Faculty Forum Papers
THE EVOLUTION OF COMPUTING AT
OREGON STATE UNIVERSITY
Curt Cook, Sheila Cordray, Ted Lewis, George Martin, Jim Richman, Michael Schuyler,
David Sullivan, Barry Shane
THE EVOLUTION OF COMPUTING AT OREGON STATE UNIVERSITY
ABSTRACT: Last month the Milne Computer Center issued a PLAN FOR MAINFRAME COMPUTING recommending the immediate purchase of another major upgrade to the Computer Center's CYBER 170 mainframe. Although the immediate upgrade would provide an increase in raw computing power, it crushes any short-term hope of improving the quality of mainframe computing on campus and seriously compromises our long-term options by investing in dead-end technology.
The PLAN FOR MAINFRAME COMPUTING implies that we face a serious CYBER capacity problem in FY 1985-1986. Our investigations show otherwise. In Section 1, we use three forecasting techniques to predict the demand for CYBER computing. These techniques show that OSU's mainframe computing usage has been flat for the last two years and predict that CYBER usage is not likely to grow significantly next year. Section 2 describes the amount of computing we can expect from the CYBER next year. In the last twelve months a number of substantial performance upgrades have increased the CYBER 170's capacity by more than 50 percent. Section 3 suggests alternative computing growth paths that provide immediate growth in computing capacity and also build toward a healthy long-term computing environment.
The Milne Computer Center is responsible for providing general purpose computing facilities for the entire university. Last month it issued a PLAN FOR MAINFRAME COMPUTING that makes a number of recommendations for the 1985-1987 biennium. Reviewing this plan carefully is important, because the decisions we make now will affect our computing environment for a long time.
The PLAN FOR MAINFRAME COMPUTING proposes a quick fix solution to a short-term mainframe computer capacity problem. It recommends that we immediately acquire another major upgrade to the Computer Center's CYBER 170 mainframe. It also suggests that we should acquire a new mainframe from an as yet unselected vendor. These two recommendations are at odds with each other. The proposed fix will purchase equipment with a brief usable life and will compound our eventual conversion problems.
The issues involved in developing a good growth path for a university computer center are exceedingly complex, and to date the issues have not been adequately discussed at the university-wide level. This memo supplies some important facts that are missing from the Computer Center's plan and makes some tentative recommendations. Overall our impression is that the Computer Center's plan would set OSU back by about a year in developing a healthy high quality computing environment.
This is not a convenient time to make long-term decisions about mainframe computing. We are in the process of hiring a new director for the computer center. Last month John Byrne proposed a new administrative structure that includes the establishment of an Associate Vice President for Computing and Communications. Administrative computing is migrating -- probably in FY 1986-1987 -- to a new centralized State System of Higher Education facility. Personal computers are invading all aspects of campus computing. Large-scale computing projects are shifting to departmental facilities and National Science Foundation supercomputers. All of these changes are occurring swiftly and promise to have a significant effect on the long-term demand for mainframe computing. Given the turbulent state of computing on campus, it is not surprising that a task force consisting exclusively of Computer Center personnel should recommend a stopgap solution to their immediate operational problems. After all, they are the ones who face the day-to-day tasks of running the Computer Center.
The PLAN FOR MAINFRAME COMPUTING is discouraging in its scope. It reacts to problems rather than seizing opportunities. Its five criteria for an acceptable plan are designed to protect existing users and ensure that not mistakes are made. These criteria are important, but are they the most important considerations? Where is a list of needs for mainframe computing? Where is the vision of what we are trying to accomplish? Where does the PLAN FOR MAINFRAME COMPUTING fit within the University's plans for personal computing, networking, printing, and communications? Only in an emergency should tactical decisions be made without reference to a master plan. Bluntly stated, we should not continue to upgrade an obsolete mainframe; we should begin converting to the next mainframe.
1. FORECASTING THE DEMAND FOR CYBER 170 COMPUTING
Forecasting demand for computing is difficult. As many different techniques as possible should be used to allow for tests of internal consistency.
