The grid pioneer

Last year Johnson & Johnson united its grids in the US and Europe. Jeff Mathers, Director of Technology Office Strategy and Delivery at Johnson & Johnson Pharmaceutical Research and Development, discusses the challenges.

When most big pharma companies were cautious about investing too much time and resources in the relatively unchartered waters of grid technology. Johnson & Johnson had already plunged headfirst into deploying grids in the US and Europe. By 2004, under the direction of Jeff Mathers, Director of Technology Office Strategy and Delivery at Johnson & Johnson Pharmaceutical Research & Development, L.L.C., (J&JPRD), part of Johnson & Johnson, his team had activated the first clinical development system to run on the grid, allowing the company to quickly push several drugs through the FDA approval process and saving millions of dollars from the bottom line.

Deploying grids in the US and Europe at a time when grid computing was touted as hype was certainly a risky gamble. The real driver in the beginning, says Mathers, was an ageing legacy of high performance computing equipment. “We knew it would pay off in terms of server consolidation and reduced maintenance costs for SGI equipment, so we knew right away that we would achieve return on investments through that alone,” he points out. “But what we’ve also seen is certain types of application and scalability capabilities beyond what we could have ever purchased that’s impacting on scientific processes. That’s really where we see the major ROI over time.”

Last year, Mathers’ team played the ace by uniting the US and Europe grids to create the single, most advanced pharmaceutical grid in the world. “Many people have different definitions of what grid computing is. Our definition is managing work and managing computing tasks efficiently through a single interface. The range of possibilities is huge. One of the things we did differently from other pharma companies was having the intention to develop a single grid globally, right from the start – one that could do both validated and non-validated computatioanal work,” he explains. “We began with two separate grids. The European grid focused on early research and the US grid focused on the clinical space. Once they were both working, we merged the two grids together.”

Mathers had confidence in the technology; the challenge was the diversity and complexity of Johnson & Johnson’s make-up. “We have so many different companies and each company, in one way or another, had their own computing equipment,” he admits. “We don’t have a single computer configuration process, whereas other pharmaceutical companies are more centralized in terms of how their computer structure is managed. So, actually making the grid stable and viable was a little bit more complicated than we’d initially thought. Surprisingly, what did turn out to be less of an issue was the impact of our wid area network. We thought that would be a big impediment to doing this project but that hasn’t turned out to be true.”

Johnson & Johnson is now running over a dozen applications on the grid, more than any other pharma company. Why haven’t more of their competitors followed suit? Mathers believes it isn’t a case of firms being reluctant to deploy grids. Rather, their rivals’ approach to implementation has held them back. “A lot of other companies went straight in and put out a huge grid, then tried to figure out how to put applications onto it. We approached it from the opposite angle and went out to find applications first,” he says. “When those applications were working we scaled the grid. We took our time and tried not to build the grid too quickly. Other companies have struggled with the infrastructure and it’s held them back from focusing on the applications.”

Double up

Plans are now underway to double the size of the network and the number of applications running on it. Mathers hopes to reach their current license maximum by the end of this year – about 2500 CPUs of computing power – and is looking to extend the number or applications being run on the grid to 15-20. “The biggest problem that we have right now is that we need more computing capacity because we have too much demand from the business,” he explains. “Some other companies have thousands and thousands of machines but we don’t see that many machines which would be good candidates for our grid. We may need at some point to look at a way to farm out work to other organizations on some level that could be combined with our grid.”

The initial hurdle has been trying to convince people to contribute their computing power to the grid in the first place. These so-called server huggers are naturally worried that sharing their PC with others could throw up all sorts of access, security and, privacy issues in an industry where fierce competition and secrecy thrive. J&JPRD’s Advanced Biology and Chemistry Discovery (ABCD) Platform, for example, will eventually contain all of the company’s research data and will be available to company scientists around the world. The grid’s features include a highly configurable scheduler, encryption, and security services that sit right inside the grid environment to balance local and grid computing demand, protect data from local access, and also protect the local machine from access from the grid software as well.

Mathers adds: “For the most part, we don’t do a lot of desktop grid computing. End users don’t normally see or contribute cycles off their laptops. We have a lot of servers and dedicated workstations on the grid focused on business units that are using the grid. That seems to be fairly effective because the groups contributing resources are getting something in return. Yet, we still have this problem of server hugging with new groups and have to work through a whole series of dialogues and meetings to get them to understand what grid technology actually is and how their computers will behave.”

However, he also highlights that one of the key reasons J&JPRD’s grid has been successful is that the scientists involved actually want to work in the grid space. “If you don’t have that I don’t care what the technology is, it just wouldn’t work,” Mathers concedes “It takes people with a really open mind and an innovative spirit to embrace grid technology and we’ve been really lucky that we’ve found people who are willing to explore the concept of changing the way they do research.”

Of course, Mathers has to take the credit for being the visionary behind the grid project and has, in fact, won awards for his pioneering work in implementing and advocating grid computing as an enterprise platform. He also has some definite ideas about how he would like to see grid technology develop in the future. “I would expect to see more commercial packages like MATLAB and SAS technology start to take advantage of grid technologies more. I think there’s a lot of opportunity for open source technology – how we look at those in the company and how they can be easily integrated into a grid environment,” he says. “As tools mature, it will be easier and easier for scientists and people in the business (statisticians and those working in bioinformatics, genomics etc.) to integrate their own software and their own informatics projects into the grid environment.

“The other big area I see is the whole concept of data grids. Once you’re leveraging a grid for producing results and scientific outputs, you then have by default a global data directory. When you know where an idea has been generated, there is the potential to exchange and combine knowledge and output. It’s a big area that’s still untapped,” he added. Also, there is still early work based on ‘pipelining’ being carried out, which is essentially what we call assembling multiple grid tasks together in a sequence that has been automated.

“We struggle with the fact that there are so many organizations carrying out work in so many different places. In two or three years we’ll have something that’s more robust. The power of pipelining and the power to see a universal data directory are going to be huge.”