1. Information about the paper
Deng, Shuo, and Hari Balakrishnan. “Traffic-aware techniques to reduce 3G/LTE wireless energy consumption.” In Proceedings of the 8th international conference on Emerging networking experiments and technologies, pp. 181-192. ACM, 2012.
The 3G/LTE wireless interface is a significant contributor to battery drain on mobile devices. A large portion of the energy is consumed by unnecessarily keeping the mobile device’s radio in its “Active” mode even when there is no traffic. This paper describes the design of methods to reduce this portion of energy consumption by learning the traffic patterns and predicting when a burst of traffic will start or end. We develop a technique to determine when to change the radio’s state from Active to Idle, and another to change the radio’s state from Idle to Active. In evaluating the methods on real usage data from 9 users over 28 total days on four different carriers, we find that the energy savings range between 51% and 66% across the carriers for 3G, and is 67% on the Verizon LTE network. When allowing for delays of a few seconds (acceptable for background applications), the energy savings increase to between 62% and 75% for 3G, and 71% for LTE. The increased delays reduce the number of state switches to be the same as in current networks with existing inactivity timers.
2. My review of the paper
- I agree that this is good enough for its goals.
- I also agree that user context might help in this case. I also mentioned in previous review that this would be scheduling problem for developers. However, now I think app developers cannot independently decide. For example, a user might receive a link in Facebook, and then decided to open it on the browser. There might be dependency between apps.
- Therefore, I think the best solution is to incorporate this in the OS (with additional user context).
- The user is the one who really knows about what he does with smartphone. A user could schedule turning on and off any feature according to his daily schedule using currently available If This Then That App (IFTT) and its equivalent. However, the goal is to do it automatically. Every automatic scheme will likely use machine learning, which cannot be accurate at all time. My solution is, in addition to machine learning, user input in the process will be really valuable. For example, if a user knows there will be significant changes in his pattern, like he moves to other city or changes jobs, instead of waiting for machine learning we could offer him a button so he could reset the training data thus accelerate the training process.
- We cannot turn off 3G/LTE while using unstable WiFi for Internet dependent application like video call. I have an experience when I connected to WiFi and tried to turn off 3G/LTE during a video call, iOS warned me that it might end the video call (FaceTime). And it did end the video call. Then I started it again with 3G/LTE already off. I got connected but with lower quality. Then I turn the 3G/LTE back on. The conclusion is we can only turn off 3G/LTE while the WiFi is stable (maybe on the 80% signal on WiFi indicator) to maintain user experience.