Cover Image


View/Hide Left Panel
Click for next page ( 331

The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement

Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 330
330 not necessarily to develop an architecture that matches the physics but to develop a high degree of parallel pro- cessing leading to a machine that is approximately 10 times faster than the fastest commercially available machine. It appears that specialized architectures embodying some features of the physics of the problem are unlikely to be important for astronomical computing in the 1980' On the other hand, very fast computers may well become available as a result of work like that being done at NASA/Ames. For the same reasons as stated in Recommendation 3 of Section IV on theoretical computing, mutual benefits accrue when astronomers are allowed access to nonastro- nomical computational facilities. - ~ - mus, any new facili ties that are developed should be open (consistent with the objectives of the facility) to use by the astronom- ical community. For its part, the astronomical community must remain aware of developments in this area and make plans for the effective use of such facilities. APPENDIX 5. A: THE "CANONICAL " SYSTEM Earlier sections of this chapter have recommended the installation of minicomputer-based facilities equivalent to 30 "canonical" systems. This Appendix provides some of the characteristics of the canonical system and expands on our concepts for the implementation of these systems. It must be emphasized at the outset that the canonical system is a concept that is used solely for the purpose of sizing and costing. Some installations, such as those used primarily to support image analysis for a small uni- versity group, may be only one third the size of the canonical system. Others, such as those at the National Astronomy Centers, which must support processing and analysis for a large number of users, may be three or four times the size of the canonical system. It may well be that no system exactly matching the canonical system is installed in the 1980's. At present, one may purchase, for approximately $220,000, a system with the following characteristics: 32-bit address space with virtual memory, 1-Mbyte physi- cal memory, complete floating-point intruction set, 50 Mbytes of disk storage, one tape transport, and a capa-

OCR for page 330
331 city of 1-2 ~FLOPS. On a system primarily used for theoretical computing, an array processor is extremely desirable. Systems used primarily for image analysis will benefit from sophisticated display and hard-copy peripherals. Of course, array processors are useful for image analy- sis, and sophisticated displays are useful for theoreti- cal computations--if one must choose one or the other, then theorists will choose the array processor while the image analyst will select the display. Either of these two options will bring the purchase price of the system to about $300,000. We believe that such a system is ade- quate to support the research of a medium-sized theoreti- cal or observational research group or department. Smaller groups can make significant progress with systems down to the $100,000 level, although the array processor, sophisticated display, and 32-bit virtual memory archi- tecture must be dispensed with (the last, only for the next few years). Larger groups, in which theory and analysis are to be done on the same machine, may require large systems with more memory, terminals, and both the array processor and display devices. Facilities to be deployed at major observatories and intended for image processing as well as analysis may be several times the size of the canonical system depending on the number of users that must be served. Our proposed implementation strategy calls for the funding agencies to provide for the purchase of systems equivalent to five canonical systems per year, which allows systems to be replaced every 6 years as they become obsolete (see below for a discussion of the need for periodic replacement). A system to be installed at university should be proposed by astronomers at the uni- versity and evaluated through a peer-review process that takes account of the scientific merit of the research performed by the potential users of the system and the impact the system will have on this research. A system to be installed at a National Astronomy Center should be evaluated in the context of the objectives of the Center, together with the scientific programs and requirements of the users and staff of the Center. In addition to the funds needed to purchase capital equipment, funds are also needed to provide for the main- tenance and operation of the systems and basic software support. These expenses may be met by cost sharing between the universities and the funding agencies in the case of university installations but must be wholly . a

OCR for page 330
332 included in the operating budgets of the National Astronomy Centers. Typical service contracts are in the range of 6-10 per- cent of the purchase price per year; we have adopted a figure of 8 percent for our cost estimates. Operations expenses include consumables such as paper, magnetic tape, power, and air conditioning. Although these expenses can amount to several thousand dollars per year, we have ig- nored them in our cost estimates. The systems that we are proposing, although relatively inexpensive, are quite sophisticated. It is our estimate that each system will require a minimum level of software support equivalent to that provided by one full-time soft- ware person. This support includes installation and main- tenance of the operating-system software, development of local system software, development of basic applications software, adaptation of software developed at other sites, and assistance to users. Maintenance and software support can be provided through several mechanisms. Some groups may wish to keep their astronomers "doing astronomy" and elect to provide these functions through a service contract and the full- time employment of a systems programmer. In this case, maintenance and software support would appear explicitly in the budget of the group. At the other extreme, some groups may elect to provide these services through part- time involvement of faculty and students. In this case, maintenance and software support would not appear in the budget and would seem to be provided free, although in reality the costs have merely been shifted to another part of the budget. We arrive at the following formulas for the steady- state costs of the systems that we have proposed: ~ _ Cm where Cc' ~m' = i2/T, NPM, = NS, and C~ are the annual costs for capitaL~equ~pment, -maintenance, and software support, respectively; N is the number of systems; P is the purchase price of an average system; T is the system ~ _ _ , _ _ _ lifetime in years; M is the annual service contract cost as a fraction of the Purchase price; and S is the salary, benefits, and indirect costs required for a computer professional. With N = 30, P = $300,000, T = 6 years, M - 0.08 per year and S = $50,000 per year, the steady- state annual costs are

OCR for page 330
333 C- _c $1.5 million/year, $0.7 million/year, $1.5 million/year. As mentioned above, we expect that the funding agencies will be responsible for gs and that cost sharing is appropriate for ~ and ~ . If, for example, 50 percent cost sharing is achieved, the cost to the funding agencies would be $2.6 million/year. We believe that funding of these systems should begin immediately and build up to this ~steady-state" value in 3 or 4 years (the buildup will not require 6 years because some systems are already in existence) and should there- after remain constant in real dollars. Of course, the computational facilities must be periodically reviewed and evaluated, and upward or downward adjustments may be required. However, the concept of a steady level of fund- ing in real dollars is important, as it provides for con- tinuous upgrades in the computational capability as new technology becomes available. Such upgrades will be necessary to handle the ever-increasing data volume and the growing complexity of theoretical computations. With the model that we are recommending, the average computa- tional capability available to the majority of the astro- nomical community advances with the state of the art, remaining approximately 3 years behind the very latest in new technology. This will be a substantial improvement over the situation today in which computers of the 1960's remain the workhorses of several major observatories. A final advantage of our recommendation is that it provides the flexibility to accommodate modest changes in research emphasis, computational requirements, or funding levels. We are not concentrating our resources in only a few large facilities. If it should turn out, for example, that we have erred in the estimates of the needs for theoretical computation relative to image processing, it will be a relatively simple task to switch resources from one area to the other. If we have somewhat underestimated the com- putational requirements of the community, more facilities can be acquired for the same level of funding simply by stretching out the replacement interval.