| | 1 | == Hints for Making New OS Images == |
| | 2 | First, 40% of our nodes are physically located 400 miles away from the boss and users servers. Please, '''never''' try to create an image from a node located in berkeley instead of one located at ISI. The ones located at berkeley and their types all begin with the letter ''b'', eg ''bpc183'' for a node, ''bpc2133'' for a type. |
| | 3 | |
| | 4 | Secondly, after you have created an image, try loading it back and watching what happens through the serial port. |
| | 5 | |
| | 6 | Consider creating a '''two''' node experiment, one to create the image and the other to load it back. |
| | 7 | |
| | 8 | There is a command called "os_load" available on the ''users'' server: |
| | 9 | {{{ |
| | 10 | users% which os_load |
| | 11 | /usr/testbed/bin/os_load |
| | 12 | users% os_load -h |
| | 13 | option -h not recognized |
| | 14 | os_load [options] node [node ...] |
| | 15 | os_load [options] -e pid,eid |
| | 16 | where: |
| | 17 | -i - Specify image name; otherwise load default image |
| | 18 | -p - Specify project for finding image name (-i) |
| | 19 | -s - Do *not* wait for nodes to finish reloading |
| | 20 | -m - Specify internal image id (instead of -i and -p) |
| | 21 | -r - Do *not* reboot nodes; do that yourself |
| | 22 | -e - Reboot all nodes in an experiment |
| | 23 | node - Node to reboot (pcXXX) |
| | 24 | }}} |
| | 25 | while the second node is reloading, watch its progress in real time using the console command from the ''users'' server, ie |
| | 26 | {{{ |
| | 27 | users% console pc193 |
| | 28 | }}} |
| | 29 | A third general suggestion is that if you think you've got a good image, but it flounders while coming up , you can create another experiment with an ns directive that says "Even if you think the node has not booted, let my experiment swap in anyway;" then you may be able to log in through the console and figure out what went wrong. An example of this is: |
| | 30 | {{{ |
| | 31 | tb-set-node-failure-action $nodeA "nonfatal" |
| | 32 | }}} |
| | 33 | A fourth suggestion is to create whole disk images on a smaller machine rather than a single partition image. |
