Using Hoarding to Increase the Availability in Shared File Systems

Paper i proceeding, 2009

Many mobile devices have reached the point where the users' (active) working set is smaller than the amount of storage available and that trend is likely to continue. Currently these resources are made available for recording new data, but we think that we could make better use of this capacity. Hoarding previously not accessed data could give better data coverage in cases of disconnected operation, when wireless networks are not available or access to them is expensive.We gathered a trace from a university file system used by more than 5000 people over a period of 16 months. This trace is used to drive a simulation model of distributed file systems. This paper studies a novel hoarding scheme that uses the access profile of other users to predict what files a user would need in the future. This hoarding scheme is shown to avoid between 30% and 75% of remote file accesses to files that are accessed for the first time. Furthermore, hoarded but not used data can be expired, because we note experimentally that the population shifts focus each month.