|
This document is aimed at end users of Theory sector systems and
describes what is backed up and when it is done.
The useful information for most people is in the section on the main backup service.
The main sector backup service
Most of the workstations and servers in the Theory sector are
backed up to the central backup servers in the Unilever
comms room, so the servers are physically separated from the machines they
back up.
The client machines are divided into groups and backed up each night
using rsync over ssh. This means that the data is encrypted in transit. The previous night's backup plus two week's
worth of incrementals are kept on disk. This means
that you can only restore things deleted less than two weeks
ago.
Certain things are excluded from the backups by default:
-
/tmp/*
-
/scratch/*
-
RESTART*
-
core
-
/mnt/*
-
/media/*
-
.netscape/cache
-
/*/*/.mozilla/*/*/Cache
The /sys, /proc, and all NFS filesystems are also excluded.
Restoring from the Backups
The backups are now user-accessible. If a machine is on the new-style backup system you should find its backups NFS-mounted on /rsnapshots on the machine itself. Just cd /rsnapshots. Some machines are still on the older system and the COs have to do restores from those. Please email support@ch.cam.ac.uk . You can see the status of the backups in the Hobbit system monitor at http://hobbit.ch.cam.ac.uk/hobbit/bb.html (Raven protected).
The Wales group have a small backup and archive server, silver.ch.cam.ac.uk. This server
holds the data from some retired Sun machines (old mek-quake, old clust, defiant, nemesis), and is also a
user-accessible backup server for the current clust, zippo, volkhan and mek-quake. It also backs up a
couple of workstations (lovebomb, steelhorn). This is the best
place to start if you have deleted things from your homespace.
There is a second small server, dw34.dow.cam.ac.uk, which is
offsite and backs up clust and mek-quake.
|
