Print

[fariszuriekat]

Hello,

I have been working on developing similar client/server architecture fro grid computing, but I have been able to perform thread monitoring on Parallal Programs ( MPI , PC, ..etc). Allowing new features like shortcutting  and concurrent execution management available.

I would be a great idea if I can embed my architecure into JPPF , meaning that JPPF will have more capabilites to run language independent programs and perform execution management and control, In addition , and most important is increasing the perfomance of JPPF.

I am interested to work on JPPF project, if my proposed idea sounds feasible, please contact me back.


Thanks
Eng. Faris Zuriekat

[Arrouan]

It's very interesting.
One of the things i need and i don't find in JPPF is a good monitoring capability for the task.
I have some tasks that take more than 10hours to compute and I can't monitore efficiently them.
What is your project ?
How can i help you to include theses functions into JPPF ?

[lolo]

Hello,

This is indeed an interesting feature to include in JPPF.
I would like to like to point to a number of constraints and or requirements that would have to be met:

• JPPF is designed to scale from small-sized networks to large ones (we're aiming at thousands of nodes). A node monitoring framework would thus have to scale to the same level and/or provide the capability to be enabled/disabled on demand. We must be aware of what level of traffic will be generated due to monitoring events.
• Unlike grid toolkits based on MPI or other standards for distributed/parallel computing, JPPF is entirely designed for Java. The Java standard for monitoring and management is JMX, and that is the road we want to follow. We are actually migrating or existing administration and monitoring tool to a JMX-based framework.
• If I understand well, we are talking about task-level monitoring, not merely node level. This means the task APIs have to be extended to enable monitoring capabilities. On the other side, the monitoring tools have to be extended to enable generic ways to represent the monitoring data in a usable form.

This is definitely an exciting feature, and it is definitely one that will require a lot of reflexion and work.

I would like to hear your input on this.
Thanks for your time,
-Laurent

[fariszuriekat]

hi Arrouan,

Yes I am really interested to get involved

My project is named PRIMM, as I mentioned before, PRIMM has the capability to solve non-deterministic problems, it has also a new O.O feature called shortcutting remotely and locally.  You can perform monitoring at any point of execution and also you can at any point change the direction of the computations.

When I am completely done with PRIMM API, I will let you know, It is going to be very easy to use though.
I will need some help if I want to import it to JPPF.

[fariszuriekat]

Hello lolocohen

The PRIMM has close architecture to JMX, PRIMM is also designed for Java, but to be language independent in implementation, PRIMM cannot distribute tasks on large scale as JPPF, but hopefully trying to cooperate with JPPF and increase its capability.

Actually I only tried it to perform interactive monitoring on MPI-Cluster. It performs a process streaming in (UDP or TCP/IP). It sends process status to clients and accepts controls messages at any time.

PRIMM may be added as a utility in JPPF,I am not sure how easy to do that, but it may be worth trying.

I will send you more details when I am done.

Thanks for replying
Faris Zuriekat