Asking this question about when it makes economic sense to build one's cluster in the cloud has led me to some interesting thoughts. Everyone seems to have a slightly different idea of what a reasonable cost per cluster CPU core is, everyone has a slightly different idea of what constitutes appropriate hardware for a given task, etc.

This has made me realize how useful it would be to get a list of the hardware everyone on the site is using, and, broadly, what they're using it for. What would also be really useful would be a rough estimate of the cost per CPU core. People could give make/model of the CPU, quantity of RAM, whether the cluster is in-house or off-site, etc.

I know that this isn't a hardware site, but that's what would make this so great. Although we all use serious computing machinery every day for our work, none of us are truly hardware people. This kind of task-specific hardware guide would be insanely useful. I for one would have loved to have had a guide like this during my last round of cluster node acquisition.

I realize that lists are the devil around here, but is there any place on the site where such a community hardware guide could live?


2 Answers 2


I don't think having a cluster hardware guide is necessarily a good idea for a question, because it does invite a list. If the list is small, curating it is not a big deal, but if it's big, it's problematic.

I'm beginning to think we should have some sort of wiki-type resource associated with the site where such information should be posted.

Instead of asking for a list of hardware, I suggest you ask questions like:

  • How do I build a cluster for (insert application or target algorithms here)?
  • How do I go about building and administering a cluster for computational science research?
  • Given a limited budget for a cluster, how should I allocate it to get maximum performance?

(I'm not an expert on clusters, so anyone who is, feel free to chime in with suggestions for better questions.)

The idea is to ask questions that will elicit the information you want without having it devolve into a list. Having a target application or algorithm in mind, and making the computational science influence apparent is really helpful also.

  • $\begingroup$ I think the "wiki-type resource associated with the site" is a pretty good idea. $\endgroup$ Jun 2, 2012 at 13:36

Your proposed question would not be a good match for SE. There are just so many ways to build a cluster that there is "no answer". Such question would go the same way of a 'which programming language' discussion, a 'which OS' discussion, or a 'which compiler'.

A cluster is more than just a list of components. The success of the facility does not only depend on the quality of the components but also in the experience and level of involvement of the support team.

What really makes a difference when building and maintaining a cluster is the experience of the builder/maintainer with the software and hardware that the cluster will use.

I think that programmer experience is what makes the difference in the 'which X' type of questions. The answers are certainly interesting but not very enlightening.

If you are interested in building clusters I recommend for you to get experience: Start small and slowly go larger. You will see that each time you increase the size of the cluster by an order of magnitude the most pressing issues of the cluster will change.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .