A hierarchical namespace is a common abstraction used for data organization within modern file systems. Fast translation of namespace objects to physical locations is necessary to carry out efficient file system operations. For reasons attributed to modularity, security, and to some extent legacy, namespace translations involves iterative translation of intervening directory objects from the root of the namespace. Namespace resolution is typically a multi-step process, potentially involving serialized I/O operations at each step. In this paper, we propose a rethink of the strategy to fetch pathname entries. Our technique, StepAhead, proactively utilizes hints about namespace translation lookup failures to enable parallel and just-in-time fetching of necessary path translation data into memory to increase cache hits significantly. With StepAhead, we measure an increase in cache hit rates for path translation data across a set of six workloads by as much as 51%, which in turn results in application speed-up of as much as 20%.
