VSFS: A Versatile Searchable File System for HPC Analytics

Emerging HPC analytics applications urgently demand filesearch services to drastically reduce the scale of the input data in real-time, so that the speed of computation and data analytics can be greatly accelerated. Unfortunately, the existing file-search solutions are either poorly scalable for large-scale systems, or lack a well-integrated interface to allow applications to easily use them for critical tasks. We believe that the time is ripe for the design of a searchable file system capable of accurate and scalable system-level filesearch functionality. In this paper, we propose a Versatile Searchable File System, VSFS, which provides a transparent, accurate and real-time file-search service through a POSIX-compatible file system namespace that can be integrated into any HPC/Big Data legacy code without modifications. Additionally, to support real-time file search, VSFS uses a DRAM-based distributed architecture to perform real-time file indexing. Moreover, a versatile index scheme is designed to adapt to the various forms of HPC datasets. The results of our VSFS prototype evaluation show that VSFS is scalable in a typical HPC environment. It achieves significantly better file-indexing and file-search performance than the popular SQL/NoSQL solutions, while it only introduces negligible I/O overhead. Finally, we integrate VSFS to a scientific analytics application to show its benefits in terms of performance and ease of use.