PRELIMINARY REPORT ON RESEARCH NETWORKING AND

COMPUTATIONAL NEEDS OF UTK



HAMPARSUM BOZDOGAN, CHAIR

Subcommittee on Research

Networking and Computational Needs

Research Council, UTK



Subcommittee Members

BOBBY BLEDSOE

RAY POPP





1. Charge



   O  To expeditiously explore the network and high performance

computing needs of the research enterprise.



   O  To communicate the results of the exploration to the full

council.



   O  To keep the Vice Chancellor for DIVISION OF INFORMATION     

INFRASTRUCTURE (DII) fully aware of the committee's activities,

progress, and conclusions.



2. Process Used



   As a preliminary step we conducted an informal, not a

"statistical", e-mail survey to acquire the input of a wide

spectrum of faculty, departments and units to explore their

networking and computational needs during the Fall Semester of

1995. Based on the responses we received, the sample data is a

good representative of cross-disciplinary areas. Therefore,   

this report is based on the inputs and responses received thus

far from different disciplines.



3. Research Networking and Computational Needs: Findings



   The quality and quantity of networking and computational needs

of researchers at UT varies dramatically from department to

department. The objective of this report is to identify and

report on the general areas of needs of the research enterprise

and make some recommendations. We classify the general research

networking and computational needs under the following

categories.



   a. Research Networking and Network Infrastructure Needs



      Use of the network provides connection between two

computers, the origin or user and a destination, service or host.

The campus network infrastructure begins at the user's wall

outlet and continues to the wall outlet of the destination system

or service. Therefore, network support ends where connection is

made to the host systems within the University or portions of

larger networks such as the Internet that are not under

University control.



      The general categories of the network activity are:



      (i)   administrative,

      (ii)  communication,

      (iii) classroom, and

      (iv)  research.



      The needs are:



      o  More dial-up lines and more gateways. With the

introduction of affordable high speed communications systems, in

the near future, the load experienced by our campus gateways will

seriously increase.



      o  Implementation of local network node to be used in the

automation of the classroom activity and teaching labs. Such

nodes will permit the exchange of process data and control

commands, between workstations and labs. As this type of

application requires a large, continuous, and rapid data flow, it

is usually configured as a separate node.  Departments or units

need to have the control of their individual system configuration

and their active network nodes.



      o  An affordable network management system to serve the

nodes and to act as a gateway.



      o  The development of a multi-platform based on the TCP/IP

network protocol. This model will allow local TCP/IP nodes to

function as autonomous agents, within a global TCP/IP based

network. The applications of this model is being targeted, for

example, for off site manufacturing facilities, and research

centers. These experimental sites will be linked via (site

dependent) high speed connections, to the internet. The

implementation of this model will integrate the process

automation community with the global communications network

(Internet) in engineering.



      o  The use of low cost, for example, Linux type network

hosts to provide network security, in a UNIX fashion. The use of

this model will also, allow LAN/WAN system administrators to

offload various duties and isolate sensitive nodes. Of course,

the development of this prototype system, will require

significant local resources.



      o  An improvement in the constancy of data thru-put, as it

sometimes slows to a crawl, and in overall bandwidth. An

affordable UNIX variant, supported on the campus might be

necessary, since there is a growing interest in network computing

across different departments.



      o  Clustered network and network infrastructure. 

Departments or disciplines, for example, the physical sciences

(such as Physics, Chemistry, and Geology) who frequently have

similar networking problems and computational demands, might need

a common network as a cluster. So based on the identifiable

similarity of different disciplines, a clustered network and

network infrastructure might be desirable.



          Efficient mechanism of accessing informational

databases (such as CRSP, COMPUSTAT, CD DISCLOSURE, DOW JONES

RETRIEVAL, IBES, etc.)used by researchers in the College of

Business Administration could be accessed through one access

point (e.g., a terminal) and processed via one mechanism (e.g.,

database server). As it currently stands each database uses its

own platform and is physically located in different areas of      

the University. It would be desirable to have all this

information be stored on a network server to be accessed and

processed through a PC environment, rather than accessing the

mainframe.



   b. Research Computing Needs



      Academic departments have three major missions:(i)

teaching, (ii) research, (iii) service and consulting (depending

on the respective disciplines).



