Prospects for Shaping User-Centric Mobile Application Workloads to Benefit the Cloud

Approaches to making cloud operation more efficient, for example through scheduling and power management, largely assume that the workload offered from mobile, user-facing applications is a given and that the cloud must simply adapt to it. We flip this assumption 180 degrees and ask to what extent can we instead shape the user-centric workload into a form that would benefit such approaches. Using a toolchain hat allows us to interpose on frontend/backend interactions in popular Android applications, we add the ability to introduce delays and collect information about user satisfaction. We conduct an "in the wild" user study using this capability, and report on its results. Delays of up to 750 ms can be introduced with little effect on most users, although this is very much user and application dependent. Finally, given our study results, we consider reshaping the application requests by selective delays to have exponential interarrival times (Poisson arrivals), and find that we are often able to do so without exceeding the user's delay tolerance.