1.1 FORECASTING BY EXAMINING GENERAL TREANDS IN COMPUTING
The PLAN FOR MAINFRAME COMPUTING correctly states that the demand for mainframe computing has been consistently underestimated throughout the history of computing. The continued growth of mainframe computing has surprised doomsayers who often call mainframe computers "dinosaurs." Thus, all other things being equal, mainframe computer services should continue to increase in importance.
Another truism is that the demand for computing is driven by the availability of useful software. It is on this dimension that the Computer Center's CYBER 170 series computer comes up short. Although the 170 series provides a tolerable programming environment, it doesn't run most applications software. Commercial and educational software is written first for IBM and DEC computers and is rarely modified to run on Control Data Corporation (CDC) computers. The PLAN FOR MAINFRAME COMPUTING says there "is an identified need for software packages such as SAS, SPSS-X, large scale LP models, report generators, versatile editors, powerful graphics tools and easy to use communications facilities." These important tools are missing from the 170 series software offerings. Other missing software categories include word processing, spreadsheet processing, and other personal productivity software packages. Now that the CYBER 170 series machines are no longer in production by CDC, there is little hope for improved software. Also the NOS operating system used by the CYBER 170 series is downright difficult to use. It totally intimidates first time users. Without improved software, there is unlikely to be a substantial increase in demand for mainframe computing.
The role of mainframe computers is shifting within the field of computing. Mainframe computers are becoming database machines that provide communications and access to a wide range of peripheral devices. Simple processing tasks, such as editing and programming, are being picked up by less expensive minicomputers and personal computers. Large-scale computing tasks are shifting to more capable supercomputers. Mainframe computers are appropriate when many people need to interact with large data sets, as with administrative computing. These considerations suggest that we will continue to see instructional and research tasks migrate from the CYBER 170 to departmental minicomputers, personal computers, and off-campus equipment.
A final trend to consider is the consistent declines in the cost of computing worldwide. For the last decade the cost of microcomputing has fallen between 20 and 50 percent compounded annually. The cost of mainframe computing has fallen less dramatically but has declined consistently. These declines in cost have caused mainframe computer usage to increase elsewhere. Against this trend has been the rising cost of computing provided by the Milne Computer Center. The rates for CYBER 170 services increased in three of the last four years. Last year the rates were held constant; the rates for next school year have not yet been set. The Task Force on Large-Scale Computing (part of the STRATEGIC PLAN FOR COMPUTING developed last year) had this to say about the cost and capability of our current mainframe:
The facts are clear. We have 1/7 - 1/2 the capacity of comparable universities…the CYBER 170 has a small central memory (132K 60-bit words), no virtual memory capability, is slow (1/20 the speed of a CDC 7600 and 1/300 that of a CRAY), has small but expensive mass storage capacity, and has costs that are 2-6 times higher than those of machines with greater capability.
In conclusion, although the potential demand for mainframe computing is undoubtedly increasing on campus, little of the potential demand is likely to be converted into actual usage on an augmented CYBER 170.
1.2 FORECASTING DEMAND BY CONDUCTING SURVEYS OF USERS
Unlike most businesses, the Computer Center does not normally survey its customers to predict next year's sales. The last two user surveys conducted by the Computer Center were prepared as part of the STRATEGIC PLAN FOR COMPUTING in June and September 1983. Clearly these surveys are of little use in predicting FY 1985-1986 computing demand. As a result we have no quantitative data about the level of satisfaction and plans of administrative, research, or outside users of the CYBER 170.
Fortunately the University Computer Committee conducted an instructional computing survey in January 1985. The survey was sent to all instructors who used the CYBER in their winter term classes. When the responses were weighted by the number of dollars spent in each respondent's class, there was more than an 80 percent return rate for the surveys. To expedite the processing of data, questionnaires for classes with at least $2,000 of winter term spending authorization were given the most analysis. Twenty-one responses representing more than 60 percent of the total spending authorization fell into this category.
The survey's results are surprising. Very few instructors of the "big 21" computing courses would elect to continue using the CYBER 170 in fall 1985 if given the option to use microcomputers. Only 24 percent of the dollars are spent by instructors who would prefer to continue using the CYBER. Seventy-two percent of the dollars are spend by instructors who would prefer to use microcomputers. Among all the courses surveyed, microcomputers were preferred over the CYBER by a 52 percent to 38 percent margin.
