ErdOS: Achieving Energy Savings in Mobile OS Narseo Vallina-Rodriguez Jon Crowcroft Computer Lab Computer Lab University of Cambridge University of Cambridge JJ Thomson Avenue JJ Thomson Avenue Cambridge, CB3 0FD Cambridge, CB3 0FD United Kingdom United Kingdom [email protected] [email protected] ABSTRACT Both hardware and operating system manufacturers found posi- The integration of multiple hardware components available in cur- tive and interesting approaches to extend the battery life of mobile rent smartphones improves their functionality but reduces their handsets at different hardware and software levels. However, their battery life to few hours of operation. Despite the positive im- efforts are limited by the strict layering of the system that makes provements achieved by hardware and operating system vendors diff cult to exploit cross-layer optimisations that might otherwise to make mobile platforms more energy eff cient at various levels, be fairly straightforward. As an example, and unlike with laptops, we believe that an eff cient power management in mobile devices the operating system does not get direct access to information about is compromised by strict layering of the system caused by com- aspects of the handset hardware such as telephony and radio hard- plex mobile business models that mitigates against cross-layering ware power consumption. The reason behind this limitation is a optimisations. However, there is a lot of room for improvement in complex business ecosystem in which multiple players (e.g. cellu- the operating system. This paper presents ErdOS, a user-centered lar network providers, content providers, cloud service providers, energy-aware operating system that extends the battery life of mo- hardware manufacturers and operating system vendors) compete to bile handsets by managing resources proactively and by exploit- retain their share of the mobile business. New open platforms like ing opportunistic access to resources in nearby devices using social Android and Nokia’s Maemo offer new opportunities for improve- connections between users. ment. This paper presents ErdOS, a mobile operating system that ex- ploits user-centered optimisations to extend the battery life of mo- Categories and Subject Descriptors bile handsets. We believe that one of the reasons behind mobile en- D.4.7 [Operating Systems]: Organization and Design—Dis- ergy ineff ciency is that current operating systems do not naturally tributed Systems control access to energy-consuming resources to applications tak- ing into account the patterns of users interactions with their hand- sets. Two techniques that ErdOS integrates to solve this problem General Terms are: Design • A proactive resources management system that predicts the future resources demands and status based on the users’ Keywords habits and preferences. Mobile computing, operating systems, energy-awareness, re- • sources management, resources sharing, opportunistic computing Opportunistic access to computing resources available in nearby devices using local wireless interfaces and informa- tion about users’ social networks (which can be obtained 1. INTRODUCTION from online services, email accounts and address book) to Current mobile platforms integrate sensors such as GPS, several provide access control policies. types of wireless interfaces, a giga-hertz range multicore CPU and a touchscreen. This trend bootstrapped the birth of rich mobile The system attempts to make optimal use of all the resources applications that, despite improving the usability of the device, can available in the environment in a distributed fashion taking into ac- become energy sinks depending on the way users’ interact with count the situation, the users’ preferences and both local and re- their handsets. In fact, the state of the art of lithium-ion batteries mote available resources. Our earlier work on understanding the clearly indicates that capacity will still be constrained by design impact of mobile users on resources demand [14] and the results parameters such as battery size and weight for years to come. we obtained by simulations (which are detailed later), clearly indi- cate that it is possible to achieve energy savings with those tech- niques without truncating user experience. However, a system like ErdOS opens new technical and research challenges such as fair Permission to make digital or hard copies of all or part of this work for scheduling algorithms for distributed resources in dynamic scenar- personal or classroom use is granted without fee provided that copies are ios, energy-aware access control policies for sharing resources, ad- not made or distributed for prof t or commercial advantage and that copies equate inter-process communication (IPC) mechanisms for access- bear this notice and the full citation on the f rst page. To copy otherwise, to ing a diverse range of remote resources and f nally, non computa- republish, to post on servers or to redistribute to lists, requires prior specif c tional intense techniques for monitoring and forecasting resources permission and/or a fee. MobiArch’11, June 28, 2011, Bethesda, Maryland, USA. demands and state. Nevertheless, we allow the user to decide Copyright 2011 ACM 978-1-4503-0740-6/11/06 ...$10.00. whether to enable or not the automatic features of ErdOS. Some 37 users may prefer to get feedback from the system about future en- as public places or urban and rural locations) with more stationary ergy limitations or resources unavailability to adapt the way they nodes and more chances of establishing opportunistic connections interact with their devices rather than enabling automatic resources with nearby devices. management in order to extend the battery life. Energy consumption per hardware module 2. MOTIVATION Near (30 cm) 36.0 mW Bluetooth Past work reported in the literature proposed a re-examination Far (10 m) 44.9 mW of some aspects of operating systems design and implementation Idle 8mW WiFi from an energy eff ciency perspective, rather than the more tradi- Full Capacity 720 mW tional target of maximising performance [13]. Researchers have Idle 58mW GSM recognized the mismatch between the original design assumptions Full Capacity 620 mW underlying the resource management mechanisms of operating sys- tems and applications’ behavior [4]. As we reported in our previous GPS 143.1 mW work [14], that incongruity is even more dramatic in mobile devices where simultaneous use of the diverse hardware systems in a mod- Table 1: Detailed power consumption on a modern smartphone ern multitasking smartphone arising from personal usage patterns (Openmoko Neo Freerunner) can limit many handsets to just a few hours of operation. In practice, a modern mobile operating system will attempt to extend the battery lifetime of the handset by making selective use All this type of evidence motivates our work on ErdOS; a mobile of the available resources. However, experience shows that this is OS that exploits contextual information to manage resources proac- not eff cient. Choices are most often implemented through the use tively by learning from the usage patterns of the mobile user (or as of static policies and by standby power states, automatic control of it has been said, provides feedback to the users about their energy the screen backlight, and actively switching particular subsystems consumption) and also capable of providing transparent access to (such as networking interfaces) on or off as demand dictates. They remote resources without impacting the user experience with mo- do not take account of the dynamism of the users’ interaction with bile applications. We claim that an operating system that supports applications and the importance of context on resources availability these features can also bootstrap many new collaborative applica- (e.g. location impact on network coverage). As a result, it is quite tions that require accessing resources that span over several mo- possible for a power-hungry application to drastically shorten the bile devices such as collaborative sensing [8] or distributed com- operating time of the handset. puting [9]. There are daily situations in which mobile applications can drain the battery due to inadequate management of resources. A common 3. PRELIMINARY RESULTS activity such as synchronising the email client is a good example. In addition to our prior work on understanding users’ interaction Performing this kind of action can be energy-expensive since it re- with mobile phone resources [14], we run a simulation to demon- quires waking up the CPU and the radio interface when the handset strate the potential energy benef ts of sharing GPS readings with is in idle mode followed by a DNS request and a connection with nearby devices 1 (one of the use cases explained in the previous the email server. Such action happens regularly even in situations section). We used the dataset collected during one of the hikes that when its execution does not improve the usability of the system were part of ExtremeCom’09 workshop in Padjelanta National Park (e.g. at night when the user is sleeping) and in scenarios where it (Sweden) [15]. may well be known that there is no network coverage from previous This adversarial environment where the hike took place lacks of experiences. any infrastructure such as power access points and network cov- On the other hand, previous experiments on human mobility erage. Padjelanta National Park is a completely desert land in the and social interaction