Is it possible to have a dynamic model for data and computations being distributed. This should be achieved by analysing Metrics, data access patterns, scheduled computations and programmatically specified access plan (if provided) and decide what is the the best way to distribute the particular data sets in the cache and associated computations.
The implication is instead of a node wide model the model would be based on sets of data and computations. (This would mean if you exclude certains from dynamic auto configuration then you can specify the caching model for the data set.) Also the cluster can dynamically re organise itself.
http://www.paralleluniverse.co/ has a complete dynamic model. But I think choice is better there you can which data is treated in which way. Also ability to force a node to have a norde wide policy.
Is it possible to have a dynamic model for data and computations being distributed. This should be achieved by analysing Metrics, data access patterns, scheduled computations and programmatically specified access plan (if provided) and decide what is the the best way to distribute the particular data sets in the cache and associated computations.
The implication is instead of a node wide model the model would be based on sets of data and computations. (This would mean if you exclude certains from dynamic auto configuration then you can specify the caching model for the data set.) Also the cluster can dynamically re organise itself.