Definition of the BIP configuration file format

 

You do not need to read or understand this section if the general requirements of the last section are fulfilled, but in case you have any problem, it could be useful to check the contents of the configuration file that have been automatically generated.

The configuration of a BIP application relies on a configuration file that must be read by every process of the application, this file is by default $HOME/.bip/bipconf, but can be overridden by the environment variable BIPCONF.

You may write manually the configuration file for some peculiar configuration, for instance using a unit different of 0 or if you have problems using the scripts biproute, bipload, and bipconf. The only strong restrictions of BIP is that two machines must be separated by at most 4 switches, and only one board can be used by BIP on each machine (this last restriction could be removed, but it has not been done yet as it seems unlikely to be really useful).

The configuration file is organized by lines:

• the first line contains the number of nodes n for the BIP application.
• Then for each node there is:
  • one line giving the hostname of the machine that will be this node and the unit of the Myrinet board to use for communication,
  • and n lines which are the routes in Myrinet topology to reach the other nodes, these routes consist in 1 to 4 numbers depending on the number of switches to cross. As a special case a line with a single 256 means no switch (and is also used to fill the route to itself).

So in summary a configuration file will contain (1+n*1+n) lines. Moreover all blank lines and lines starting with # are ignored and can be used for comments. See file bip/examples/bipconf.ex to see an example of a configuration file, file bip/examples/bipconf.noswitch is an example when using two nodes linked directly without any switch.

At BIP initialization (in fact in bip_init), a process first gets its BIP logical number by searching the name of the machine (as given by uname -n) in the configuration file, reads the following routing information, and initialize the bip_mynode and bip_numnodes variables. And then will execute a small synchronization algorithm to synchronize with the other nodes, so that the processes can be started in any order and with any delay between them.