Launcher

In this panel you select the number of trajectories that you would like to generate and the number of cores to use. On startup Cain detects the number of cores in your computer and sets the parameter to this value. You may reduce the number of utilized cores if you want to reserve more processing power for other applications. For example, this may be a good idea if you want to watch a video while running your simulation.

Launching simulations.
There are two ways to launch simulations; you can use either the launch button   or the compile and launch button . Depending on the method you have selected, the launch button   will either launch the simulation using a C++ solver or a Python solver that can handle events. Consider the methods that use C++ solvers. If your model uses only mass-action kinetics, then a pre-built mass-action solver will be launched. Otherwise, a custom solver will be compiled and then launched. One may force compilation of a custom solver by clicking the compile-and-launch button. Compilation typically takes a few seconds. If you entered any propensity functions which are not proper C++ expressions, you will be notified of the compilation errors. Note that you can set the compilation options through the preferences button   in the main tool bar. The custom solvers are typically a little faster than the pre-built mass-action solvers, but the difference is usually not significant. However, if you are generating many trajectories it may be worthwhile to compile a custom solver. The mass-action solvers use a function that can evaluate kinetic laws with any stoichiometry. Evaluating this function is not as fast as evaluating a specific propensity function.

Stopping simulations.
When a simulation is running, the fraction of trajectories that have been generated is shown in the progress bar. You can abort a running simulation with . This will wait for each processes to finish generating its current trajectory and then exit. The trajectories that have been generated up to that point will be stored. You can also kill a simulation with . This will kill the solver processes and store the partial results if possible. Note that you can repeatedly launch suites of simulations to accumulate more trajectories. You don't have to calculate them all in a single run.

Running from the command line.
You can run simulations from the command line if you want. This may be useful if you want to use several computers to generate the output. (See the Command Line Solvers section.) First export a solver with . You have the option of exporting a custom solver or a generic mass-action solver. A custom solver is specific to the selected model. A generic solver may be used with any model that has mass-action kinetics. Next export ascii input files with the export jobs button . This will write an input file for each process; the trajectories will be split between the processes. Each file contains a description of the selected model and method as well as the number of trajectories. Suppose you export a solver to solver.exe. Then you enter 1000 trajectories and 4 processes in the launcher window and export the job with a base name of batch. This will create the solver inputs: batch_0.txt, batch_1.txt, batch_2.txt, and batch_3.txt. You can generate the trajectories for the first batch with the command:

./solver.exe <batch_0.txt >trajectories_0.txt
You can import the simulation results with the import trajectories button . (Make sure that you have selected the correct model and method before doing so.) See the File Formats section for specifications of the input and output file formats.

Using Mathematica.
If you select the "Mathematica" method in the methods editor, the export jobs button changes to the export to Mathematica button . The Mathematica notebook defines the ODE's that describe the reactions and species populations as well as commands for numerically solving the set of ODE's and plotting the results. The final section in the notebook has commands for saving times series data from the solution in a text file. You can import this in Cain with the import trajectories button .

Importing statistics.
If you have selected either the "Statistics, Transient Behavior" or the "Statistics, Steady State" output category in the method editor, then all of the buttons in the launcher, except for the import button , will be disabled. This is because there are no Cain solvers for these output categories. They serve as placeholders for externally generated solutions. One may import the solutions in a text file. See the Method Editor section for details.

Now we will consider two advanced topics: task granularity and job priority. If you are an advanced user conducting computationally expensive simulations, then these topics may be of interest. Otherwise, just ignore the two sliders in the launcher panel and skip the rest of this section.

Task granularity.
The first slider in this panel lets you select the task granularity. Suppose that you want to generate 20,000 trajectories using 2 processing cores. Cain will launch two solvers and then send them tasks (requests to generate a certain number of trajectories) until all of the trajectories have been generated. Note that there are many ways of dividing up the 20,000 trajectories. Of course the simplest approach is to request that each core generate 10,000. However, this may not give the best performance because one core may be busy with other tasks. Also, some trajectories may take longer to simulate than others. Dynamic task scheduling often improves performance. Here one breaks the problem of generating all of the trajectories into a number of tasks. Cain hands out tasks and waits for the solvers to finish with them. When a solver finishes, it receives another. This continues until the job is complete. So how many trajectories should comprise a task? If generating a single trajectory is computationally expensive, then trajectories should be generated one at a time. This minimizes the time at the end of the job when one core has finished with all of its tasks, but the other core is still working. Unfortunately this strategy does not work well when generating a single trajectory is cheap. In this case the overhead of handing out tasks could be more expensive than generating the trajectories. Now it might be best for a task to be 100 or 1,000 trajectories.

The size of the tasks (relative to the whole job size) is the task granularity. Small tasks are fine grained; large tasks are coarse grained. If one puts the granularity slider to the left, the trajectories will be generated one at a time. If one puts it to the right, the trajectories will be divided amongst the available cores and generated in a single task. The default position of the slider is in the middle. Here the task size is the square root of the number of trajectories divided by the number of cores. For our example above, trajectories would be generated 100 at a time. The default granularity is fine for most scenarios. However, if you are generating many trajectories and each one is expensive, it is probably best to put the slider to the left.

Job priority.
The slider at the bottom of the launcher panel allows you to set the priority of the solvers. (Mac OS X and Linux users may be familiar with the nice program, which allows one to set the priority of a process.) By default, the solvers are launched with the lowest possible priority. This way your computer will remain responsive. You can continue to work with Cain, check your email, or surf the web. If your computer is not busy with other tasks, launching with a low priority has a negligible effect on the running time of the simulations.