Requesting Cluster Computing framework for High Performance

137 Views Asked by At

I would like to build a cluster for getting high performance and to be used in a case of volunteer computing. I'm finding some volunteer computing platforms such as BOINC L, Beowulf cluster. I am a little bit familiar with Apache Hadoop, Spark cluster. Some of these cluster platforms would have to use MPI and other parallel processing libraries. However, I don't have any experiences with parallel prigramming. Moreover, I couldn't afford to write parallel programs for the cluster because I would like to make network simulations, image processing, programming testing etc. on my cluster. Therefore, I am finding a cluster framework that can run any applications distributed among the nodes. I found that openMosix cluster framework will be suitable with my requirements but I found that it stopped its cluster development. So, I would like to request respectfully some suggestions from the great seniors to direct me which kind of cluster framework is suitable to improve processing power in case of executing any applications. Thank you so much.

1

There are 1 best solutions below

0
Rytis On

(Disclosure: I am one of the BOINC developers, I also administer several BOINC projects)

As long as your applications are parallelizable (e.g. you can split the dataset or modify the application's command line parameters to split the search space), multiple simple options are available for you. If your nodes run Linux and you can SSH into them, simple tools like GNU Parallel might be suitable.

BOINC in general is tailored more to volunteer computing, aggregating heterogeneous resources. It has a bit of a learning curve, and focuses on security a lot (for example, all code running on the platform has to be signed, thus complicating deployment of "any" application). If you are looking at a real cluster-style use, where you will be running dozens of different apps, BOINC might not be for you (note that there exist ways to submit application binaries as part of work, but it is not trivial).

If you do decide to try BOINC, there is a somewhat old Virtual machine with BOINC server preinstalled, or just get a latest release of the code from Github.