Frustrated with the state of the world I’ve decided to last out at the RHSHADOWBOX project. Don’t get me wrong there is a certain amount of entertainment value from spending $5000 on a compute cluster and putting it in a fancy case and then raking it around to the various club meetings and bragging about haw cool it is.
Certainly if you are in the business of providing or consumer of dynamic compute resources then these appliances are interesting. They clearly give you a chance to practice and improve the aspects of your three legged economic stool. But for the average consumer or business idle capacity is either lost revenue or additional expenses.
One interesting note in the shadowbox project page was:
The biggest TODO right now is replacing the bad node. The buildout as it currently stands makes replacing a node problematic. You have to remove the entire sub-assembly of server nodes and partially disassemble it. We have a better design in mind. We just haven’t had a chance to implement it yet. This is the current blocker.
One thing I find interesting is that this project uses new hardware. I’m certain thet RedHat has plenty of laptops that could be recycled. Next, the demo was computing pi rather than demonstrating some actual function. (see economics) And frankly they built something as complex as the space shuttle when a weather baloon would have been sufficient.
I think about the weight of that case and the burden of carrying it around when clearly a raspberry pi, while not as intense as the Intel i3, would easily demonstrate the principles. For that matter there is something to be said for the sopine cluster… if you just want to compute pi.
The project had one other complaint about hardware failure. The case construction hindered their ability to replace damaged boards. This is where I get particularly snarky. Their project was not designed to be pluggable or resiliant. When you look at the software stack they are giving up a lot of time, money and effort just to keep the cluster running. The infrastructure has a cost. Building with OpenStack is like using cinder blocks when lego would be sufficient. Partly because the tools we have today work and have already passed the test of time. And while I’m a fan of the new stuff there is yet another cost if you make the wrong choices.
I’m staring at my workshop and I just cannot find a single use for a cluster.
Correction… I cannot find a single use for a COMPLEX cluster. (oh docker swarm my docker swarm)