Keeping the top job
Lawrence Livermore National Laboratory (LLNL) has a reputation second to none in high-performance computing. It builds and manages the two largest computers ever built and was one of the technical powerhouses that developed the internet. But running the computing side of a laboratory with such a reputation is not about sitting back on past laurels. Dona Crawford, as associate director for computation, probably has the top job in scientific computing management, and it is her job to keep it at the top.
Her combination of technical ability and people skills have made her a dynamic force in keeping the laboratory at the leading edge of scientific research. It has not lost touch with its core mission of advancing the technology of the most deadly weapons on earth. But countless swords have become ploughshares, and the weapons have served as an effective deterrent.
Steve Ashby, LLNL's deputy associate director for applications and research, says: 'Dona is an extraordinarily personable and likeable person. When you are in a work or social setting she has an energetic personality that stands out. She engages very readily and she makes other people feel very comfortable. She has good people skills and so she is not your stereotypical maths geek. When you are part of her management team you feel that you are working with her rather than for her. She is my boss, but she never makes me feel that way. That makes for a more effective management team.
'Before she came were had a very strong organisation, but I don't think we were pulling in the same direction all the time. What Dona has been able to do, through her management and leaderships skills, is get the team moving in a more common direction. Before she came, I think, each manager felt they had their piece of the world and wanted to do it their way; there was not much of a sense of team within the directorate. Dona has made it clear that each department has to be successful and the laboratory as a whole has to be successful. She is willing to rely on her team, she gives us a certain level of responsibility and we are accountable for that.
'She is not afraid to make difficult decisions. When she came in she surveyed the lie of the land and decided she needed to make changes. She did that quickly to get the team she wanted. She has told us that she expects us to be successful in our areas but that we are part of one laboratory. We have gone from having a variety of different business practices to having much more consistent practices. There was a lot of rumbling at the time but, in fact, she anticipated the direction the whole laboratory was going and, by acting in good time, we were in much better shape when the laboratory started going down a similar path about 18 months later.
'The other thing I have found most impressive is that she often argues the case for going in a particular direction and people wonder why, and then we later find we are out ahead of everyone else, and better able to respond to things than if she had not led us there.
'What she has tried to do is to engage more of the laboratory in working with the computational directorate. She has kept us at the forefront of high-performance computing, but at the same time she has put a lot of emphasis on leadership training and making sure that we are developing the next generation of leaders to call upon - which, historically, was not something that the laboratory has done well at.
'You will never silence the grumblers, but a fair observer would conclude that she has been right more times than she has been wrong.'
Crawford was born in Fort Wayne, Indiana. Although there was a farm in the family she lived in the suburbs, but played on the farm at weekends. She admits she was not a very sociable child and spent a lot of time sitting in a tree and reading. Her father worked in the commercial side of Shell Oil. Her parents broke up when was a child and she moved with her mother and two sisters to live near relatives in Long Beach, California, where she went to high school. Her mother found work as a secretary, but making a living was tough.
Crawford says: 'It was at that point in my life that I decided I would work really hard and never be in a position where I could not support myself and my family if the need arose.'
She was a straight-A student at school. Her hero was Madame Curie and maths was her favourite subject, so she wanted to study maths.
She says: 'My adviser told me that if I studied something like German as well then I could get a job as a teacher. German was the language you studied in the US at the time and I found it mathematical, so it was like chewing bubble gum - it was fun. Doing maths and German was fun to me.”
In 1969 she went to Redlands University, which was near where she lived and got through with the help of a scholarship and a series of jobs including as a waitress, work in an old people's home and canning fruit. She did well and stayed on to get a teaching credential. After graduation she was worried about how well she spoke German, so she spent time at Middlebury College in Vermont, where students immerse themselves in the language for three months and you are not allowed to speak or read English. She then went to Mainz in Germany for a year.
She returned to the US in 1975 with a masters degree in German and her teaching qualification, and she was ready to start teaching maths and German. The only problem was that, at that point, there was a glut of teachers and she could not find a job. So she started looking around for other things she could do.
She says: 'I thought, maths, German; these are great things. Lots of government jobs need these skills, so I applied to those things you could not even say in those days, like the CIA, the NSA, LLNL, Sandia National laboratory, and I got a job at Sandia.'
Sandia is a Department of Energy weapons laboratory, which does the weaponisation of nuclear devices designed by LLNL or Los Alamos.
She admits that getting involved in the nuclear weapons programme was not a fashionable choice. She says: 'My family had some issues with this. My idea was it was better to have a person like myself, with good Mid-West values, who was a patriot, rather than some hawk who would actually want to use them. Most of the people at weapons labs know that the use of nuclear weapons is as a deterrent. Nobody wants them to be used, but they want them logically used as a deterrent. They have to be credible and have to work should the need ever arise, and people have to know that they will work.
'People have to get their head around it to work there, but it depends what part of the lab you are working on whether you worry about it every day. Parts of the laboratory support the weapons, but are not interacting with them daily, and there are other sponsors of research. The underpinning science may be nuclear weapons but there are programmes related to radiation transport, which is important in global planet modelling, for instance.'
Her first job was in the solar energy programme, as she had to wait for her security clearance. She had to model a solar trough system using a computer, the first time she had contact with computers, so she had to teach herself Fortran, which in those days used punch cards. She managed to find a closed-form solution to all the equations and was hailed a hero. But it took her forever to turn this into a program that worked, so her star fell a bit.
