Distributed online optimization for latency assignment and slicing

A non-transitory computer readable storage medium comprising a computer readable program for latency assignment in a system having shared resources for performing jobs, wherein the computer readable program when executed on a computer causes the computer to perform the steps of:

• computing a new resource price at each resource based on latencies in a previous iteration;
• sending the new resource price to a task controller that has at least one subtask running at the resource, wherein each subtask has an initial deadline;
• determining new deadlines for the subtasks in a task based on the new resource prices at the task controller;
• and
• sending the new deadlines to the resources where at least one subtask in the task is running to improve system performance.

A method for latency assignment in a system having shared resources for performing jobs, comprising:

• computing a new resource price at each resource based upon latencies in a previous iteration;
• sending the new resource price to a task controller that has at least one subtask running at the resource as feedback, wherein each subtask has an initial deadline;
• computing a path price for each path of the task at the task controller based upon latencies in the previous iteration;
• determining new deadlines for the subtasks in a task based on the resource prices and the path prices at the task controller by maximizing a Lagrangian of a constrained objective function describing subtask latencies;
• sending the new deadlines to the resources where at least one subtask is running;
• and
• iterating to update deadlines.