It is hard to tell how much the instructional computing survey tells us about next year's demand for CYBER 170 services. At least in the short-term, nearly all of the funds that are granted on a university-wide basis for instructional computing (over $450,000 per year) are held captive by funding and administrative procedures that effectively require using the CYBER 170 even if it is grossly inappropriate for an instructor's class.
1.3 FORECASTING DEMAND BY EXAMINING RECENT USAGE FIGURES
Trends in computer usage are easiest to evaluate with the help of graphs. Figure 1 shows the Mine Computer Center's quarterly accounts receivable billings for "Central Processor Time." Figure 2 is similar to Figure 1 except that it graphs "Total Processor-Related Charges," an account grouping that includes central processor time, file storage, and a wide range of other miscellaneous services.
Both Figures 1 and 2 show a consistent yearly usage pattern. Computer usage is low in the summer and grows to a peak in the spring. Both graphs reflect dollar amounts, not constant amounts of computing services. The Computer Center's rates increased by an average of 5 percent between Spring 1983 and Summer 1983. Thus, if a true comparison of the demand for mainframe computer services is desired, the height of the first four bars in each of these graphs should be increased by 5 percent. Overall, Figures 1 and 2 show that the total demand for mainframe computing at OSU has been remarkably constant over the last two years.
In August 1984 the Honeywell computer was turned off after having been used for only four years. It is still sitting in the Milne Computer Center, unused. Presumably, the Honeywell users switched to begin using the CYBER this school year. The graphs show that CYBER central processor usage has risen by about 10 percent over a two year period, because the Honeywell computer is no longer being used.
Figure 3 tracks "CYBER SRU Usage" for the three year period from January 1982 to January 1985 as reported in the Milne Computer Center's Access newsletter. This bar chart tells essentially the same story as the earlier figures. Timesharing usage appears to have stayed about the same; batch usage varies more dramatically month-to-month and has risen slightly.
2. FORECASTING THE SUPPLY OF CYBER 170 COMPUTING
Unused central processing time is wasted computing capacity. Processing capacity cannot be stored like wheat until needed. This creates the problem of leveling peaks in demand to match a constant source of supply. The Computer Center's rate structure helps by making evening and weekend processing rates significantly cheaper than prime time rates. Still, no matter what is done, the demand for computing on Friday at 4:00 p.m. of dead week will exceed the supply. Thus, it is not surprising that the PLAN FOR MAINFRAME COMPUTING reports response times "in excess of 10 seconds and waiting queues of 30 or more users" during peak periods of use. From this perspective the necessary amount of processor capacity is clearly a judgment call.
A computer's capacity depends heavily on the capabilities of its peripherals. The PLAN FOR MAINFRAME COMPUTING does not mention that the Computer Center has purchased and leased four major upgrades to the CYBER 170 in the last 12 months. In May 1984, a memory upgrade was purchased for $20,000 with a monthly maintenance cost of $489. This increased central memory by 48 percent from 132K 60-bit words to 196K. It is an interesting commentary on the field of computing that an equivalent memory upgrade for an IBM Personal Computer (512K 8-bit words) costs about $300. In August 1984, a CDC front end communications process with a 64-channel capacity was leased for $3,478 per month including maintenance. In November 1984, a disk drive and controller were leased for $2,825 per month including maintenance. This increased disk capacity by 76 percent and provided faster disk access. Finally, in March 1985, the number or peripheral processing unity (PPUs) was increased from 10 to 14 at a monthly lease cost of $1,795. This resolved the last input/output bottleneck and left the central processing unit as the limiting performance factor during peak processing periods.
The cumulative effect of these upgrades has been to raise the number of users the CYBER 170 can handle from 80 to 130. In the April 30th open forum discussing the PLAN FOR MAINFRAME COMPUTING, Tom Yates (Director of the Milne Computer Center) said that these upgrades had caused "certainly more than a 50 percent increase in capacity." Yet the graphs in the previous section show nowhere near as large an increase in actual usage. It appears that we already have enough CYBER 170 capacity for FY 1985-1986.
