The Math Department file server, math.rice.edu, is a Dell Poweredge 2970
running Red Hat Enterprise Linux. Data is stored on a RAID volume, and
it has many redundant features and failsafe mechanisms to protect
department data.
Nonetheless, hardware and software failures occur even on
state-of-the-art equipment, and the best policy is to back up data
frequently, with multiple copies. To accomplish this, Rice IT has
provided disk-to-disk backups according to the following schedule:
- Daily disk-to-disk rsync (Sunday-Friday evening at 9pm)
- Weekly disk-to-disk rsync (Saturday evening at 5pm)
Rsync backups are made over the network to a separate storage device
with redundant capabilities (i.e., RAID5 plus hot spares).
Backups are made of the application tree (/divsite and /local), home
directories, and /etc (the configuration directory). These are copies
made with rsync's "delete" option, which means that if a file is deleted
on the source disks, it is also deleted on the copy at the next sync
time. This helps to conserve space on the backup media.
In practical terms, the mirror means users can recover an accidentally
deleted file if it was created a day or more earlier and they realize
they want it back before the next day's sync. The weekly copy means
that any files created prior to a week ago, which were accidentally
modified or deleted, can be recovered until the next weekly sync.
This type of backup scheme is really designed to do two things, in order
of importance:
- Disaster recovery (i.e., math server and its data are damaged,
destroyed, or lost, we have a physical mirror on another disk set and at
most a day's work is lost)
- Give users a reasonable (but not 100%) chance of recovering any data
that they accidentally delete, provided that they notify IT as soon as
they can to ask for the files.
This is *not* designed to guarantee that a user can always recover
accidentally deleted files, nor is it designed to maintain longer term
archives of data. It also does not provide "offsite" disaster
recovery, in the event of a catastrophic loss of the entire Rice data
center. The very best data protection scheme for any user is to have
personal copies of your most critical data in more than one location.
Specific requests for help or information should be addressed to
helpdesk@rice.edu
to the attention of Joseph Ghobrial.
