Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
SUMMARY Present Activity in Crystallographic Computing in the U.S.A. Present crystallographic uses of computers include instrument control, data handling and reduction, application of direct methods of structure solution, refinement of both general and detailed features of structural models, and display of the results in graphic form including three- dimensional dynamic displays utilized in an interactive mode. The total annual cost of crystallographic computing being done in the U.S.A. is estimated at 10 million dollars of which, probably, only about half is charged directly to crystallography budgets. Those crystal- lographers heavily engaged in structure determination spend about 80% of their computer time on refinement of structural models. The computing cost per structure varies too greatly to be a meaningful figure for planning purposes; costs in Hamilton's survey reported at this symposium range from less than $1,000 to more than $100,000, the differences being attributable to the number of structural parameters being refined and this number, in turn, being dependent both on the number of atoms in the unit cell and on the detail in which the refinement was to be carried out. It is a truism that the cost per calculation goes down as the size of the computer goes up. A rule-of-thumb suggested is that a factor of two increase in machine cost is accompanied by a factor of eight increase in computa- tional speed in solving crystallographic problems. However, the actual gain depends on the particular calculation and machine combination. Further, it refers only to computation time and not to input-output time. Hamilton's survey shows that crystallographers, as a group, have experience with a great variety of computer facilities. The size of the computers used ranges from stripped-down minicomputers to the largest com- puters now generally available. Of the responding crystallographers, 62% reported experience with remote terminals, 377o had used computers at locations other than their home installations, and 42% had made use of program lists. Needs Although no poll was taken, the combination of the papers presented and the vigorous discussion of them suggest the existence of a fair consensus on the following needs, perhaps among others. 1. There is a need for computing capacity one to two orders of magni- tude larger than is presently available. Refinements of protein structures are now becoming possible, and such refinements will make significant demands on large-core, very fast computers. Such large-capacity computers are also needed for application of direct methods of structure solution to large structures, i.e., containing more than 40 to 50 atoms in the asymmetric unit. A third area in which the extra capacity is needed is in refinement of
structural models accurate in greater detail; such detailed information may be important to molecular biological function, on the one hand, or to fundamental understanding of the solid state, on the other. In Hamilton's survey, 227o of the respondents reported that their work is seriously limited by lack of adequate computer capacity and another 21% reported that they are moderately unhappy with their computer resources. The capacity for full-matrix least-squares refinement of up to 240 parameters is typical of one of the major types of large computers now in use (e.g., CDC 6400). For gross structural refinement this number of parameters corresponds to only about 60 atoms; for detailed refinements it corresponds to only about 25 independent atoms. The need for larger capacity is particularly evident when it is noted that the magnitude of the calculation goes up as N^, where N is the number of parameters. The parallel vector processing machines now coming into operation may have a significant impact on this problem. 2. There is a need for improved approaches, algorithms, and computer programs that will do more efficiently, or in better ways, what is done most, i.e., refinement of structural models now done by least-squares methods. Concomitant with this need are expressed needs for improved information and program exchanges, and for overcoming compatibility problems that presently limit the general utilization of desirable programs, including systems approaches, developed by certain leading groups. 3. There is a need for local computing facilities that would give faster response and would be more convenient to use. Two routes to this goal are indicated, (i) through local computer centers that are more responsive, with fewer bureaucratic problems and communication failures, and (ii) enhance- ment of computational capabilities through minicomputers on line with data- collection instruments in the crystallographic laboratories. 4. There is a need for attention to the computing problems of the crystallographer by the funding agencies, especially with regard to the sys- tematic development of computing resources (as differentiated from continued use of present resources in the present manner); there is now little programmatic support in the agencies for such development. Trends From a combination of the invited presentations and the discussion, certain trends can be identified that will affect substantial portions of the crystallographic community: 1. Increasing computer usage by crystallographers, both because of new computational techniques and because the number of crystallographers working on larger structures is steadily increasing. 2. More work on protein structure, as it becomes possible for protein structures to be refined.
