Multiple PCs to work around a single task

[KAIRU26]

I often use my PC for quite heavy task (conversions of large files, compression, animation...), and that during long periods (1 week for certain animation for example)

And so I asked whether there is a solution for multiple PCs connected together can be used to achieve the same task. I am indeed in my 3 PC equivalent of a power, which could, if I can gather, greatly reducing the computation time.

I think especially in a software as it should be able to work with various applications (blender, VLC...).

[Zachary]

What you're trying to do is called clustering.

Some software allows it, its even more common (3DS Max, to name him). But this is indeed the software itself known to, you can not distribute a calculation "that way". What you ask is just like if you wanted to use two PC "means" to turn a game super greedy: it will not work because the game is not designed to run on multiple machines simultaneously.

Look in the docs of your software, if they accept or not distributed computing (search "distributed computing" or "clustering" in the docs). If the software can do it, then it will be possible. Otherwise, it will not be.

[KAIRU26]

I imagined an application (independent software) installed on each machine to work together, which would be a server. The "intermediate" application requests to load the processor, which would send the application server, which in turn distributes the calculations on other machines.

Where other machines would send their results to the server, which manage to save results to a file ...

Ok maybe bombarding but I know myself enough in the area of communication between application processor and whether it is feasible or not.

[Zachary]

I imagined an application (independent software) installed on each machine to work together, which would be a server.

This is exactly what happens with applications designed for it: you installed on machines "slaves" a smaller version of your program (an option in the installation, often), and you bump on your machine "master" normally. Except that you have configured the application to know that it has N machines available for calculations.

The "intermediate" application requests to load the processor, which would send the application server, which in turn distributes the calculations on other machines.

Impossible: it would be a hook at the CPU, and it is not possible to do this, simply. Something working as low level simply could not distinguish whether it is your application running, or the OS or another application. And even in your application: it could be both refreshing code GUI that calculation, or even read requests to disk!

Where other machines would send their results to the server, which manage to save results to a file ...

Yes, it is indeed the principle: machines slaves are waiting, prepare the master machine (more or less) data and send pieces of calculations to do (often called chunks). This is usually done by TCP/IP protocol with a more or less specific to the application above.

Basically, the same principle that thread, except that unlike a thread that has access to the total memory process, then it is imperative to copy (and send) explicitly the initial data, and retrieve (and write/recombine) results. Of course, it is worthwhile only if the time transmission and copying is negligible compared to computation time, otherwise you lose performance instead of winning so we must be careful at the granularity of treatment.

Ok maybe bombarding but I know myself enough in the area of communication between application processor and whether it is feasible or not.

An application does not communicate with the CPU: the CPU, which executes, in addition to many other things. From this perspective, the CPU is perfectly neutral with regard to what is running, the only thing it can know is the level of protection/performance data (in very rough, it can hardly distinguish if the code is executed in kernel mode or user mode).

But look very carefully in the documentation of software you use, they may offer such a calculation.