      Computers are fundamentally changing the way all these

missions are accomplished. In 1993, an Instructional Technology 

Advisory Committee was formed and undertook the task of

developing recommendations for an instructional technology

initiative at UTK. However, we do not see an equivalent committee

on Research Computing Advisory Committee. This area seems to be

neglected and needs to be fostered. One can argue that perhaps    

the most essential justification for computational needs is for

research support. In instructional programs, the computer aided

education will substantially improve the quality of instruction

and enrich the learning environment, but most teaching duties can

be accomplished at some level without them. This is not the case

with research computing, where the infrastructure required to

meet research computing needs at UTK consists of:



      (i)   Bread and Butter Computing (B&B) (dubbed by Division

of Biology)which consists of distributed workstations in

client-server domains administered by individual departments or

colleges or by a central computing organization, and 



     (ii)  High Performance Desktop Computing (HPD).



     (iii) High Performance Parallel Supercomputing (HPPSC).

These consists of high performance computing resources, such as

powerful massively parallel supercomputers, that are beyond the

scope of what can be acquired and maintained by individual

departments or colleges.



      As we are now halfway through the second decade of

"personal computing" many desktop applications have matured to

the degree that they have become fundamental tools to scientist

and engineer in areas of imaging, design and composing of

typography, statistics and statistical modeling and forecasting,

automated manufacturing, controlling nuclear and thermal power

plants, modeling environmental, social and behavioral systems, to 

mention a few. Consequently, many mainframe applications have

been ported to the desktop.



      So the needs are:



      o  To establish a mechanism of  acquiring B&B Computing

platform for departments with very limited operating budgets.



      o  To establish a standard policy for the acquisition and

upgrading applications software. A University-wide licence would

be a very cost effective way to ensure that the many potential

units of UTK that would use a certain application software can do

so at low cost to any single department. Presently, there seems

to be the repurchasing of a site license for software already on

a site license by many departments which is not cost-effective.

As these applications software have become more widely available,

so as the diversity of the problems which they solve. As a result

something more than a B&B is required. To this end and using the

B&B system as a foundation, 



      o   The HPD system needs high-end processors in the range

of 120MHz to 150MHz, PCI bus, 1.6Gb hard disk, 6X CDROM, 3.5"

diskette drive, 16-32MB DRAM, 17" color display, a 2MB VRAM video

controller,and a backup tape drive. Furthermore, the basic

software and network applications packages are also needed as

discussed above.



         For those users who require the highest level of desktop

performance, the HPD should to be modified to include a

symmetrical multiprocessing SMP) capability allowing the

utilization of from two to four processors. SMP operating

systems, such as NT or UNIX should be provided.



         Another alternative would include a UNIX system. One

should consider the just released Sun Ultra 1. This 64bit

processor running at 167MHz includes a wide and fast memory bus,

288 bits at 300 Mbits per second with a burst rate to 1.3Gbits

per second. In addition, this box includes a 2D/3D frame buffer

equipped with 15MB of 3DRAM. All this does not come cheap

however, at $28, 000.00 per copy. One needs ample justification

for these based on their research computing needs.



   c. Distributed, On-Site Technical Support Needs



      o   Of overriding importance for the continued operation of

research networking and computation is the need for on-site

technical support.  As we solve the hardware and applications

software problems and needs, technical support staff will be a

universal need. It is one which must be emphasized when a

department is considering its own computational facility. So a

on-site support mechanism must be established to properly support

the resources.



      o   While management and support of the network must be

centralized, a distributed approach using existing campus unit

technical groups can facilitate problem solving. This concept

would establish a "first responder" program to solve minor

problems with the network and act as liaison between specific

campus units and the central network management team. This

concept could result in management of the network in a proactive

fashion. It could facilitate the resolution of likely problems

before they become actual problems.



      o   Centralized provision of low demand but high value

resources such as math, statistics, and other technical areas of

expertise are needed.



