Network Working Group Barry M. Leiner
Request for Comments: 1015 RIACS
July 1987
Implementation Plan for Interagency Research Internet
STATUS OF THIS MEMO
The RFCproposes an Interagency Research Internet as the natural
outgrowth of the current Internet. This is an "idea paper" and
discussion is strongly encouraged. Distribution of this memo is
unlimited.
OVERVIEW
Networking has become widespread in the scientific community, and
even more so in the computer science community. There are networks
being supported by a number of the Federal agencies interested in
scientific research, and many scientists throughout the country have
Access to one or more of these networks. Furthermore, there are many
resources (sUCh as supercomputers) that are accessible via these
networks.
While many of these networks are interconnected on an informal
basis, there is currently no consistent mechanism to allow sharing
of the networking resources. Recognizing this problem, the FCCSET
Committee on Very High Performance Computing formed a Network
Working Group. This group has recommended an administrative and
management structure for interconnecting the current and planned
agency networks supporting research. The structure is based on the
concept of a network of networks using standard networking
protocols.
This report elaborates on the earlier recommendation and provides an
implementation plan. It addresses three major areas; communications
infrastructure, user support, and ongoing research. A management and
administrative structure is recommended for each area, and a
budgetary estimate provided. A phased approach for implementation
is suggested that will quickly provide interconnection and lead to
the full performance and functionality as the required technologies
are developed and installed. While this report addresses the
interconnection of agency networks, and cooperation by certain
federal agencies, some discussion is presented of the possible role
that industry can play in support and use of such a network.
Work reported herein was supported by Cooperative Agreement NCC 2-
387 from the National Aeronautics and Space Adminstration (NASA) to
the Universities Space Research Association (USRA). This report was
prepared in response to a request from John Cavallini, Chairman of
the Networking Working Group of the FCCSET Committee on Very High
Performance Computing.
INTRODUCTION
Computer networks are critical in providing scientists access to
computing resources (such as supercomputers) and permitting computer
supported interaction between researchers. Several agencies,
recognizing this need, have established networks to provide the
needed communications infrastructure. The need for this
infrastructure, though, cuts across the various agencies. To that
end, the FCCSET Committee on Very High Performance Computing Network
Working Group has recommended the formation of an Interagency
Research Internet (IRI) [1].
The purpose of this report is to suggest an implementation plan for
such an IRI. It addresses three major areas; communications
infrastructure, user support, and ongoing research. A management and
administrative structure is recommended for each area, and a
budgetary estimate provided. A phased approach for implementation is
suggested that will quickly provide interconnection and lead to the
full performance and functionality as the required technologies are
developed and installed. Finally, some discussion is presented on a
possible role for industry in supporting and using such a network.
Motivation
The prime responsibility for providing the required infrastructure
for successful research lies with the researcher, his/her
institution, and the agency supporting that research. Thus, the
individual agencies have installed and are continuing to enhance
computer networks to allow their researchers to access advanced
computing resources such as supercomputers as well as being able to
communicate with each other via such facilities as electronic mail.
However, there are a number of reasons why it is advantageous to
interconnect the various agency networks in a coherent manner so as
to provide a common "virtual" network supporting research.
The need to make effective use of available networks without
unnecessary duplication. The agencies each support researchers in
many parts of the country, and have installed equally widespread
resources. Often, it is more effective for a scientist to be
provided networking service through a different agency network than
the one funding his research. For example, suppose several
scientists at an institution are already being funded by NASA and
are connected to a NASA supported network. Now a scientist at the
same institution but supported by NSF needs access to an NSF
supercomputer. It is much more effective to provide that
connectivity through an interconnection of NASA and NSF networks
than to establish another connection (to NSFnet) to the same
university.
The need to establish communication infrastructure to permit
scientists to access resources without regard to which network they
are connected but without violating access controls on either the
networks or the resources. A scientist may be supported by multiple
agencies, and therefore have access to resources provided by several
agencies. It is not cost-effective to have to provide a separate
network connection to the scientist for each of those agency
resources.
The need for a communications infrastructure to encourage
collaborative scientific research. One of the primary functions of a
computer network supporting science is the encouraging of
collaboration between researchers. Scientific disciplines typically
cut across many different agencies. Thus, support of this
collaboration should be without regard to agency affiliation or
support of the scientists involved.
