Optimizing mobile computational offloading with delay constraints

Abstract

Computation Offloading, sending computational tasks to more resourceful servers, is becoming a widely-used approach to save limited resources on mobile devices like battery life, storage, processor, etc. Given an application that is partitioned into multiple tasks, the offloading decisions can be made on each of them. However, considering the delay constraint and the extra costs on data transmission and remote computation, it is not trivial to make optimized decisions. Existing works have formulated offloading decision problems as either graph-partitioning or binary integer programming problems. The first approach can solve the problem in polynomial time but is not applicable to delay constraints. The second approach relies on an integer programming solver without a polynomial time guarantee. We provide an algorithm, DTP (Deterministic delay constrained Task Partitioning), to solve the offloading decision problem with delay constraints. DTP gives near-optimal solution and runs in polynomial time in the number of tasks. Going beyond prior work on linear delay constraints that apply only to serial tasks, we generalize the delay constraints to settings where the dependency between tasks can be described by a tree. Furthermore, we provide another algorithm, PTP (Probabilistic delay constrained Task Partitioning), which gives stronger QoS guarantees. Simulation results show that our algorithms are accurate and robust, and scale well with the number of tasks.