Print

[iammahendra]

Hi All,

I see the two major areas in computing to be supplementing each other -
a) Autonomic Computing
b) Grid Computing

The way i see this going forward is we have grids which has -
a) Self configure
b) Self deploy
c) Self heal

autonomic capabilities.

I see the point b) self deploy is being currently handled with JPPF. Please correct me if i am wrong. If we can add the a) and c) features (again please correct me if i am wrong, if they are already there) then we can be a true differentiator as Open source grid providing all the autonomic capabilities.

Any thoughts ?

Regards
Mahendra

[lolo]

Well, I do have a few thoughts - I'd feel naked if you caught me without one  
It's actually mostly questions:

1) what do you mean exactly by self-configure? I can think of a few functionalities, like adaptive performance tuning, automatic peer dscovery and things like that, but I'm not sure it's what you hint at. Am I close?

2) I think we do agree on self deployment, even though we still have to extend it to a federation of server type of architecture, which is not a trivial affair.

3) Self heal: currently nodes, server and clients have some failover/recovery features.
Nodes and clients automaticaly attempt to reconnect to the server, but they do not restart themseleves in case of crash. The server could be extended to restart itself under specific conditions, rather than just relying on request from the admin interface.

I don't know if it helps a lot,  but I'd be grateful for your feedback.

-Laurent

[iammahendra]

I like your sense of humor    but you are surely very near and not caught naked :wink:

1) What do you mean exactly by self-configure? I can think of a few functionalities, like adaptive performance tuning, automatic peer dscovery and things like that, but I'm not sure it's what you hint at. Am I close?

3) Self heal: currently nodes, server and clients have some failover/recovery features. Nodes and clients automaticaly attempt to reconnect to the server, but they do not restart themseleves in case of crash. The server could be extended to restart itself under specific conditions, rather than just relying on request from the admin interface.

Yes this is close. Thought is to come close to system which if requires based on performance metrics or some provided guidelines can re-configure or self-configure to maintain the desired output.

This is also tied to self healing i.e. once it can self configure to desired output or guidelines defined it is self healing application.

In grid computing frameworks its not a trivial affair (as you mentioned). May be this is a far off thing but i believe it is really worth to have -

a) Nodes and Servers configurations based on some predefined desired output say e.g. OS can be reconfigured and recover back.

b) A control feedback loop which keeps checking which node is deviating and provide a feedback to the node to maintain the optimum value say by changing the configurations. There can be ways other than changing configurations.

This is something coming from my research pre-work, which i am going to start in Autonomic computing area soon.

Lastly, i am not too sure if these ideas are -

a) Needed for the grid frameworks
b) How easily can these be designed and implemented them for grid frameworks.

Hope this helps in clarifying my thoughts !!

- Mahendra

[lolo]

a) Needed for the grid frameworks

Definitely

b) How easily can these be designed and implemented them for grid frameworks.

It sure isn't easy, but that why this project is so exciting to me.

Regarding the self configuration, you might want to discuss with Domingos, he is the one who designed and implemented the automatic task bundle size tuning. That gives us an adaptive performance tuning, with a strategy similar to a Monte Carlo. Each node has its own tuner, which focuses on computing the bundle size to minimize the network I/O overhead. Each bundle execution provides feedback data that is immediately integrated.