Is there any benefit to having your own clustered system vs just using AWS or other cloud services?
For example, a friend of mine had four 4 core, 8GB machines and tried to configure a hortonworks cluster, but the system spent so much time just running the cluster state, that you could only do trivial computing on top of the cluster before the nodes started timing out.
He then spent ages disabling unnecessary services, re configuring components to be more memory efficient - and spent so long doing all of this maintenance he never actually learnt how to compute on top of the cluster, which was his original goal.
So it's best just to pay a few hundred a month to amazon and run your own small cluster that you can spin up and down as you need
A clustered system could be a load balancer and a few servers running percona cluster, or a couple of ldap masters with read-only relicas and proxy-writes.
Doing it on bare metal also gives exposure to the parts amazon specifically don’t
For example our deep learning clusters do not use hadoop, but use multi-process python daemons (written by me) and communicate with protocol buffers. Writing these high performance daemons myself has taught me a lot about cluster computing and some of the tradeoffs that have to be made in order to keep the GPU pipelines full.
For anyone else reading: Pyro is quite fun if you're in a limited environment and want to learn https://pythonhosted.org/Pyro4/ Also, setting up a spark https://spark.apache.org/ environment can be done on smaller machines too
The vast majority of software developers will interact with, if it’s running clustered it’s because the project needs higher availability for that service (typically the database) and potentially higher performance in production.