Posts Tagged Supercomputing


Over the past few years, I’ve learned a great set of new words from my UK counterparts.  Many unfamiliar terms used in the UK have a great history, and sometimes they’re just… so perfect for what we do as system administrators.

Bespoke is such a term. It has many meanings, essentially being “fully custom” or “hand built”, or “hand made”. But is also has deeper meanings, alluding to “exactly fitting your personal needs”, “crafted”, or “personal touch”.  It comes of course, from the tailoring world, where it evokes a sense of old world craftsmanship, a very personal garment, hand measured and hand sewn, just for you.

Bespoke is at one end of the scalability spectrum.  A very few garments, made exactly for a single person.  Expensive and not very many will be made from any single pattern.  Like hand-built systems.

As you move to larger scale production you see much lower costs, but many fewer options and features.  Exactly like the IT world.

Bespoke is fully custom, exactly what you desire, no matter the cost.  Options and extras.  “High touch” support.  At the other end of the spectrum you have high scale, mass production: what you need, no more, no less, but at a more reasonable price.  It is a compromise solution to meet your most important requirements, but one you’ve decided to accept, usually for faster delivery or lower cost.

There is a time and place for bespoke, but increasingly we need to achieve high scalability, as we are increasingly pushed to do more with less.

There are some problems that just can’t be solved at the low end of the scale solution, or aren’t cost effective, or aren’t widely available.  For example, High Performance Computing (HPC) clusters of commodity computers are a high-scale alternative.  Commodity clusters brought supercomputing the masses, or at least to most research group, or smaller companies.  These high-scale solutions are compromises, but good enough and more widely available and affordable.

These cluster are great alternatives to the bespoke supercomputers of the past, such as the early Cray machines.  Our need to solve larger and larger problems, such as the HPCC Grand Challenge problems, eventually required more horsepower than a single, hand-built, bespoke machine could deliver at any affordable cost.

Moving away from bespoke supercomputers allowed us to scale in two ways:  we are able to make very capable systems widely available (at a reasonable cost) and we can grow systems at the high end where cost is not as important, but we need ever-larger capabilities.

As we explore solutions we have to ask ourselves where we need to be on the “bespoke” scale.  Most can’t really afford bespoke, and truly most won’t need it.  Automation allows us to build high-scale systems that provide most (or at least enough) of the features of a bespoke solution, but at an affordable price.



, , ,

Leave a comment

%d bloggers like this: