A relaxed synchronization primitive for macroprogramming systems

Some sensor networks have large, non-deterministic communication delays which can be problematic because nodes must decide how long to wait before acting. A conflict arises when deciding on how much information is necessary: waiting a long time will improve accuracy but is detrimental to timeliness and acting quickly will improve timeliness but worsen accuracy. We present a relaxed barrier synchronization primitive that allows the programmer to make this tradeoff. A key challenge of a relaxed barrier is correctly setting exit conditions. For example, the number of nodes, radius of influence, and deadline can be adjusted to ensure the proper tradeoff between timeliness and accuracy. We provide a solution for discovering these values that combines both simulation and hill climbing. We show that by utilizing our primitive, application accuracy can be improved and maintained for many different scenario variations.