Smartphone batteries don’t last as long as they used to. Why? Applications consume a lot of energy. However, the authors show that, besides the display, which is the most power-hungry service, all tasks related to the radio interface consume large amounts of energy.

The authors propose a prototype of a service called smart on demand (SOD). This service runs the radio interface layer daemon (RILD) in a low-power secondary microcontroller. This service reduces energy consumption when no other apps are active. The prototype has been tested under real user traces of representative smartphone usage.

Phone calls are managed by the RILD. This service checks for any incoming phone call to start the connection. The authors tested deactivating and resuming RILD for different configurations. For a LG V10 smartphone, the power consumption before the RILD suspension is 70mW. On the other hand, the power consumption of the system after RILD suspension is reduced to 39mW. However, the RILD suspension procedure itself consumes as much as 200mW. Therefore, it is necessary to take a long suspension time to compensate the suspension power overhead. They show that RILD should be suspended for a period longer than 60 seconds to obtain power consumption benefits. Using the user traces of smartphone usage, the authors show that two consecutive phone calls in an interval shorter than 60 seconds has a probability between five and ten percent (with 17 percent in the worst case). Therefore, suspending RILD will lead to large power savings, as most intervals are longer than 60 seconds.

However, RILD is also responsible for cellular data management and updates. RILD uses the cellular network for sending and receiving updates for the smart applications running in the background. Most applications require less than 15 seconds to update all the information. The authors, in testing, suspend RILD for intervals ranging from one to 30 minutes. These values include most update intervals of the most common applications (such as WhatsApp, ranging from five to 25 minutes; Facebook, 30 minutes; and Outlook, 15 minutes). Their experiments show that using an RILD suspension interval of 20 minutes results in an increment of five percent of power consumption (compared to a no update at all policy), or a 15 percent of power consumption (for the same no update policy) using a five minute interval.

The authors provide a very large set of experiments to test the savings of the SOD proposal in different environments and configurations. Those experiments show an average battery life extension of 2.5 times the standard configuration. Moreover, they show, with real user traces, that SOD is able to extend the battery life from 0.3 to 1.4 days in one dataset, and from 0.25 to 1.6 days in the other dataset, compared to the standard configuration using a big.LITTLE application processor.

This paper may represent a quantitative jump in the battery life extension of smartphones using a very simple technology. I recommend this interesting proposal for any smartphone enterprise designer.