Energy aware processing load distribution system and method

A method of operating a plurality of servers in a plurality of racks, comprising:

• maintaining an empirical thermal model of energy dissipation for each server, responsive to activity counters and software-accessible registers of each server;
• predicting a thermal trend of each server dependent on at least the empirical thermal model of energy dissipation of each server, and a workload of tasks allocated to each server;
• and
• allocating tasks from a queue of unexecuted tasks to respective ones of the plurality of servers, in dependence on at least a predicted effect of the allocated task on each server according to the empirical thermal model of energy dissipation for each server, an operating temperature of each server, and an energy efficiency of processing of the allocated tasks and the controllable proactive cooling for each rack dependent on at least the allocation of tasks from the queue of unexecuted tasks to respective ones of the plurality of servers;
• and
• proactively adjusting a cooling for each rack, responsive to at least a predicted thermal trend of each rack and the allocation of tasks, wherein the predicted thermal trend of each rack is dependent on at least the tasks to be performed by the servers in each rack and the empirical models of energy dissipation of the servers in each rack.

A datacenter, comprising:

• a plurality of racks;
• a plurality of servers in each rack;
• a cooling system configured to cool each rack;
• and
• a scheduling system configured to:
• maintain empirical thermal models responsive to activity of each server;
• predict a thermal trend of each server dependent on at least the empirical thermal models, and a utilization of each server;
• and
• allocate tasks from a queue of unexecuted tasks to the plurality of servers, in dependence on at least a predicted thermal effect of processing the task on each server, and
• a predicted energy efficiency of the processing of the allocated task and cooling of each rack;
• and
• adjust the cooling system for each rack to proactively cool each rack based on at least one task in the unexecuted queue of tasks prior to execution, responsive to at least a predicted thermal trend of each rack comprising the predicted thermal trends of the plurality of servers in each rack, and the empirical thermal models.