Anzere: replication policy scaling for personal storage networks

Опубликовано 17 августа 2016, 21:28

While managing a user's personal data (photos, contacts, music collection, etc.) with online service providers like Facebook, Google, or Yahoo offers many advantages, it is not without its drawbacks: a loss of privacy and control, lock-in, and vulnerability to provider failures due to attacks or insolvency. In this talk I will show how we built a personal storage system, called Anzere, where parts of the system use the heterogeneous ensemble of a user's devices, and parts are acquirable: the storage overlay has the option of acquiring one or more (virtual) resources, at some cost, if the benefit in data invulnerability and accessibility makes it worthwhile. Anzere makes personal data selectively available based on a set of user-specified replication policies. Compared to existing systems, it extends the expressivity of the policy language to include device-independence - referring to devices by predicate rather than name - yet still scales complex policy calculations up to very large numbers of data items. Core elements of Anzere include an overlay network, monitoring infrastructure, constraint logic solver, knowledge base, data replication, and Paxos. I will present initial results from trials of the system across personal computers, phones, and dynamically-acquired virtual machines from Amazon EC2 and PlanetLab.