E
E
Evgeny Dolbanovsky2018-09-30 14:55:37
Debian
Evgeny Dolbanovsky, 2018-09-30 14:55:37

How to create a cluster of multiple Raspberry Pis?

So, good people, I googled on this topic, but I didn’t understand this:
Is it possible to combine several single-board computers into a cluster with power pooling to complete the task?
That is, one system on the first board takes the resources of another board or sends a task to it, part of the task, in short, create a single system based on 4 computers with automatic distribution of resources among the boards to complete the task.
Well, or how does this cluster work in general, I can’t understand the principle, parallel computing or in one thread?
Throw on the topic links to articles, manuals, videos.
I want to build a stand

Answer the question

In order to leave comments, you need to log in

3 answer(s)
S
Sanes, 2018-09-30
@Sanes

Can be distributed. This is the task of the application.

N
Nazar Mokrinsky, 2018-09-30
@nazarpc

Assembling a cluster from a Raspberry Pi only makes sense for fun. For performance, it is more advisable to take a Xeon of some kind not of the newest generation.
How to process data in a cluster depends very much on the application, and there is no general effective solution that will allow you to make a virtual machine with at least 5x performance compared to a single system with a dozen Raspberry Pis.
The principle is the same as you described: something takes data from somewhere, processes it, and sends the result somewhere. It is impossible to say more precisely from such a general question. For example, mining using a pool is one example of how a bunch of computers can work on a common task.
Professional data centers have fast networks, direct memory access and a bunch of other tricks that are not even approximately available on the Raspberry Pi. I think that for training it would make more sense to raise virtual machines on the desktop and set up interaction between them over a virtual network.

R
Roman Sokolov, 2018-09-30
@jimquery

Perhaps the most effective method in this case would be to launch separate instances on each piece of hardware that will process data from one network storage (database)

Didn't find what you were looking for?

Ask your question

Ask a Question

731 491 924 answers to any question