If we follow the PLAN FOR MAINFRAME COMPUTING and upgrade the CYBER 170 again, the central processing speed will increase by another 90 percent at a purchase cost of $64,000, plus $1,433 per month for maintenance. It is unrealistic, however, to expect that total throughput would increase by 90 percent. It can be expected that some part of the system other than the central processor would limit performance first.
3. A RECOMMENDED COMPUTING GROWTH PATH
Upgrading the current mainframe is only one way to increase the University's computing capacity. Better alternatives exist. For example, in the last few months the University Computer Committee has put together numerous proposals designed to establish microcomputer laboratories for instructional computing. All of these proposals have been postponed indefinitely based on uncertainties dealing with administrative issues, such as location, budgeting, and ownership. On a substantive level, it is undeniable that microcomputers provide more computing per dollar and offer better software for most instructional purposes. Instructional computing demand is quite "peaky" in nature, so removing it from the mainframe would improve the mainframe's response time for all other users.
Oregon State University needs better mainframe computers that provide a wider range of applications software and work well in a distributed processing environment. This probably will require using machines such as an IBM 4300 or a number of networked DEC VAXs. Rather than upgrade the current mainframe, we should begin the process of converting to a new and better computing environment. Upgrading the current mainframe will only make the conversion process more painful later.
In order for OSU to enjoy the benefits of technological change, we must have a university-wide master plan to provide better access to modern computer equipment and software. Last year's STRATEGIC PLAN FOR COMPUTING identified many of our long-term need; what we need now is a plan of action. If the plan is to succeed, it must be enthusiastically backed at all levels of the university. We need to keep OSU on the forefront of the information age.
Curt Cook Sheila Cordray Ted Lewis George Martin
Computer Science Sociology Computer Science Accounting
Jim Richman Michael Schuyler David Sullivan Barry Shane
Oceanography Chemistry Business Management
PLANNING FOR MAINFRAME PROCESSING IN
A DISTRIBUTED COMPUTING ENVIRONMENT
May 13, 1985
The Faculty Forum offering entitled THE EVOLUTIONOF COMPUTING AT OREGON STATE UNIVERSITY asks, "Where does the PLAN FOR MAINFRAME COMPUTING fit within the University's plans for personal computing, networking, printing, and communications?" The answer is that advances in all of those areas (including mainframe computing) are proceeding in accordance with the recommendations set forth in the July 1984 report OREGON STATE UNIVERSITY STRATEGIC PLAN FOR COMPUTING. The report was produced by a committee appointed by Vice President Parsons, who directed the committee to "…help develop institutional goals and strategic plans for computing."
This year's accomplishments have been directed toward achievement of an effective distributed computing environment. Examples of the accomplishments include,
Personal Computing - more than 800 microcomputers have been acquired through the Faculty PC Program and the Resale Program.
Networking - the first phase of a broadband local area network (LAN) has been installed. Subscription to the BITNET international network has been initiated.
Printing - dial-up service has been added by the Department of Printing. Use of the Computer Center's Laser Printer has expanded.
Communications - higher speed data communications services have been provided for the Cyber. The LAN offers a 4-8 fold increased upper limit for intercomputer file transfers. DATALINK communications software has been refined.
Mainframe - the number of users, applications software field size limit, available disk space, and the speed of data communications have all been increased. A full-screen editor has been implemented.
The mainframe plan proposed for the next biennium advocates continued emphasis on enhancement of distributed computing by acquiring another mainframe which has outstanding applications software and will interface well with the hundreds of work stations on campus. The plan also includes a cpu upgrade to the Cyber 170/720, which is now overloaded, partially as a result of the aforementioned mainframe changes. Those changes have added significantly to the overhead 'cost' borne by the cpu; as more micro-to-mainframe communications traffic occurs, the cpu will become saturated. Deferring the cpu upgrade would introduce an unacceptable level of risk of mainframe failure.
Thomas L. Yates
Milne Computer Center.
Opinions expressed by authors of Faculty Forum articles are not necessarily those of the OSU Faculty or Faculty Senate.