The need for a cooperative research and development program to
evolve and enhance the IRI and its components where appropriate.
Scientific research is highly demanding of both the computing and
networking environment. To assure that these needs continue to be
met, it is necessary to continually advance the state of the art in
networking, and apply the results to the research networks. No
individual agency can afford to support the required research
alone, nor is it desirable to have inordinate duplication of
research.
Summary of previous report
These reasons led to the formation of the FCCSET Commitee on Very
High Performance Computing and its Network Working Group. This group
began in early 1985 to discuss the possibility of interconnecting
into a common networking facility the various agency networks
supporting scientific research. These discussions led to the report
issued earlier this year [1] recommending such an approach.
The report used the "Network of Networks" or Internet model of
interconnection. Using a standard set of protocols, the various
networks can be connected to provide a common set of user services
across heterogenous networks and heterogenous host computers [2,
3,4]. This approach is discussed further in the Background section
below.
The report goes on to recommend an administrative and management
structure that matches the technical approach. Each agency would
continue to manage and administer its individual networks. An
interagency body would provide direction to a selected organization
who would provide the management and operation of the
interconnections of the networks and the common user services
provided over the network. This selected organization would also
provide for coordination of research activities, needed
developments, and reflecting research community requirements into
the national and international standards activities.
Overview of Implementation Plan
The general structure of the proposed IRI is analogous to a federal
approach. Each of the agencies is responsible for operating its own
networks and satisfying its users' requirements. The IRI provides
the interconnecting infrastructure to permit the users on one
network to access resources or users on other networks. The IRI also
provides a set of standards and services which the individual
agencies, networks, and user communities can eXPloit in providing
capabilities to their individual users. The management structure,
likewise, provides a mechanism by which the individual agencies can
cooperate without interfering with the agencies' individual
authorities or responsibilities.
In this report, an implementation plan for the IRI is proposed.
First, some background is given of the previous efforts to provide
networks in support of research, and the genesis of those networks.
A description of the suggested approach to attaining an IRI is then
given. This description is divided into two sections; technical and
management. The technical approach consists of two components. First
is the provision of an underlying communications infrastructure;
i.e. a means for providing connectivity between the various
computers and workstations. Second is provision of the means for
users to make effective use of that infrastructure in support of
their research.
The management section elaborates on the suggestions made in the
FCCSET committee report. A structure is suggested that allows the
various agencies to cooperate in the operations, maintenance,
engineering, and research activities required for the IRI. This
structure also provides the necessary mechanisms for the scientific
research community to provide input with respect to requirements and
approaches.
Finally, a phased implementation plan is presented which would allow
the IRI to be put in place rapidly with modest funding. A budgetary
estimate is also provided.
BACKGROUND
The combination of packet switched computer networks,
internetworking to allow heterogeneous computers to communicate over
heterogeneous networks, the widespread use of local area networks,
and the availability of workstations and supercomputers has given
rise to the opportunity to provide greatly improved computing
capabilities to science and engineering. This is the major
motivation behind the IRI.
History of Research Network
The Defense Advanced Research Projects Agency (DARPA) developed the
concept of packet switching beginning in the mid 1960's. Beginning
with the Arpanet (the world's first packet switched network) [5],a
number of networks have been developed. These have included packet
satellite networks [6,7], packet radio networks [8,7], and local
area networks [9].
Although the original motivation for the Arpanet development was
computer resource sharing, it was apparent early on that a major use
of such networks would be for access to computer resources and
interaction between users [10]. Following the Arpanet development,
a number of other networks have been developed and used to provide
both of these functions [11]. CSNET was initiated to provide
communications between computer science researchers [12,13]. CSNET
was initiated by the NSF in cooperation with a number of
universities, but is now self-sufficient. Its subscribers include
universities throughout the world as well as industrial members
interested in interacting with computer scientists.
CSNET makes use of a number of networking technologies including the
Arpanet, public X.25 networks, and dial-up connections over phone
lines, to support electronic mail and other networking functions. In
addition to the basic data transport service, CSNET and Arpanet
operate network information centers which provide help to users of
the network as well as a number of services including a listing of
users with their mail addresses (white pages) and a repository where
relevant documents are stored and can be retrieved.
