|Volume 7, Issue 1 -
A Dangerously Flawed Congressional Report
Ken Kennedy, Director, CRPC
I have just finished a detailed reading of the Congressional Budget Office (CBO) study entitled "Promoting High Performance Computing and Communications." The stated objective of the report is to focus on the role of high performance computing in promoting new industries, concentrating on obstacles, principally on "the demand side," that might prevent computing and communications industries from capitalizing on HPCC technologies. Unfortunately, the report suffers from serious shortcomings that make it more likely to be harmful than useful.
In its critique of the computation component of the HPCC program, the report questions the economic value of the HPCC concentration on scalable parallel systems, asserting that these "massively parallel" machines will not have much impact on commercial computers. The primary reason given for this assertion is that scientific workstations have a much larger market than supercomputers and the technology flow is no longer from supercomputers down to workstations but rather in the opposite direction because massively parallel systems increasingly use the same single-chip processors as workstations. But this misses the truth by a wide margin. If you examine the architectural strategies being used to achieve high performance in workstations-pipelining, memory hierarchy, instruction look-ahead, multiple instruction issue, multiprocessing and so on–every one of them had its origin in high-end computer designs. Furthermore, many of the software techniques that are making workstations more efficient also have origins in supercomputing.
This highlights one of the CBO report's most fundamental errors: its attempt to separate high-performance workstations from the scope of the HPCC program. HPCC research has focused on the high end because that is where significant problems are first encountered. Although the report acknowledges this strategy, it denies its effectiveness, failing to recognize that workstations will eventually reach the point where bus- based shared memory will not support the number of processors needed. At that point, all the dollars spent on high performance networks, scalable architectures, parallel programming systems, and scalable parallel algorithms will be appreciated as an excellent investment.
Despite its goal of impartiality, the CBO report seems to advocate a specific point of view--namely, that if the Congress is investing in the HPCC because of its benefits to economic competitiveness, the money is largely being wasted. Much of the problem arises because of the report's narrow view of the benefits of HPCC. From the outset, the report limits itself to the economic benefits to the computing industry, intentionally omitting benefits to other industries that use high-end supercomputers and benefits to society, such as those that will accrue from the solution of grand challenge problems. The report compounds these omissions by deliberately factoring scientific workstations out of the scope as well. As a result, the report is improperly one-sided.
There are many things about the implementation of the current HPCC program that are open to criticism. However, the distorted view presented by this CBO report is extremely dangerous, because it encourages Congress to throw out the baby with the bath water. Therefore, I call for the Congressional Budget Office to withdraw the report and rewrite it to achieve an appropriate standard of accuracy, balance and impartiality. In the meantime, I suggest that everyone read the report carefully so you can contribute an informed opinion to the debate that is sure to follow.
Table of Contents