The SMFS architecture is illustrated in Figure 5. It works as follows. Clients access file system data by communicating with a file server (e.g. using the NFS protocol). File servers handle writes in a similar fashion to LFS. The log is updated by traversing a file system in a depth-first manner, first appending modified data blocks to the log, storing the log address in an inode, then appending the modified inode to the log, and so on [27,29]. Reads are handled as in any conventional file system; starting with the root inode (stored in memory or a known location on disk) pointers are traversed to the desired file inode and data blocks. Although file servers provide a file system abstraction to clients, they are merely hosts in the storage system and stable storage resides with separate storage servers.