning
2007-07-30 - phase I tests report
2007-08-09 - phase II tests report
2007-08-09 - evaluation of tests results
- Participants: James Casey, Ian Neilson, Wojciech Czech, Piotr Nyczyk
- Goals achieved: Planning to test prototype in SAME configuration
2007-09-19 - Grid Messaging System Presentation
2007-10-25 - ActiveMQ extended Tests Presentation and Report
- Participants: Max Boehn, James Casey, David Collados, Gavin Mccance, PiotrNyczyk, Daniel Rodrigues
- Summary: Presentation of the 1st month of work on the MSG ( Messaging Service For Grids )
- Final Remarks:
- Double Check inconsistent results on the ActiveMQ broker capabilites
- Further test with more fine-grained measurements, namely consumers throughput and latencies.
- Stick to ActiveMQ 5.0, unless unusable
- Time visualization on impact of aditional producers or subscribers. ( graphs of messages/second per each producer/ consumer )
2008-01-29 - Openlab Presentation ( MSG: An Overview on a Messaging System for Grid )
- Summary: Presentation on test results and further directions for the MSG in Openlab Quarterly Meeting
- Further Steps:
- Test with different topics, queues, + master slave configurations;
- Implement SAM messaging (based on Piotr Prototype
- MSGOpenlab_v1.2.pps: MSG Openlab Presentation
2007-01-30 - Test Plan
Further test plans:
clients on different machines, improvedClient |
1 Consumer, 1 Producer |
3Consumers, 3 Producers |
5 Consumers, 5 Producers |
Queue Performance |
1 Consumer 1 Producer |
3 Consumers 3 Producers, 1 queue |
3 Consumers, 3 Producers, different queues |
Master/Slave |
1 Consumer 1 Producer Queue |
3 Consumers 3 Producers, 1 topic |
100 Producers, 1 Consumer, Queue |
Master/Slave,SAN |
|
|
|
Master/Slave,JDBC |
|
|
|
*Network Of Brokers |
|
|
|