Features:

INTERVIEW WITH BOEING'S HPC SERVICE MANAGER SURESH SHUKLA
by Alan Beck, Editor-in-Chief

HPCwire's Editor-in-Chief Alan Beck recently interviewed Suresh Shukla, HPC Service Manager of the Boeing Company, about the decision making aspects involved in the HPC industry.

Q: You gave a presentation on the HPC decision-making model "CART" at the HPC User Forum in Europe. What is CART, and what necessity did you see for introducing such a decision-making model in the first place?

Shukla: At the HPC User Forum in the USA, we discuss many elements pertaining to the decision-making process for HPC requisitioning. The concept of CART evolved in my mind because of these discussions. The CART model categorizes the drivers behind such decisions, and provides a perspective that streamlines the decision-making process. It helps decision makers to weigh the drivers appropriately.

CART stands for four components: Cost, Applications, Requirements, and Technology. Cost is controlled by Management, Applications by in-house or third-party developers, Requirements by users, and Technology by HPC vendors. A decision happens when all of these components are agreed upon at a given point in time. This CART process moves in time, and makes decision-making more difficult due to the continuous changes happening in these components.

Q: What's wrong with the way HPC buying decisions are often made today within US Industry?

Shukla: There was a time when enough importance was being given to user requirements amongst the four CART components. In the last five years that has changed. Management has become the driver, and the importance of HPC environments has shifted from user requirements to Cost. There is nothing wrong in buying a low-cost solution if it fits the need; but when cost becomes the primary driver, vendors and application developers tend to spend their energies and R&D dollars to satisfy these cost requirements, and the users' requirements to solve HPC problems are not given due significance.

Q: What are the negative outcomes of this kind of decision-making process?

Shukla: The low-cost solutions tend to address problems that were solved by using HPC platforms of yesterday. New grand-challenge problems are left unanswered. This has stunted progress in the HPC Industry during the last five years in the USA, which is obvious from articles published about Earth Simulator by people in the know.

Q: Why are you stressing the component of "user requirements" in CART so much?

Shukla: Computing, that addresses user requirements at any given time which cannot be addressed by low-cost computers, is how HPC needs to be defined. User requirements are always expanding. For example, I solved 8,000 equations 10 times in a day with a CDC7600 in 1973. That was an HPC problem in 1973; but it is not today. HPC problems in the years 1980, 1990, and even in the year 2000 are not what they are in the year 2003. Problems in the year 2004 wouldn't be what they are in the year 2003 either. All this is the result of expanding user requirements. Without it, HPC wouldn't exist.

I think we need to start talking about HPC with a subscript of a year following it. For example, HPC1980, HPC2003, etc. By the definition mentioned above, the HPC platforms have to cost more than commodity computers. Giving cost the premium consideration it gets today would lead to not having HPC capacity within the organization.

Q: So, do you think many organizations are buying based on price, rather than on what they really need to get their work done?

Shukla: Yes. But that is not the only unfortunate part of it. They are proud that they are doing a good thing for their organization by saving a lot of money. This would reduce the costs for the organization in the short run; but it stunts the growth of thinking of amongst the users. That would not be helpful to the organization in the long run. Many are oblivious to this phenomenon. Unless users in an organization empower themselves to state their case, the pendulum wouldn't switch around. That is a difficult task, as even some users have joined the cost bandwagon recently.

Q: How can users exert a stronger influence on HPC buying decisions?

Shukla: Firstly, by realizing what's happening, and then forming "user teams" within their organizations. This is to ensure that they are jointly able to convince management about the necessity of solving the HPC problems of today and tomorrow, not yesterday.

Q: Why aren't more users doing this today?

Shukla: As I said, this is a cost-bandwagon effect. Management controls the dollars. It is difficult to argue with them. But the more united and empowered the users are, the more management would listen.

Q: How do I know that the decision I make today won't turn out to be the wrong decision six months or a year from now?

Shukla: This is an interesting question. The decision one makes today would always be wrong to a certain extent six months or one year from now. If that doesn't happen, we are not dealing with HPC Technology. It is the nature of HPC Technology to change, and change very fast, as the CARTwheel moves through the space of time. Costs can go down for the same platform or for a competing platform, a new algorithm may make the previous efforts worthless, a vendor may start having troubles in supporting the purchased platform, and anticipated problems can become so large or complex that the purchased platform proves to be inadequate again. One needs to develop a comfort factor with these changes, and make the best effort to anticipate the changes, and make the decisions with some futuristic user requirements in mind.

Q. Does the CART model stop at identifying four components?

Shukla: No. The CART model starts with the identification of four components. It then categorizes many elements under each component. The purpose of this decision- making model is to have all elements identified in an organized way so that we don't lose sight of any element while making the decision. An organization can decide to give different weights to different elements to suit their specific situation. If some people are interested in looking at all the elements, they can go to the site http://www.idc.com/hpc.

Note: The views noted in this interview are Suresh Shukla's own and do not necessarily represent views of others in The Boeing Company.