PerCom 2013

I've been in San Diego for the last few days for PerCom 2013. Besides enjoying nice weather and environment, I presented two of our workshops papers at PerCol and PerHot both on the first day of the conference. First one is about enabling proximity alert service on smartphones for pervasive applications, and the second one reflects our group's vision for ubiquitous computing.

Energy Efficient Proximity Alert on Android
Proximity alert service is an enabler for various pervasive applications including but not limited to "personal location-based reminders", "safety apps notifying dangerous neighborhoods", "auto-check-ins" and "social pervasive apps". In one of our recent study, i.e. LineKing, we utilized proximity alert service on smartphones to estimate line wait-time in coffee shops.

However, proximity alert inherently depends on frequent checking of locations to determine if the device enters or exits from a point of interest (POI). Thanks to the advanced localization capabilities of current smartphones, one can use various location providers (i.e. GPS, Wi-Fi APs (WAPs), Cell towers), to obtain device's location. Albeit, these providers have different accuracy and energy expenditures and irresponsible usage could easily drain the battery in a few hours.

In this work, we first identify the problems in the current smartphone operating systems and propose design principles to overcome the limitations. We specifically focus on Android, however the design principles that we talk about could easily be applied to other smartphone operating systems too. In Android, we find out three limitations that cause proximity alert service to spend too much battery. These are, static use of location providers, static and frequent periodicity of location updates and lastly the underutilized inertial sensors that could be useful for saving energy.

Afterwards, we design a clever algorithm that uses user's distance from the POIs and user's transportation mode (i.e. idle, walking, driving) to choose appropriate location provider and optimal location sensing frequency. We implement our proposal as a middleware service in Android by modifying the AOSP. Our results for a realistic scenario improves the battery consumption by 75% comparing to the baseline proximity alert service in Android and put the applications that use proximity alert service into the realm of possibility.

Eywa: Crowdsourced and Cloudsourced Omniscience
In this work, we propose an open publish subscribe framework for pervasive computing. Although Mark Weiser called computers to come outside and be fabrics of our daily lives, the trend went the other way; most of our physical services become virtualized and moved into the computer or internet. Physical mail replaced by e-mail, shopping replaced by online shopping, even friendship got virtualized and moved into social networks such as Twitter, Facebook etc..

However, this doesn't mean that Weiser's vision was lacking or incomplete. Instead, it was the natural flow of the things. A ubiquitous computing system has both physical and virtual component and the virtual component has gotten the most attention because it was more easier to design and implement. However, we've also seen an advancement in the physical component too. Especially with the advent on smartphone technology, location-based services and systems are getting more prevalent.

In this work, we argue that although the progress is astonishing, it is still not complete. Although, we have devices and the tools to realize ubiquitous computing vision, lack of an infrastructure to task, coordinate and collaborate these devices jeopardize the realization of ubiquitous computing vision. Therefore, in this vision paper, we propose Eywa which is an open publish-subscribe framework and mainly consists of 4 main components: Sensing & Aggregation component which sense and collect information from various smartphones and sensor networks, Computing and Machine Learning Component which handles the large-scale computation and provides the intelligence of the system, Social networking component which enables the socially relevance tasking/computing and finally Crowdsourcing component which mainly responsible for tasking the devices.

In order to keep the system sustainable, we also propose a virtual coin market where an app that needs data, should first pay for it. Similarly, if an app provides data to the system, then it can make virtual coin out of its data based on the value it provides. This mutual symbiotic relationship keeps the system in equilibrium and makes Eywa sustainable.

Others notes from PerCom 2013
Besides from my presentations, I would like to share some of my thoughts related to PerCom 2013. Although, I just attended a few conferences so far, in general I really like conferences. Especially, the interaction it provides is invaluable. However, I have to admit that I don't get enough benefit from most of the presentations. The lack of English in some of the presentations are so prevalant.  Besides, slides in most of these presentations are amateurish and cluttered with lots of texts which obviously make them hard to understand. Conferences are for interaction and getting to know about works people published. However, some of the presentations (maybe most) in the conference are total failures. Although, it would have been harsh, I wish, papers can be rejected from the proceedings if the presentation does not meet some certain criteria. And finally, I was surprised to see that posters are so much effective than I thought. Besides from getting feedback and advertising your work, it also provides a mean for interaction. This makes me to rethink about poster submission for the upcoming conferences.