With the installation of supercomputers came the desire to provide
network access for researchers. One of the early networks to
provide this capability was MFEnet [11]. It was established in the
early 1970's to provide DOE-supported users access to
supercomputers, particularly a Cray 1 at Lawrence Livermore National
Labs. Because MFEnet was established prior to widespread adoption of
the TCP/IP protocol suite (to be discussed below), the MFEnet uses a
different set of protocols. However, interfaces have been developed
between the MFEnet and other networks, and a migration plan is
currently under development.
NASA Ames Research Center has long been in the forefront of using
advanced computers to support scientific research. The latest
computing facility, the Numerical Aerodynamic Simulator, uses a Cray
2 and other machines along with a number of networking technologies
to provide support to computational fluid dynamics researchers [14].
This system uses the TCP/IP protocol suite both locally and remotely
and provides easy access through advanced workstations.
Recognizing the importance of advanced computers in carrying out
scientific research, NSF in 1984 embarked on an ambitious program to
provide supercomputer access to researchers. This program involved
both the provision of supercomputers themselves (through purchase of
computer time initially, and establishment of supercomputer centers)
and provision of access to those supercomputers through an extensive
networking program, NSFnet [15]. The NSFnet uses a number of
existing networks (e.g. Arpanet, BITNET, MFEnet) and exploratory
networks interconnected using the TCP/IP protocol suite (discussed
below) to permit scientists widespread access to the supercomputer
centers and each other. The NSFnet is also taking advantage of the
widespread installation of campus and regional networks to achieve
this connectivity in a cost effective manner.
The above are only a small number of the current and existing
networks being used to support research. Quarterman [11] provides a
good synopsis of the networks currently in operation. It is obvious
from this that effective interconnection of the networks can provide
cost-efficient and reliable services.
Starting in the early 1970's, recognizing that the military had a
need to interconnect various networks (such as packet radio for
mobile operation with long-line networks like the Arpanet), DARPA
initiated the development of the internet technologies [16].
Beginning with the development of the protocols for interconnection
and reliable transport (TCP/IP), the program has developed methods
for providing electronic mail, remote login, file transfer and
similar functions between differing computers over dissimilar
networks [4,3]. Today, using that technology, thousands of
computers are able to communicate with each other over a "virtual
network" of approximately 200 networks using a common set of
protocols. The concepts developed are being used in the reference
model and protocols of the Open Systems Interconnection model being
developed by the International Standards Organization (ISO) [17].
This is becoming even more important with the widespread use of
local area networks. As institutions install their own networks,
and need to establish communications with computers at other sites,
it is important to have a common set of protocols and a means for
interconnecting the local networks to wide area networks.
Internet Model
The DARPA Internet system uses a naming and addressing protocol,
called the Internet Protocol (IP), to interconnect networks into a
single virtual network. Figure 1 shows the interconnection of a
variety of networks into the Internet system. The naming and
addressing structure allows any computer on any network to address
in a uniform manner any computer on any other network. Special
processors, called Gateways, are installed at the interfaces between
two or more networks and provide both routing amongst the various
networks as well as the appropriate translation from internet
addresses to the address required for the attached networks. Thus,
packets of data can flow between computers on the internet.
Because of the possiblity of packet loss or errors, the Transmission
Control Protocol (TCP) is used above the IP to provide for
reliability and sequencing. TCP together with IP and the various
networks and gateways then provides for reliable and ordered
delivery of data between computers. A variety of functions can use
this connection to provide service to the users. A summary of the
functions provided by the current internet system is given in [4].
To assure interoperability between military users of the system, the
Office of the Secretary of Defense mandated the use of the TCP/IP
protocol suite wherever there is a need for interoperable packet
switched communications. This led to the standardization of the
protocols [18, 19, 20, 21, 22].
+---+ +---+ +---+ +---+ +---+ +---+
FS SC SC SC SC SC
+-+-+ +-+-+ +-+-+ +-+-+ +-+-+ +-+-+
--+-------+-----+-----+-------+--LAN-- --+------+-+---+----LAN--
+-+--+ +-+--+ +-+--+ +-+--+
WS WS WS WS
+-+--+ +-+--+ +-+--+ +-+--+
+-+-+ +-+-+
G G
+-+-+ +-+-+
+--------------+ +--------------+
Agency +-+-+ Agency
Network -- G -- Network
+------+-------+ +---+ +------+-----+-+
+-+-+ +-+-+
G G
+-+-+ +-+-+
/ / +-------+
/ / TS
/ / +-+-----+
+--------------+ +--------------+ ...
