Version 10 (modified by 16 years ago) (diff) | ,
---|
Work-handling daemons
A BOINC project includes of a set of daemons for generating and handling work. Each program should be listed as a daemon in the config.xml file. Most daemons have the command-line options:
- -d N
- Sets the verbosity level. 1 = critical messages only, 2 = normal messages, 3 = detailed debugging info.
- -one_pass
- Process all available items, then quit.
- -mod n i
- Handle only workunits for which mod(id, n) = i. This lets you run the daemon on arbitrarily many machines. (available for feeder, transitioner, validator).
- -sleep_interval N
- How long to sleep when there's nothing to do.
Work generator
There is one work generator per application. It creates workunits and the corresponding input files. It is application-specific, and uses BOINC library functions for registering the workunits in the database.
During testing, you can create a single workunit using create_work, then use the daemon program make_work to copy this workunit as needed to maintain a given supply of work.
Feeder
This program is supplied by BOINC and is application independent. It creates a shared-memory segment used to pass database records to CGI scheduler processes. This data includes applications, application versions, and 'work items' (an unsent result and its corresponding workunit). It has the following command-line options:
- -random_order
- Enumerate work items in order of increasing result.random
- -priority_order
- Enumerate work items in order of decreasing result.priority
- -priority_order_create_time
- Enumerate work items in order of decreasing result.priority, then increasing workunit.id
- -sleep_interval N
- Sleep N seconds if nothing to do
- -allapps
- Interleave work items from all applications. Weight applications according to the value of their 'weight' field; if all weights are zero, results are interleaved uniformly. Without this option, runnable results are enumerated in an indeterminate order, and there may be periods when only results from one application are available for sending.
- -purge_stale X
- remove work items from the shared memory segment that have been there for longer than x minutes but haven't been assigned
- -appids N1, N2, …
- Get work only for the given apps. Useful if your project uses both locality and job-cache scheduling.
If a user's project preferences include elements of the form <app_id>N</app_id>
then the scheduler will send the user work only from those applications.
Transitioner
This program is supplied by BOINC and is application independent. It handles state transitions of workunits and results. It generates initial results for workunits, and generates more results when timeouts or errors occur. Options:
- -sleep_interval N
- sleep N seconds if nothing to do (default 5)
- -one_pass
- do one pass, then exit
- -d N
- debug level N (1, 2, 3=most verbose)
- -mod n i
- process only WUs with (id mod n) == i
Validator
There is one validator per application. It compares redundant results and selects a canonical result representing the correct output, and a canonical credit granted to users and hosts that return the correct output. Depending on your application, you may be able to use a BOINC-supplied validator, or you may have to develop a customized validator.
Assimilator
There is one assimilator per application. It handles workunits that are 'completed': that is, that have a canonical result or for which an error condition has occurred. Handling a successfully completed result might involve record results in a database and perhaps generating more work. See more details.
File deletion
The application-independent program file_deleter deletes input and output files when they are no longer needed. Unlike the rest, this daemon isn't strictly necessary; the server will still work without it, but unneeded data will accumulate in download and upload directories.
Database purging
The application-independent program db_purge removes work-related database entries when they are no longer needed. This keeps your database at a constant size even when your project runs for a long time.