Hi,
I would like to know, how the tasks are distributed to multiple nodes, when no capabilities are defined.
Best regards
Yannik
Daily Usage Issues
How does NCrunch distribute work on grid nodes when no capabilities are set?
Started by ysteiger on 6,583 views
Remco NCrunch Developer
#6542
14 Oct 2014 20:58 UTC
Hi Yannik,
This is quite a broad question - but I'll do my best to answer it in summary :)
The protocol works via a pull from the grid nodes. They request work from the client routinely in order to fill up their own local processors. The work requests are a round robin between all connected clients, which allows processing power to be shared fairly evenly between multiple connections.
When work is requested by a grid node, the client will make the choice as to which task is appropriate for the node. This takes into account activities, exclusively used resources, etc. If no work is available, the client simply returns a 'work request rejected' message.
At present there is no intelligence regarding the physical performance of the individual nodes. NCrunch doesn't yet know which nodes are faster/slower or more appropriate for certain types of tests, unless you disclose this via the capablities.
I hope this helps. Are you experiencing problems with your grid?
Cheers,
Remco
This is quite a broad question - but I'll do my best to answer it in summary :)
The protocol works via a pull from the grid nodes. They request work from the client routinely in order to fill up their own local processors. The work requests are a round robin between all connected clients, which allows processing power to be shared fairly evenly between multiple connections.
When work is requested by a grid node, the client will make the choice as to which task is appropriate for the node. This takes into account activities, exclusively used resources, etc. If no work is available, the client simply returns a 'work request rejected' message.
At present there is no intelligence regarding the physical performance of the individual nodes. NCrunch doesn't yet know which nodes are faster/slower or more appropriate for certain types of tests, unless you disclose this via the capablities.
I hope this helps. Are you experiencing problems with your grid?
Cheers,
Remco
Hi Remco,
thanks for your answer. Its not that I am facing problems with the grid. I do a second Demo of NCrunch in my company and I know there will come up questions about work distribution, when I come to the topic "Distributed Processing".
So, if I get it right:
As long as I do not specify restrictions on tests and grid nodes, the idle nodes (idle in terms of processing Ncrunch tasks) do request work chunks from the attached clients and its relatively random, which node gets the work.
Best regards
Yannkk
thanks for your answer. Its not that I am facing problems with the grid. I do a second Demo of NCrunch in my company and I know there will come up questions about work distribution, when I come to the topic "Distributed Processing".
So, if I get it right:
As long as I do not specify restrictions on tests and grid nodes, the idle nodes (idle in terms of processing Ncrunch tasks) do request work chunks from the attached clients and its relatively random, which node gets the work.
Best regards
Yannkk
Remco NCrunch Developer
#6546
15 Oct 2014 21:47 UTC
ysteiger wrote:Hi Remco,
So, if I get it right:
As long as I do not specify restrictions on tests and grid nodes, the idle nodes (idle in terms of processing Ncrunch tasks) do request work chunks from the attached clients and its relatively random, which node gets the work.
Technically the distribution of work isn't random. It will follow a circular pattern through the connected clients .. although I would say that if you were to observe it, it would probably seem very random :) So yes, I think you can safely say this!
Post a reply
Log in to reply.