Backup Server

fakefs can be used to implement a backup system for a workgroup of machines in geographically diverse locations. The process works in stages:

1. At regular intervals, each machine which is to be backed up (backup clients) will contact the backup server, which runs fakefs. Each machine will upload itself into a subdirectory of the fakefs filesystem using a modified or LD_PRELOAD-hacked rsync, or some similar utility.

2. Backup clients can access the backup server, either through a fakefs NFS server, or through an LD_PRELOAD-hacked Samba server, or through a fakefs-aware web server, to retrieve individual files should they be lost, or browse through previous versions of the filesystem.

3. The entire contents of the backup server are dumped to tape at regular intervals. Note that fakefs uses friendly filenames and directory structure, so you don't get such data restoration disasters as:
   • missing hard links,

   • entirely missing files due to confusion caused by hard links,

   • case-mangling backup file formats (e.g. ISO-9660 or anything touched by Microsoft),

   • poor performance on slow files (e.g. millions of 4K files transferred over NFSv2, or anything touched by Microsoft),

   • loss of Unix-specific file attributes, and

   • entirely missing directory hierarchies due to insufficient credentials for the backup process.

As long as the Berkeley DB guidelines for database backup are followed, it should be possible to back up a live fakefs filesystem. The main concern is the non-Berkeley data (external files). Fortunately there isn't much of this data.