A Rational Approach to Raspberry Pi Clustering

Any good use cases?

So far maybe the best one is as a compile server farm? It could be useful when building large software projects for Pi, where multiple cores can be used. I wouldn't know how to set that up - in old times, it could be done using gcc using the -j switch but it probably needed some setup/configuration somewhere.

As for other use-cases, the article was clutching at straws it seemed. There are better use-cases. One would be as test environment, because sometimes people want to simulate a scenario with multiple devices. Once I wanted to simulate a small business. To do that I 'borrowed' 25 laptops.. Today, that could have been done with 25 Pi's.

Any good use cases?

So far maybe the best one is as a compile server farm? It could be useful when building large software projects for Pi, where multiple cores can be used. I wouldn't know how to set that up - in old times, it could be done using gcc using the -j switch but it probably needed some setup/configuration somewhere.

As for other use-cases, the article was clutching at straws it seemed. There are better use-cases. One would be as test environment, because sometimes people want to simulate a scenario with multiple devices. Once I wanted to simulate a small business. To do that I 'borrowed' 25 laptops.. Today, that could have been done with 25 Pi's.

Hello Shabaz,

I am preparing a blog post with several scenarios (but need some days as I should make it in two languages ) There are many cases where a cloud is useful and there are very few examples All explain the "right" solution, I have just afforded one based on some initial considerations as explained in my previous post. Then no serious examples are shown around. I just list some, maybe I can't provide all but some are really interesting.

Data collection

1. When an IoT or any other devices manages a large amount of data collected from the real world, maybe saved in a json file for simplicity (but with some more difficult also a csv or xml format will be good), with few lines or R code (you know R language, yes, I am sure ) you can plot very fast many results and crossing data with very low effort.
2. Another kind of data collection management is selecting the bare metal information and standardize them to be passed to some kind of machine learning process that should use the data for training.

Large batch image processing

1. Using some efficient image manipulation tool like imagemagick one of the most time-consuming processes for which a cluster is ideal support is when you should resize an entire set of images, for example, to prepare them from the high-resolution camera (18, 20 Mp or more) for web use or to make thumbnails.
2. Starting from a collection of time-lapse images to assemble a clip sequence

Signal management for testing

About this topic (suggestions are welcome), I have not a clear idea but I have the sensation that a Pi Cluster can do something that very few other kinds of clusters can do. Not only you can make parallel software processes (including large compilation) but you can also execute multiple processes that use as many GPIO as many operational nodes you have (where the master node does not participate to the pool, as you know).

Enrico