How to Switch Networks
For users of the ATOMIC LAN in Div7
J. Touch / July 1, 1996

This document addresses how users of the ATOMIC LAN in div7 can address and correct disturbances regarding failures of the ATOMIC LAN.

Classes

There are currently five classes of hosts in the division.

REGULAR
hosts with one address on either (but not both) of:
- zephyr-net (128.9.160.x)
- darkstar-net (128.9.128.x)
The entire floor originally consisted of only REGULAR hosts. Four additional types of host configurations have been added:
DUAL
hosts with two addresses,
- primary address on either (but not both) of:
  - zephyr-net (128.9.160.x)
  - darkstar-net (128.9.128.x)
- secondary address on:
  - atomic-net (128.9.112.x)
ATOMIC-ONLY
hosts with one address on:
- atomic-net (128.9.112.x)
These are limited to GRA hosts in offices where the entire ethernet could not be relocated to atomic-s-net
DYNAMIC
hosts with two addresses,
- primary address on:
  - atomic-s-net (128.9.192.x)
- secondary address on:
  - atomic-net (128.9.112.x)
FREELOADER
hosts with one address on:
- atomic-s-net (128.9.192.x)
NB: there are a small number of these non-ATOMIC hosts, usually laptops, PCs, or Macs located in offices with DYNAMIC hosts.

To discover what kind of host you have, use the script:

/usr/atomic/hosttype

This script works only on Unix-style workstations.

                            +----  rest of ISI / Internet
                            |
                            |     
                        +-------+ 
                        | cisco | 
                        +-------+               
                          /   \                 
          darkstar-net   /     \   zephyr-net  
          128.9.128.x   /       \  128.9.160.x 
                       /         \  
            REGULAR---/        ___\---REGULAR 
                     /        /    \ 
                    /__    DUAL  +-----+ 
                   /   \    |    | IDN |            
                      DUAL  |    +-----+            
                       |     \_____/  \             
                        \_________/    \            
                                 /      \---FREELOADER
                  ATOMIC-ONLY---/        \ 
                               /          \   
                              /--DYNAMIC---\
                             /              \
                       atomic-net     atomic-s-net
                      128.9.112.x     128.9.192.x

Failure Modes

There are two basic failure modes for the ATOMIC LAN; either the LAN can fail, or the isolating router (IDN) can fail. The following table describes the failure of each type of host, given each type of failure:

	may require IDN to get to:	may require ATOMIC-LAN to get to:
REGULAR	ATOMIC-ONLY DYNAMIC FREELOADER	ATOMIC-ONLY
DUAL	ATOMIC-ONLY* DYNAMIC* FREELOADER	ATOMIC-ONLY
ATOMIC-ONLY	DUAL DYNAMIC* FREELOADER	anywhere
DYNAMIC	anywhere	ATOMIC-ONLY
FREELOADER	anywhere	ATOMIC-ONLY

(* = only if ATOMIC LAN becomes partitioned)

How to Get Home

In the rare case that a new class of host (DUAL, ATOMIC-ONLY, DYNAMIC, FREELOADER) fails due to a fault in the ATOMIC LAN system (either the LAN or the isolating router, IDN), users may want to reconfigure their host to be usable until the fault can be corrected.

Users can contact Joe or Ted for help with this procedure. The following documents the procedures they follow, for use by ACTION or users familiar with system management.

This should be necessary only for ATOMIC-ONLY hosts. IDN should be comparatively reliable.

REGULAR:
- no reconfiguration is required
DUAL: (replace host-a as appropriate, e.g., mex-a)
- shutdown the ATOMIC interface,
- remove routes to the ATOMIC LAN, both direct and via redirect
  - ifconfig myri0 down
  - route delete net atomic-net (host-a)
  - route -f add default `cat /etc/defaultrouter` 1
ATOMIC-ONLY: (move back to zephyr-net, darkstar-net as appropriate)
- rewire office ethernet from atomic-s-net to zephyr/darkstar
  REBOOT AND HALT IN SINGLE-USER, DISCONNECTED MODE
- change hostname.le0
  - from host.isi.edu
  - to host-e.isi.edu
- change default router
  - to /usr/atomic/myri_configure IPATOM3 cisco5-160
  - or /usr/atomic/myri_configure IPATOM3 cisco5-128
DYNAMIC: (move back to zephyr-net, darkstar-net as appropriate)
- rewire office ethernet from atomic-s-net to zephyr/darkstar
  REBOOT AND HALT IN SINGLE-USER, DISCONNECTED MODE
- change hostname.le0
  - from host.isi.edu
  - to host-e.isi.edu
- change default router
  - to /usr/atomic/myri_configure IPATOM3 cisco5-160
  - or /usr/atomic/myri_configure IPATOM3 cisco5-128
FREELOADER: (move back to zephyr-net, darkstar-net as appropriate)
- rewire office ethernet from atomic-s-net to zephyr/darkstar
  REBOOT AND HALT IN SINGLE-USER, DISCONNECTED MODE
- change host
  - from host to host-e
- change IP address
  - from 128.9.192.X to 128.9.160.X or 128.9.128.X
- change default router
  - from idn-s (128.9.192.72) to cisco2-160 (128.9.160.5) or cisco5-128 (128.9.128.5)

Reboot & Halt in Single-User, Disconnected Mode

Rebooting a host for single-user, disconnected mode requires some subtlety. Hosts configured for network-mounted file systems will hang if the network is not available, preventing the reboot from completing. The hang cannot be interrupted without restarting the reboot (L1-A works, ^C does not).

For hosts with off-the-shelf configurations, the solution is to interrupt the reboot during the file systems check of the local disk. The user needs to hit ^C during the "fsck", IMMEDIATELY AFTER SEEING:

"checking filesystems"

The next two operations, mounting / and /usr, can be interrupted too. However, once "umount -at nfs" occurs, the system will hang if the network is unavailable, requiring a complete reboot.

This procedure has been simplified on ATOMIC hosts, which have a 2-second delay after fsck'ing. The ^C can be hit within that time, after seeing the message:

"The pause that refreshes."

Then the local file system needs to be mounted:

mount -o remount /
mount -o remount /usr

The local network can be configured as well, using

/sbin/route ...
/etc/ifconfig ...

Note that the /sbin/route command is required, because it is statically-linked. The /usr/etc/route is dynamically linked, and will hang, requiring network access to retrieve the dynamically-loaded libraries.

NOTE:

For hosts with off-the-shelf configurations, after the interrupt, the halted fsck operation must be completed:
fsck
(answer Y to any questions)

If successful, this will return to the single-user prompt. If not, the system will automatically reboot.

Go back to the ATOMIC-2 home page. / Go back to the ISI home page. ISI Logo

This page written and maintained by the ATOMIC-2 group.
Please mail us any problems with or comments about this page.
Last modified July 18, 1996.