The Microsoft "get the facts" report I discussed yesterday wasn't the only one I read - today's subject, a case study on a decision by people at the University of Cincinnati Genome Research Institute (GRI) to migrate their HPC clusters from Unix/Linux to Windows, is another winner.
Here's Microsoft's write-up of the problem GRI faced:
About 95 percent of the institute's computers run the Windows® XP Professional operating system, with identity management and authentication managed through the Active Directory® service on Microsoft Windows Server® 2003 Standard Edition. The remaining five percent are UNIX or Linux-based high-performance computing (HPC) clusters. These clusters are used for 'in silico' experiments that is, experiments performed using computer simulations.
Wortman estimates that only 15 people at GRI actively used the UNIX and Linux computer clusters. However, that group of core users regularly ran computational jobs on behalf of approximately 60 other researchers who lacked the technical skills to use UNIX and Linux.
"We were trying to engage basic research biologists with in silico drug discovery, but the tools we had were far too technically complex for many of them," says Wortman. "Users had to authenticate to various computers using Secure Shell and run their jobs from the command line. It was pretty difficult, and so we would run jobs for them. We wanted to move away from the core facility approach and enable people to work directly with the computer clusters."
"The heavy reliance on Linux for high-performance computing required GRI to have a skilled Linux administrator on call to maintain the systems, typically someone who did that job in addition to normal duties at the institute. The mixed computing environment was complicated to manage because identity management wasn't consistent across all computers; users had separate accounts in Linux, UNIX, and Windows[-based] environments," says Wortman. "We wanted to make IT management easier by bringing all of our accounts into Active Directory and standardising on an operating system that everyone was familiar with."
Read the case carefully and you'll notice first that the transition to full time Windows support has freed the organization from the horrors of having to rely on a user with a real job for Linux expertise, and secondly that there's some interesting temporal juxtapositioning going on here. Look at the third paragraph above again and you'll see that everything's in the past tense: "We were..we had..users had to.. we wanted to move away.." and that makes sense because this stuff is about why they embarked on the project. But look at the wording used to announce the project's success and you'll see that almost everything's in the future tense:
"To turn research biologists into supercomputer users, you have to make it easy for them, by providing systems that are familiar and easy to learn," says Wortman. "Our vision is to make high-performance computing into the resource that our users didn't even know they needed until now. Once people start using the new HPC system, I expect that we'll see a cascading effect where others will see the benefits, and start using the new technology. That will lead to more applications being developed, which will lead to even more users." (Emphasis added)
If you're writing about a successful project you put everything into the past or present tenses: "we did this, we achieved that, users are spreading the word". In other words, this case study announces the success of a project that hasn't succeeded yet.
Of course if the reality is that they've installed a Windows HPC cluster but it isn't attracting users, you'd expect another shoe to fall: something to make it more attractive to users:
Wortman adds, "Eventually we'll move toward a Web browser interface. Researchers work a lot from home and on the road. Providing a hosted Web-based application is a way of uniting resources and making them easy to access and use." (Emphasis added)
There are two big problems with that: strategically this is playing the Sun/Google game because moving the interface to web technologies frees the computing resource to move too.
Secondly, it leaves the reader wondering whether Wortman et al either don't know, or are just so Wintel centric that they don't care, that the industry standard open source gridengine portal already offers easy access to far more functionality than he's talking about developing - including unitary identity management, multiple concurrent host support, research portability, role and zone based data protection, regulatory compliant tracking, and automated job migration to national level compute resources - none of which run Windows.
When you write a puff piece you're naturally entitled to pick and choose what to include from notes on what the customer said and your own interpretation of what you saw on the site, but in this case the result is more suggestive of arrogance than smart marketing. The reason I think this is that every reported decision - from making the institution less attractive to researchers while creating unnecessary support burdens by porting science applications to Windows, to insisting on ssh command line operation for Linux- is an obvious bad choice: each one made in apparent ignorance of better, cheaper, and more strategic alternatives.
As a result I got the uncomfortable feeling reading this case study that the people writing up this project had to know better - and carefully picked both what they said and what they quoted Wortman as saying to create a kind of secret gloat: a celebration of themselves as smarter than the customer, instead of a marketing document.