Misadventures in HPC Cloud Migration #4

Before I start pointing out all the things that were missing from my last picture of what HPC looks like on premises (in #FinancialServices but may apply more widely), let’s start by correcting a big misconception that was on the picture first.

I know you’ve spent the last decade in various long term strategic programmes to rationalise, reduce, normalise, (call it what you will) the number of #risksystems you have. I know you think you’ve got this down to a number that you can count on one hand. If that’s the number you think you’re migrating to cloud though, you’re not even close.

You may well have reduced the number of codebases for your risk systems to a number that is only 2 digits. Each of those codebases has enough configuration to create a Christmas special sized box of Quality Street in variations of risk system. Want to know what gave it away? The number of distinct HPC “apps” presented to the scheduler.

So, being generous, you’re probably thinking you have a low 2 digit number of applications to migrate. You don’t. And let’s be honest, I know some out there thought that the “grid” is just the one application to move.

A large systemically important bank (the “too big to fail” crowd) will have a number that is probably three digits long. Four if you’re unlucky or still trading in a large range of derivatives.

Use that as your starting assumption and you might have something more than a mince pie’s chance of survival at a British office Christmas party.