Q: Your bio says that you are vice president, Deep Computing & Web Servers, responsible for Deep Computing Solutions, Linux clusters, and supercomputer strategy. How is Deep Computing relevant to business, not just supercomputing in scientific areas?
TUREK: Deep Computing is actually our effort to take people away from an esoteric view of supercomputing, and to understand the performance of complex calculations on massive amounts of data in order to
support business decisions. Problems are complex in all walks of life. Deep computing, from petroleum companies to financial institutions, refers to a
corporate ability to use technologies to solve their diverse problems. Our strategy is not to build separate computers, but to have a robust architecture for solving problems, and we think clusters are especially appropriate for that. So this is not just about scientific stuff, but about businesses with terabytes of data to mine, to discover such things as credit risks.
Q: Where does clustering stand in businesses today?
TUREK: Clustering is in full evidence in the commercial space. The approach is being used for server
consolidation, Web serving, standard database applications, and data mining. It's used in businesses such as manufacturing and logistics.
Q: What is IBM's outlook for growth in the business adoption of Linux clusters?
TUREK: Our belief is that the marketplace in the Linux area will evolve. You now can see a lot of acceptance in the technical and scientific space. Otherwise, the situation is fairly straightforward: Linux clusters are being adopted today by early technology adopters. Institutions adopting Linux clusters have common attributes: They're likely to have lots of in-house skills, with control over their own applications, and showing healthy attitudes toward risk. These kinds of companies are confident that they can build their own clusters.
Q: Some examples?
TUREK: The petroleum industry, for one, shows
evidence of having tremendous in-house technical skills. The more sophisticated financial enterprises
are another case in point . Just pick your favorite investment banks: They have a lot of Ph.Ds, and aren't waiting around for third-party applications.
Q: What still needs to happen regarding Linux?
TUREK: We still observe a lag of applications moving to Linux. We don't have all the databases and all the middleware there.
Q: One pain point is simply in daring to say "Linux clustering" in the same breath. Some traditional clustering experts contend that most Linux software doesn't meet the definition and should be relegated to tags such as load-balancing software or parallel-computing software.
TUREK: I share your pain. In general, I've found
some clustering definitions to be, well, confusing and perhaps self-serving. While purists get caught up in
matters of degree, I am simple at heart. The debate is a "single-system image" debate, arguing for single-system solutions. Everyone discovered that it was very costly to implement. People backed off, and manufacturers did not embrace the approach. Our view is that a cluster is a set of standalone devices capable of being managed as a cohesive unit. Under that umbrella, you can refine for personalized capabilities, under the higher-level construct of a cohesive management of a set of distinct computers.
Q: How key is load balancing as a business goal for moving toward Linux clusters?
TUREK: Load balancing is important, but it's only an incremental function and not the real issue. If you're talking about a business case for clustering, then it is in placing a rational management construct on to a set of servers. Look what happens in the absence of that. You get a company buying servers on a weekly basis with no scheme in hand for managing such things. The real clustering issues are computing costs and manpower. If I do an excellent job in load balancing but a poor job in user administration,
everything will be for naught.
-David Turek is IBM's vice president, Deep Computing & Web Servers, Enterprise Systems Group. Prior posts at IBM included directing systems development for the IBM RS/6000 SP.
| 
