Inspired by projects like folding@home and SETI@home, this is a framework on top of Rails that allows the server to distribute tasks to its clients (browsers). This is best suited for highly parallelizeable tasks such as monte-carlo simulations or other embarassingly parallel problems.
- Provide tasks.js files in app/tasks
- task*.js must define
var nextTask = new Task(main_function_name)
- The result of
main_function_name
will be saved in the db
Already completed tasks will run again on a new machine and the results will be compared (about 10% of the time). This gives some additional confidence in the results, but is by no means a security guarantee.
Some tasks (e.g. random generations) are not well suited for
verification. For these tasks, the rake tasksjs:repeated
command
takes an additional argument to omit verification.
A Rakefile has been provided to make some common tasks easier. The
namespace is tasksjs
.
Note: For rake operations that add tasks to the database, rake will
also look for a parameters file parameters_task*.txt
in the same
location, and will use parameters on a single line. See more at
parameters
rake tasksjs:init
will clear all tasks from the database, then go through all the tasks in app/tasks and add them to the databaserake tasksjs:repeated
will take an argumentnum
and add all the tasks in app/tasksnum
times.- as noted above
rake tasksjs:repeated
will take an additional argument to disable verification. If you want verification to be disabled, you should pass "true".
The parameters_task*.txt
file is expected to have
all the needed comma separated parameters on one line. This will then
be applied to the function in task*.txt
, once for every line.
A parameters_task*.txt
like the following:
0, "james", "sally"
5, "john", "jerry"
will result in two tasks being added to the queue:
task*.js
with args0, "james", "sally"
task*.js
with args5, "john", "jerry"
An examples
folder has been provided in app/tasks
. These simple
examples can be moved into app/tasks
if you'd like to use them. Some
have parameters, others need to have verification disabled (monte
carlo).
- browsers execute arbitrary tasks and return results to browser
- server manages queue from db
- things are added to queue only by the server (clients adding tasks is a non-goal for now)