Multicore Runtime Management Using Process Affinity Graphs

A computer storage medium having computer-executable instructions stored thereon for a method which, when executed by a computer adapt the computer to manage processes in a multicore processor using process affinity graphs, the method comprising:

• monitoring message traffic between processes in the multicore processor;
• determining affinity scores between the processes based on the monitored message traffic;
• establishing a process affinity graph based upon the affinity scores;
• and
• allocating one or more processes to one or more processor cores of the multicore processor based upon groupings identified within the process affinity graph.

A computing system configured to manage processes in a multicore processor using process affinity graphs, the computing system comprising:

• a multicore processing unit including multiple processor cores;
• a memory associated with the multicore processing unit;
• and
• a processing module stored in the memory and configured to execute on the multicore processing unit such that the multicore processing unit is adapted to
• monitor message traffic between processes in the multicore processing unit,
• determine affinity scores for processes based on the monitored message traffic,
• establish groupings of the processes based on the affinity scores, and
• allocate each grouping to a respective core of the multicore processing unit.

A computer-implemented method for managing processes in a multicore processor using process affinity graphs, the method comprising:

• tracking interprocess traffic between two or more processes executing on the multicore processor;
• determining affinity scores for the processes based on the tracked interprocess traffic such that a first pairing of processes has a higher affinity score than a second pairing of processes in response to the first paring of processes having higher tracked interprocess traffic than the second paring of processes;
• and
• allocating processes to processor cores within the multicore processor according to affinity scores such that processes having higher affinity scores execute on physically proximate processor cores to support reduced interprocess messaging latency.