German became a hobby, so Sandia encouraged her to get a masters degree in something more relevant. The lab sponsored her to do one year at Stamford but, as there were no one-year programmes in computer science, she did a masters in operational research in 1977. When she returned to Sandia she worked on a number of modelling problems.
She says: 'As you start modelling algorithms, you start getting closer and closer to the workings of the computer itself, which got me into operating systems. We got our first VAXs and first Cray in 1980 and they needed people to teach other people how to use them and I moved into the user services side, specialising in VAX. Livermore and Los Alamos had developed a timesharing system for the Cray and I moved over to supporting that, so I started spending time at Livermore.
'I met Seymour Cray many times back in Minnesota. I can remember several conversations with him about getting away from vectors towards massively parallel systems. That did not go terribly well; it was tough to tell Seymour Cray that it was the way of the future. He was big on vector computers and they were doing very well. But they were getting more and more expensive and, as it turns out, we needed something that was more mass produced because only a few people could afford them.'
Crawford eventually gained management responsibility for all the operating system work. She started to rise up the management chain, until eventually she was managing the whole computation department. She found she was good at management, something not normally associated with those who were strong technically.
Sandia decided to consolidate its computing operations at Livermore and its other location in Albuquerque, with Crawford as overall director. This led to her commuting by plane every week between Livermore and Albuquerque for 10 years, with her family staying in Livermore while she lived mostly in hotels.
Sandia was very important in the emerging high performance computing field and Crawford played a major role in that community. She was general chair of the 1997 Supercomputing Conference, and programme chair for Supercomputing '93. She started the Heterogeneous Bandwidth Challenge, which became a regular feature of the Supercomputing conferences. Eventually she persuaded the various participants to do this for real outside the conference, and the National Information Infrastructure Testbed was born.
At Sandia she moved on to become director of the model-based design and manufacturing integration program, which took her into a broader role in the laboratory around applying technology to improving the operational systems.
The defence laboratories started to fall behind other industries in high performance computing by the mid-Nineties and the end of the Cray Era. When Randy Christensen and Dave Novak from Los Alamos proposed using computer simulations to replace underground nuclear testing, Crawford became one of the Sandia representatives on the pioneering Accelerated Strategic Computing Initiative - which was jointly managed by LLNL, Sandia and Los Alamos. Its first funding came in 1995. After discussing the problem the group concluded that they needed a system capable of working at 100 teraflops (described as the 'entry level'). Nobody had broken the one-teraflop barrier at the time so it was quite a tall order.
Crawford had a lot of contact with LLNL. She always thought of it as the lab for computing. She had originally applied to LLNL when she got her job offer from Sandia and she always thought that maybe she would work there one day. Crawford had discussed some jobs there, but was always in the middle of a project she was deeply committed to and nothing came of it. When the job of associate director for computing came up she thought the time was right to make the move.
Crawford moved across the street to LLNL in 1999. She has 1,000 people in her organisation and a huge legacy of excellence in computing to live up to.
Crawford says: 'We are number one in computing in the world and I would like to keep us there - and that is not easy because, when you are number one, everyone else is trying to pull you down. We are fielding two of the largest systems in the world, including one that has 131,000 processors, but at the same time we are looking at what we are going to do next to keep on the edge. We are hoping to get to a petaflop by 2010 at the latest.
'My job is to create the environment that allows other people to do their job. That does not just mean getting the funding. An associate director is expected to know something about what they are in charge of, and it gives you credibility. It also means you can help sometimes because you know the issues they are facing.'
Being the only woman at the meeting for most of her career never fazed Crawford, or held her back from reaching one of the top jobs in scientific computing. She says: 'There were very few women in my maths class and I was told that I could become a teacher, which very few of my male colleagues were told. I have always been accepted for my abilities; I cannot say the same is true for some of my female colleagues in other science domains, but I am oblivious to it.
'There were very few women staff members at Sandia - and those that were either were not married yet or had put off having families. But I had my first child in 1981 and I started something then for women at Sandia Livermore, which is working part-time. I did it for five or six months, then went back to full-time.'
Crawford now spends time trying to encourage young women to go into careers in science.
She loves what she does but she does not believe that it would be good for the organisation to stay in her post indefinitely. The Department of Energy contract to run the LLNL, currently held by the University of California, is up for renewal in October 2007. Crawford said that whatever happens there will be a new 'bid team' running the laboratory, even if the University of California is still a partner.
She says: 'Since I am an associate director, then that could be one of the positions that changes. Since I will be on one bid team and the other may win, I might not have a job here.' Her colleagues are sure that she will get something, whatever happens.
Dona Crawford CV
1973 BA Maths and German, with teaching qualification, University of Redlands, California.
1975 MA German, Middlebury College, Vermont
1977 MSc Operational Research, Stanford University, California
1976-99 Sandia National Laboratory, Livermore, California, and Albuquerque, New Mexico, joining as a programmer doing mathematical modelling, then Cray Time Sharing System project leader, operating systems division supervisor, computation department manager, scientific computing centre director, director of national information infrastructure research and development, director of distributed information systems, and finally director of the model-based design and manufacturing integration programme
1999 to date, associate director, computation, Lawrence Livermore National Laboratory, California.