Petter Reinholdtsen <pere@td.org.uit.no>, 2000-04-17
Administrating the CIIPS Network - the easy way
Some suggestions to make it easier to maintain and
improve the CIIPS network.
To maintain a multi-user network it is vital to reduse the possibility
for human errors as much as possible.  It is also necessary to make
sure any error get as little impact as possible.
To achieve this, I believe the following must be done:
 - Keep things consistent
- Make sure knowledge is generic and not related to the different
 computers.  Make sure every supported program is available on all
 platforms and operating systems, with the same version and the same
 configuration. (cfengine, Store)
- Automate as much as possible.
- By leaving the repetitive tasks to the computer, one assures they
 are done the same way everywhere, every time. (cfengine)
- Make it possible to back out changes.
- When problems arise, make sure it is easy to get back to a
 earlier state when things where working.  Use version control systems
 to keep track of all human-edited configuration files, and make sure
 it is easy to remove a newly upgraded software package if problems
 are discovered. (CVS, Store)
- Detect problems as early as possible.
- If potential problems are detected and fixed before they become
 problems, less work is required to fix the resulting domino-effect
 when one system fail.  (Example: a full disk can stop the backup
 system from working properly, or a failing NFS server can hang
 processes on other servers and eventually bring other servers to
 their knees).
 
 Make sure to monitor all hosts and services, and warn when
 something is about to go wrong.  Keep statistics to find repeating
 problems. (Palantir, mon)
- Keep the users informed
- Make sure the users know where to find information on current
 problems, and the status of their reported problems.  This is easiest
 solved using a database and web based problem tracking
 system. (Bugzilla)
- Prepare for disaster
- Some suggestions:
 
  - Hard-disks fails after a few years.  Keep only new HDs in the
  servers, and reuse the old server HDs in workstations until they
  fail.
  
- Back up everything, and back up often.  Make it easy to get
  files back from backup, and let individual users get their files
  back on their own.  It saves support personnel the hassle, and make
  it easier for the users to recover from their personal disasters.
  
- Make it easy to replace workstations.  A HD mirroring system to
  make it trivial to reinstall a workstation and easy to replace a
  broken computer. (Autosetup)
 
 
Tools
I've already installed Store in the robotics lab, and it will take ~15
minutes to set it up on other hosts as well.  With proper setup, this
will make most hosts more independent from the NFS servers.