Ask HN: Cluster for side projects

[gnikflow]

I've been doing a lot of distributed computing work over the past year, so nowadays when I get home my side project thoughts look like this:

- I want my own mesos cluster. - I wonder if I should use apache aurora or chronos for scheduled jobs in our next release? I want to install both and take a look. - I'd love to check out spark streaming off of a kafka topic. - etc.

What's a cost effective way to have a personal cluster (big enough to really apply distributed techniques in a meaningful way, maybe ~10 nodes)? The most obvious solution is to use digitalocean or aws and only spin up instances when I use it. But I'd actually rather spend more money upfront ($1000-3000) to just have an always on cluster that I can keep going and not think about spinning up/down. Could I just buy a few servers with alot of ram/cpu and then put VM's on them, effectively giving me more instances to use? I could also go the google route, which is buy more commodity hardware at a bargain instead of fewer higher powered servers.

I'm rambling at this point - any input would be appreciated!