Energy Aware Information Retrieval from Mobile Phones

On mobile phones, Information Retrieval (IR) applications that search over remote collections consume significant energy as a result of network activity. Typical IR applications mainly focus on retrieving and presenting relevant documents to users in an effective manner and do not take into account the cost of network data transfer. Despite the improvements in network access, energy consumption of network activity can seriously reduce battery lifetime constraining the use of IR applications. Therefore, on mobile phones in addition to being effective an IR application must also be energy efficient. In this paper, we study the energy consumption of a typical IR application on a mobile phone for the task of retrieving documents residing on a remote server. Our measurements indicate that the network activity of typical IR process consumes a substantial amount of energy and can reduce battery life (up to 50%) when compared to a completely local IR application. To explore opportunities for saving energy, we first build a model for the energy consumption of network activity induced by the IR application. Motivated by the characteristics of this energy model, we explore two opportunities for reducing energy consumption of a typical IR application: 1] Reducing the overall amount of data transferred using relevance feedback and 2] Reducing the number of data transfers for each query by bundling more data at each transfer opportunity. The modifications we propose provide substantial savings in energy (up to 25%) with additional benefits and trade-offs incurred in dead time and latency. Actual implementations of the proposed modifications confirm our model based findings.