Diversity in HPC Won’t Improve Until We Start Counting


I have chosen to work in HPC because the work we do makes the world a better place. As SC has illustrated effectively over the past two years, #hpcmatters.

Given the impact of HPC on the world we share, it is important that we provide the very best tools possible for our users. And the odds are very good that we aren’t doing that today.

To first approximation the world is half female and half male; within those broadest categories, we can then subdivide into a bewildering array of nationalities, ethnicities, preferences, identities, and social backgrounds. HPC is arguably a sub-category of computer science, and the literature today tells us that CS and related fields are overwhelmingly male-dominated (the National Girls Collaborative Project has excellent pointers to these studies if you want to dig in at http://www.ngcproject.org/statistics). If you assume that HPC participation rates by women match the participation rates of women in CS (roughly 18% at the college level), then we are selecting from a pool of ideas that is far too limited by employing a pool of workers that is far too homogeneous.

It will seem intuitively obvious that we increase our odds of getting the better solutions if we increase the number of good ideas available to choose from, but there is more than intuition here. A study summarized in Harvard Business Review – and others like it, some of which are highlighted in a recent New York Times article – demonstrate that teams with more women produce better results (translating these findings into the construction of more effective teams, and understanding the degree to which other dimensions of diversity are a factor, are areas of ongoing research).

We can easily stipulate that the HPC workforce is not diverse. But most of the evidence available to us is anecdotal – I work with some women, but far less than half of my colleagues are women. I identify as a majority ethnicity male in my home country, and most of my colleagues do the same.

But this isn’t really data, is it? To my knowledge, there is not a comprehensive set of data depicting the demographics of our community, even by large conferences such as SC. If we are going to try to create more diverse teams we have to experiment, and we must be able to measure the extent to which a specific action produces a change.

And this is where we run into an actual roadblock.

Many of the diversity markers (gender, ethnicity, etc.) that we wish to measure and improve are markers against which significant bias has already occurred. There is data in both the lived experiences and in the social sciences showing that STEM professionals who are not majority ethnicity males (in the U.S. this would be “white males”, but whites are not the majority in all countries) experience differences in everything from starting salaries and rates of promotion to grant awards and publication rates – and different usually means lower.

There is a strong and entirely rational motivation for individuals from these groups to not draw attention to their difference from the majority in order to preserve their careers.

So what are we to do as a community? It is irrational, given limited resources to fund HPC research and provide services, to expect managers to fund diversity programs in the absence of data on the mere hope that diversity will improve.  Indeed, doing so would be an irresponsible use of funds. And it is likewise irrational to expect individuals to put their own careers at risk in the hope that doing so will yield some positive result to the HPC community in aggregate.

But someone has to take a first step. I believe that HPC organizations today have both the data and the responsibility to act first, and the good news is that it isn’t even hard.

I propose that every HPC center, company, and research organization create a diversity page, modeled on Google’s example at www.google.com/diversity. You already know who works for you and where they come from; just start counting. And by providing the data in aggregate you are protecting the individuals behind the data from a public identification that they may not yet be ready to make.

I have enough experience managing large programs to know this is easier to say than to actually get done. You’ll have to have your legal review the data, your HR person will freak out, your PR person won’t like you, and you’ll have to convince your boss or your board that it’s a good idea. But it IS a good idea, because without these data none of those in a position to fund programs that might broaden our workforce have the data to justify doing so. And creating a more diverse workplace is one of the very few things we can do that we know will produce better answers to the challenges we face.

I’ll go first:  I am volunteering as the general chair of SC16, and I pledge that when my conference year officially starts at the conclusion of SC15 in November, my website will launch with diversity data on the SC volunteer community. If you’re reading this in the future, go check it out at sc16.supercomputing.org/diversity, and ask yourself why you don’t already have a page like this.

Better yet, ask your boss.


John West is Director of Strategic Initiatives at the Texas Advanced Computing Center, part of the University of Texas at Austin. TACC provides HPC services and expertise to the open science community. Prior to joining TACC, John was director of a large supercomputing program for the federal government.

Current rating: 3.2