Supercomputer or distributed services?

When you do something, day in, day out, you tend to stop seeing the little quirks, the complexities, the oddities. Till one day you work on something that looks the same, smells the same, but suddenly... it isn’t. It talked like a duck, it walked like a duck till  it bit your finger off like a crocodile.

Sometimes the differences between a HPC system and a large-scale distributed system aren’t really evident from an high level architecture diagram. They both have lots of boxes doing the same or similar things after all. 

And if the diagram you’re looking at happens to be a financial risk system… well, all bets all off. You see, financial risk systems are not only HPC (ok perhaps HTC, for the purists reading this) system. They also contain a distributed service based architecture. Fun huh. Yea I know. It’s what makes working with HPC in FSI a little unique.

There is normally a delineation between the two. The service-based architecture acts as the keeper that feeds the HPC beast.  It’s not always a clean separation. Ok, let’s be honest, its usually a bit muddy! But there is a separation. You don’t try and run your calculation workload on a supercomputer as though it were a host to a services based architecture.

Makes the service based architecture sound a bit like a parasite doesn’t it. Slowly invading the supercomputer. Eeew.

Getting back to my original point though, when you see these architectures regularly you kind of assume that’s what you’re looking at when someone draws one you’ve not seen before. I mean they look the same, lots of boxes talking to each other. Till you start digging a little deeper. Suddenly something doesn’t feel quite right. And then you see it. Those few small, but critically important differences that push something from being a good HPC workload to being a Netflix or Google problem instead.

What happens when that separation is too blurred. Or just isn’t there at all? Do you still have a supercomputer? Answers on a postcard.