3. Increasing use of direct methods for large structures. 4. Further program development to take up significant parts of the structure-refinement load with faster (less costly) methods than the full-matrix least-squares refinement method, which still is needed in the final stages. 5. Increasing demands for accuracy in details of structures; examples are the determination of small differences asso- ciated with biological function, and the precise determination of thermal motions and of electron densities for their contributions to fundamental understanding of the solid state. 6. Increasing up-grading of minicomputers used for on-line diffraction-instrument control (i) so that they can be programmed conveniently to operate the instrument with better optimized and more flexible data-collection strat- egies, and (ii) so that the relatively large portion of the computer's time presently unused can be applied to data processing and interpretation including, in some cases, least-squares refinement of structural models. 7. Further emphasis by some groups on facilitating program exchanges. 8. Growing interest in computerized data banks, including banks remotely addressable. 9. Increasing use of graphic displays generated and manipulated by computer, sometimes in an interactive mode. 10. Increasing emphasis at state, city, and multiply-located- company levels on state, regional, or company-wide computer- communication networks; fueled by visions of economy in the totality of all computing, this emphasis may not always work to the advantage of the crystallographic or other scientific user. 11. Growing budgetary pressures from sponsoring agencies toward development of either (i) regional or national crystallo- graphic computing centers addressable from remote terminals or (ii) regional or national computer-communication networks accommodating heirarchical computing attuned to crystallogra- phic needs.
Pros & Cons of Remote Computing Most of the discussion of remote computing dealt with networks, the ARPA network being used as the prime example. A "network" in this context is actually a computer-communication network. It consists, basically, of single or multiply redundant communication links between various traffic- controlling computers which, in turn, communicate with other larger computers at their respective sites. A computational problem entered on the net can, in principle, be assigned to the available computer most suited to that particular kind of problem. A second type of remote service is the regional computing center which is addressable from remote terminals located at the user sites. The con- templated national computing center for theoretical chemistry was cited as an example. Such a discipline-oriented facility could be expected to optimize its program holdings and service capabilities for that particular discipline. Although these two types of remotely usable computing services differ considerably in their desirability for various crystallographic computing purposes, some advantages and disadvantages applicable to both were brought forth in the symposium. In favor of the remotely usable service, the follow- ing five points can be summarized: 1. Protein crystallographers, and some others, need access to larger computers than are now ordinarily available in any one laboratory. 2. A quantum jump in crystallographic productivity might result from what would amount to a new dimension of computing capacity made available through a network, or a fully implemented regional center. 3. Sufficiently reliable high-speed data transmission, available either now or in the near future, may obviate most needs for physical transport of data or results, possibly including graphic displays. 4. The best computing system for carrying out the items of highest cost in crystallographic computing, primarily least-squares refinement of structural models, would be available to all in an operating and fully checked-out form. 5. Information retrieval would be possible from a central library of crystallographic data, in the forms of both hard copy and graphic display, such as a stereographic view of the structure. Several points of disadvantage of remote-service systems were also brought out: 1. There is a cost threshold that could keep out the small user. Figures cited indicate a basic cost of about $12 000 per year, exclusive of computing costs, for a remote job-entry terminal (card reader and line printer) such as might be desired for access to a regional computing center.
For the ARPA network, the basic cost varies with the type of entry. For connection of a local campus computer (through an Interface Message Processor, or IMP), the cost is presently $21 500 per year just for the prorated share of communication and maintenance costs. [On the other hand, information provided several months after the symposium suggests that some usersâin this case, of course, doing ARPA-approved workâcould be accommodated from their own termin- als through a 32-port Terminal IMP, or TIP, with a lower data-transfer rate (4800 baud) at much lower cost. In this case the costs, apparently, would be comparable to those for terminals remotely accessing regional computer centers via voice-grade telephone lines. For the most rudimentary type of terminal, the fixed costs could be an order of magnitude less than those quoted, but, of course, with a considerable compromise in utility.] 2. The full real cost of the computing service would be charged to the user. This could be a considerable disadvantage to many present users in institutional environments where real costs of computing are partially subsidized. 3. The growth of local resources for large computational problems could be impaired both through the diversion of dollar support to the remote- user system and through a sapping of local interest in program development. 4. The effects of less-than-optimum program operation or computing services would be magnified on a regional or a national scale. One weakness could be lack of dynamic program development. Another could be the wide- spread effects of any programmatic error that remained undetected for a time in a program used by a large number of remote users. Still another could be the disaster wreaked if, for political or economic reasons, the network or regional center failed or stopped rendering service. Recommendations Although no formal recommendations per se, were made by the symposium group, some that probably would have met with approval can be inferred. First, it would be recommended that sponsoring agencies, and others in position to influence these matters, give particular attention to the needs mentioned. Second, it would be recommended that all parties, whether sponsors, scientists, or administrators, be sensitive to the trends in crystallographic computing with a view toward enhancing and exploiting those that are desirable. Third, the symposium group's demonstrated interest in, and apparently positive attitude toward, the potentialities of networks and regional centers for crystallographic computing lead to the inferred recommendation that the possibilities be seriously investigated by some competent group. This compe- tent group should be appropriately related to organized crystallography and should be funded as necessary to carry out substantive investigation of all factors involved, including attitudes and preferences of the crystallographic community, of which the members of this symposium were not necessarily a representative sample. 7
