Recent advances in smart device technologies have enabled a new computing paradigm in which large amounts of data are stored and processed on mobile devices. Despite the available powerful hardware, the actual capabilities of mobile devices are rather limited as they are often battery powered. This work explores data caching for k-out-of-n computing in mobile cloud environments, with the goal of distributing data in a way that the expected future energy consumption for nodes to retrieve data is minimized, while preserving reliability. More specifically, we propose to place data caches (in addition to the originally stored data) based on the actual data access patterns and the network topology. Consequently, we formulate the cache placement optimization problem and propose a centralized caching framework that optimally solves the problem and a distributed solution that approximates the optimal solution. The distributed caching framework (DC) learns data access patterns by sniffing packets and informing a resident cache daemon about popular data items. Extensive evaluations are carried out through both simulations and a proof-of-concept hardware implementation. The results show that our proposed DC effectively improves the energy efficiency by up to 70% when compared with a no-caching framework, and even outperforms the centralized framework when taking the overhead into account.
Recent advances in smart device technologies have enabled a new computing paradigm in which large amounts of data are stored and processed on mobile devices. Despite the available powerful hardware, the actual capabilities of mobile devices are rather limited as they are often battery powered. This work explores data caching for k-out-of-n computing in mobile cloud environments, with the goal of distributing data in a way that the expected future energy consumption for nodes to retrieve data is minimized, while preserving reliability. More specifically, we propose to place data caches (in addition to the originally stored data) based on the actual data access patterns and the network topology. Consequently, we formulate the cache placement optimization problem and propose a centralized caching framework that optimally solves the problem and a distributed solution that approximates the optimal solution. The distributed caching framework (DC) learns data access patterns by sniffing packets and informing a resident cache daemon about popular data items. Extensive evaluations are carried out through both simulations and a proof-of-concept hardware implementation. The results show that our proposed DC effectively improves the energy efficiency by up to 70% when compared with a no-caching framework, and even outperforms the centralized framework when taking the overhead into account.
