1 FICUS: A File System for Inter Cluster Unified Storage Arun Ramakrishnan and Mario Lauria Dept. of Computer and Information Science Ohio State University, OH 43210 (ramakria,lauria)@cis.ohio-state.edu Abstract There are no high performance file systems that allow sharing of data between different clusters presently. In this paper, we address this issue by developing FICUS (File System for Inter Cluster Unified Storage). FICUS provides the convenience of file system level access to remote data while preserving the performance of striped file systems such as PVFS. We achieve a bandwidth of 80 MB/s for local access using four I/O nodes, and 75 MB/s in accessing the same number of I/O nodes on a remote storage cluster. A parallel data intensive application achieves comparable performance when accessing 6 GB of data in local and remote storage. We show that careful pipelining of data transfer and proper integration with underlying file system and communication layers are crucial for preserving the performance of remote access. 1.Introduction The advances in inter- disciplinary sciences and technology trends have resulted in data- intensive applications coming to the forefront of the high performance computing (HPC) community. A rapidly growing number of applications in domains such as genomics/proteomics [3, 7, 9, 13, 15], astrophysics [4], physics [1], computational neuroscience [21], or volume rendering [2], need to archive, retrieve, and process increasingly large datasets. Therefore, the storage of large amounts of data has become a primary concern in the design of architectures for HPC. At the same time, clusters of PCs have become a popular platform for high performance computation at all institutional levels – from a single research group to a supercomputer center or a government laboratory. It is the advent of “killer micros” and “killer networks” that has allowed clustered commodity machines to compete with supercomputers in aggregate performance. However not all areas of the cluster technology have enjoyed the same dramatic advancements – one aspect that has lagged behind is the performance of the I/O subsystem. Early models of disk storage organization are still being used today and have not been significantly improved over the years. Two models account for most of the I/O system organization found on contemporary clusters. In the simpler approach, the Networked File System (NFS) is used to export shared file systems from one or more file servers to all the nodes, on each cluster node a local disk is used to store the operating system and for temporary local storage. Since today distributed file system services are integral part of every operating system, such scheme has the benefits of immediate installation and simple