Fat Server vs Raspberry Pi for automated jobs

More and more I am using small machines (SoC) like Raspberry Pi to run automated workflows (like automated marketing workflows). They run independently, connected to each other and Internet with high-speed LAN or WiFi. I feel it much more comfortable than running everything on one big fat server.

SoC / Raspberry Pi

SoC’s (System on Chip) are easy to restart, update independently and has dedicated resources (CPU & RAM). RAM, meagre though it may be, is often enough to run single processes. Very importantly there are no Software conflict issues, applications are naturally separate. Each SoC consumes less than 10 watts. Also, I can leave them running for very long periods without thinking about cooling or face fan / SMPS issues. Periodic backing up or restart (never faced it) is very easy.

Docker / Virtual Machine

It can be argued that I can run such processes on Docker containers / Virtual Machines. I find them more complex to set up, secure, operate and manage. The workload from the processes are non-homogeneous and IO access often becomes a bottleneck. You can typically run less number of such machines compared to the theoretical potential. What I find most annoying is that a single errant or heavily used process often affects multiple processes, not to mention the overhead from each. Then there is the issue of being single point of failure.

Not a panacea for all

There would be processes which simply cannot fit in the constraint resource model of SoC’s (primarily memory).

You may need to run multiple processes. I prefer addressing it by using multiple SoC’s, communicating via LAN.

Summary

I have put forward few points I personally find important to pursue the strategy of using multiple SoC’s to run long running automated jobs. I am interested in knowing your views and any issues I may have missed, both for and against this approach.