The biggest human temptation is to settle for too little.

― Thomas Merton

Lately at work I find myself wondering, “is this really the best we can do?” “Why aren’t people interested in better methods?” “Are the current methods and models good enough?” The reasons for these questions are the relatively low-bar we set for achievement these days. The outcome of all this lack of ambition is astonishing levels of mediocrity coming from ridiculously large investments. Ultimately it reflects our choice to invest too much in hardware and not enough in intellectual capacity.

Some men are born mediocre, some men achieve mediocrity, and some men have mediocrity trust upon them.

― Joseph Heller

titan2The problem with the money spent on computational science is one of lack of balance and aggression. There is plenty of money; the problem is what that money is spent on. Too much focus has been made on hardware with a reliance on Moore’s law for progress. Hardware improvement has been a sure thing for half a century, and thus a low risk. Recent investments in scientific computing have largely forgotten the history of scientific computing. A lot of issues remain unaddressed by current work. This is a theme I’ve touched on before (https://wjrider.wordpress.com/2014/09/12/the-dangers-of-good-enough-thinking/, https://wjrider.wordpress.com/2013/12/26/what-makes-a-computational-simulation-good/, https://wjrider.wordpress.com/2014/03/20/legacy-code-is-terrible-in-more-ways-than-advertised/, https://wjrider.wordpress.com/2014/10/06/supercomputing-is-a-zombie/).

Compromise is a sign you’ll pass on the road to mediocrity.

― Tim Fargo

1280px-Columbia_Supercomputer_-_NASA_Advanced_Supercomputing_FacilityThere is a significant practical problem with maintaining the progress provided by Moore’s law. It has become very hard to do. This is because Moore’s law is already dead; at least its dead in any practical sense. For a tiny and relatively unimportant, impractical application we can still make it work, but at a ridiculous cost. For most of the things we really use computers to do, Moore’s law died about 10 years ago. To keep it alive we make computers that are incredibly hard to use, to build and maintain. They use too much power and cost way too much money. The resources going into this “fool’s errand” are starving all the work that actually makes the computers useful. This downward spiral needs to end. The commercial computing world has already divested from Moore’s law, and now focuses on software and communication capability for value. The hardware is improving, but modestly compared to the past.

Another critical issue is the complexity of the software. The software is much like an infrastructure that is crumbling. Like roads, we patch them, but rarely rebuild the software from the ground up. As a result we use the same basic software for years legacy-code-1beyond the date it should be rewritten. It becomes less useful and more expensive to maintain. These costs amplify over the long-term, but in the short-term, the patch and kick the can down the road approach is viable. The problem simply gets worse every year. We are unwilling to deal with rewriting the software because of the investment they represent. The problem is quite analogous to the Nation’s physical infrastructure problems.

Cray XE6 imageWe also have issues regarding methods and models. As code becomes more capable, it becomes harder to develop new codes because it is so expensive to measure up. The old codes can do marvelous things for practical problems. Testing new methods and models becomes nearly intractable. Ideas to make methods and models abstract with “components” have largely failed to provide a path forward. Part of the issue is the inability for component-based methods to solve “real” applications, which requires a lot of dirty work (robustness, efficiency, reality). As a result the older methods and models have become engrained. As this happens the community working on methods and models becomes estranged from computing. Additionally, the effort to put these older codes on new computers has become extremely difficult and expensive. This is compounded by the size of the code base. Together we have the recipe for disaster.

The highest level than can be reached by a mediocre but experienced mind is a talent for uncovering the weaknesses of those greater than itself.

― Georg Christoph Lichtenberg

This ends up in the loss of expertise at the level of depth needed for excellence as improving methods and models developed those experts, which isn’t happening. This provides all of the makings for a viscous cycle. Eventually we will hollow out our expertise leaving us with old methods, old models running on new computers tlegacy-codehat are outrageously expensive, but less useful every year. It simply cannot be sustained. The short-term thinking and the lack of tolerance for risk keep us from solving any of these problems. We end up of settling for mediocrity as a result.

The only sin is mediocrity.

― Martha Graham

I think that the idea of not having Moore’s law operating terrifies some people. They 23611-004-41D33A72won’t have the security of doing nothing and making progress. The problem is that they don’t recognize the exorbitant costs of propping up Moore’s law for the last decade, or the cost of what has been scarified. The terrible thing is that the costs and risks of the path we’ve taken are far higher. We are moving toward a collapse of astounding proportions. Instead of building a sustainable future we are building on the past while losing sight of how we actually got here. For decades the mathematics and physics were miles ahead of computers. During the late-70’s computers caught up and for fifteen or twenty years there was a glorious balance of computing hardware and intellectual capital. We have lost sight of what made all of this possible, and we are taking a huge risk moving forward.FortranCardPROJ039.agr

Caution is the path to mediocrity. Gliding, passionless mediocrity is all that most people think they can achieve.

― Frank Herbert

technicaldebtAt some level it all stems from fear of failing, which ironically, leads to actual failing, or at least success that is so modest that it seems indistiquishable from failure for success-minded folk. I simply don’t see an appetite for progress that can overwhelm the desire to never appear to fail. This outcome is assured by the belief that we can manage our way to success (and manage away failure), and the short-term focus for everything.

As long as I have a want, I have a reason for living. Satisfaction is death.

― George Bernard Shaw

 

Advertisements