Replica placement for p2p redundant data storage on unreliable, non-dedicated machines

P2P architecture appears to fit for enterprise backup. In contrast to dedicated backup servers, nowadays a standard solution, making backups directly on organization’s workstations should be cheaper (as existing hardware is used) and more efficient (as there is no single bottleneck server). However, non-dedicated machines cause other challenges. Update propagation algorithms must take into account frequent transient failures (computers turned off for nights or weekends). Hardware and network topology are heterogeneous. A replication system must not only handle resulting constraints (such as e.g. varying amount of disc space), but also take advantage of the diversity (and e.g., distribute replicas geographically). We present a p2p backup system using pairwise replication contracts between a data owner and a replicator (storing one of the replicas). In contrast to storing the data directly in a DHT, the contracts allow the system to optimize replicas’ placement depending on features of the data owner, the replicators and the network topology. To cope with replicators’ transient failures, the system propagates notification about the updates through asynchronous messages. Each peer has a few synchro-peers that, in case the peer is off-line, store these notifications. Once back online, the peer contacts its synchro-peers, and then explicitly requests latest versions of modified files it replicates from the owner, or from other replicas. We implemented the system and tested it on PlanetLab as well as 150 computers in the faculty’s student computer labs. We experimentally assess the cost of unavailability showing how the time needed for data backup increases non-linearly with machines’ unavailability. Individual workstations are certainly less reliable and less efficient than a dedicated backup server. Yet, as our experiments show, given a sufficient number of such workstations, a p2p backup system outperforms a centralized solution, benefiting from better scalability and distributed topology. P2P is a cheap and robust alternative for enterprise backup.