4. Recommendations



     The quality and quantity of research computational needs and

resources available to researchers at UTK varies dramatically

from department to department, and even within the same

department. Therefore, the research networking and computational

needs of the UTK research enterprise seems to be

discipline-oriented with departments having very limited

resources for networking, hardware, and applications software

needs. We found out that currently, here at UTK, as we move from

one mainframe platform to another, and to distributed computing

platform, it seems that all the costs have been passed to the

departments. Many of the costs passed on are initially hidden.



     The cost of software purchases, the learning of new methods

of operation, the need to recompile, etc. are not factored in to

establish a healthy and responsive computational environment.

Therefore, the overriding questions emerge are: "Who is going to

pay these costs?","How will they be borne?"



     Based on the informal e-mail survey responses and our

findings, we arrived at a series of recommendations directed,

variously, at the University administration, especially to the

Office of Vice Chancellor for the DIVISION OF INFORMATION

INFRASTRUCTURE (DII), and at Departments and Colleges. These are

as follows.



  a.  Recommendations to the Office of Vice Chancellor for the

DIVISION OF INFORMATION INFRASTRUCTURE (DII)



      o  A portion of the University's research computing budget

should be directed toward the research networking and

computational needs in acquisition of hardware, software, and

network facilities of the research enterprise.



      o  A process similar to the distribution of the Incentive

Funds administered by the Office of Research Administration (ORA)

needs to be established at the VCDII level on a per proposal

basis from different disciplines, departments, etc.



      o  To oversee such a process, a Research Computing Advisory

Committee needs to be formed similar to the Instructional

Technology Advisory Committee to make recommendations to VCDII on

research networking and computational needs of the UTK

enterprise. We strongly recommend that such a committee should

contain a diversified group of researchers and administrators. It

needs to have in its membership individuals that use a computer

solely for library searches and e-mail to those that have complex

models and know what the research computing is about. 

Furthermore, it needs to contain individuals that understand and 

protect departmental funding to those desiring a centrally funded

system.



      o  A cost sharing mechanism needs to be established between

the portion of the University's research computing budget and how

much the departments and/or colleges can contribute to meet their

research networking and computational needs. This will establish

a distributed financial support mechanism. As the computer base

become more distributed, so should the support.



      o  A balance needs to be achieved between the needs of

Instructional Technology and the Research Technology at the VCDII

level. Current direction seems to favor the Instructional

Technology more. We view this a more balanced strategy to follow.



      o  Establish and implement a University-wide license

agreements with software venders to ensure cost-effective site

licencing and eliminate duplications at the departmental or unit

level.



      o  VCDII should play a role of identifying potential

sources of funding (Federal, and, possibly, state or local,

government agencies, industrial sponsors, private foundations)

for implementation of the development plan for the Instructional

and Research Technology and in approaching them.



      o  Establish and implement a University-wide policy in

helping the transition of research computing from VAX cluster and

the IBM machines to UNIX environment during the migration process

by providing assistance of compiling, for example, the FORTRAN,

PL1, and other programs of the faculty with no charge. This seems

to be one of the major frustrations of the faculty in having

their research agenda being delayed, not alone the hidden costs

created for the Departments which cannot be absorbed.



     o   Establish contractual agreements with more of the

reputable brands of PC manufacturers rather than going to no-name

brand of PC's. Study, cost-effectiveness of leasing strategy and

its pros and cons against purchasing PC's.



     o   Forums should be established where questions and answers

can be attained. These forums would create ongoing dialog between

faculty and students and those planning for our research needs.



  b. Recommendations to the Departments/Colleges



     We recommend:



     o  That the Departments and Colleges formulate a systematic

development plan for building their specific research networking

and computational needs, computing resources, including careful

internal records of computing costs by category.



     o  That the Departments and colleges should budget resources

for their research networking and computational needs and

programs as they relate to desktop networking and computing.



     o  That the Departments and Colleges seek advice from other

Departments and Colleges which have already developed their

resources to the level planned.



     o  That the development plan ensure access to facilities for

all of the Department's Faculty and Graduate Students.



     o  That the development plan place the highest priority on

best computing hardware, software, and networking solutions

available to meet the needs of the Department and individual

Faculty. This should also account for the connectivity of

activities between the Departments and with others around the

globe to establish a cost-effective research computational

environment.