Regional Commercial T T T
Network Network
+-----+--------+ +------+-------+
+-+-+
G
+-+-+
+-+-+
H
+---+
----+------+-----+-----+------LAN----
+-+--+ +-+--+ +-+--+ +-+--+ +-------------------------+
WS WS WS WS H - Host
+-+--+ +-+--+ +-+--+ +-+--+ WS - Workstation
SC - Supercomputer
TS - Terminal Server
FS - File Server
G - Gateway
+-------------------------+
Figure 1: Internet System
Thus, the TCP/IP protocol suite and associated mechanisms (e.g.
gateways) provides a way to interconnect heterogeneous computers on
heterogenous networks. Routing and addressing functions are taken
care of automatically and transparently to the users.The ISO is
currently developing a set of standards for interconnection which
are very similar in function to the DARPA developed technologies.
Although ISO is making great strides, and the National Bureau of
Standards is working with a set of manufacturers to develop and
demonstrate these standards, the TCP/IP protocol suite still
represents the most available and tested technology for
interconnection of computers and networks. It is for that reason
that several agencies/programs, including the Department of Defense,
NSF and NASA/NAS, have all adopted the TCP/IP suite as the most
viable set of standards currently. As the international standards
mature, and products supporting them appear, it can be expected that
the various networks will switch to using those standards.
TECHNICAL APPROACH
The Internet technology described above provides the basis for
interconnection of the various agency networks. The means to
interconnect must satisfy a number of constraints if it is to be
viable in a multi-agency environment.
Each agency must retain control of its own networks. Networks have
been established to support agency-specific missions as well as
general computer communications within the agency and its
contractors. To assure that these missions continue to be supported
appropriately, as well as assure appropriate accountability for the
network operation, the mechanism for interconnection must not
prevent the agencies from retaining control over their individual
networks.
This is not to say that agencies may not choose to have their
individual networks operated by the IRI, or even turned over to the
IRI if they determine that to be appropriate.
Appropriate access control, privacy, and accounting mechanisms must
be incorporated. This includes access control to data, resources,
and the networks themselves, privacy of user data, and accounting
mechanisms to support both cost allocation and cost auditing [23].
The technical and adminstrative approach must allow (indeed
encourage) the incorporation of evolving technologies. In
particular, the network must evolve towards provision of high
bandwidth, type of service routing, and other advanced techniques to
allow effective use of new computing technology in a distributed
research environment.
Communications Infrastructure
The communications infrastructure provides connectivity between user
machines, workstations, and centralized resources such as
supercomputers and database machines. This roughly corresponds to
communications services at and below the transport layer in the ISO
OSI reference model. There are two different types of networks. The
first are local networks, meaning those which are internal to a
facility, campus, etc. The second are networks which provide transit
service between facilities. These transit networks can connect
directly to computers, but are evolving in a direction of connecting
local networks. The networks supported by the individual agencies
directly are mainly in the category of transit (or long-haul)
networks, as they typically provide nationwide connectivity, and
usually leave communications within a facility to be dealt with by
the facility itself. The IRI communications infrastructure thus
deals mainly with the interconnection of transit networks.
The internet model described above provides a simple method for
interconnecting transit networks (as well as local networks.) By
using IP gateways between the agency networks, packet transport
service can be provided between computers on any of the various
networks. The placement of the gateways and their capacity will have
to be determined by an initial engineering study. In addition, as
the IRI evolves, it may be cost-effective to install one or more
wide area networks (or designate certain existing ones) to be IRI
transit networks, to be used by all agencies on a cost sharing
basis. Thus, the IRI communications infrastructure would consist of
the interconnecting gateways plus any networks used specifically as
transit networks. Using IP as the standard for interconnection of
networks and global addressing provides a common virtual network
packet transport service, upon which can be built various other
network services such as file transfer and electronic mail. This
will allow sharing of the communication facilities (channels,
satellites, etc.) between the various user/agency communities in a
cost effective manner.
To assure widespread interconnectivity, it is important that
standards be adopted for use in the IRI and the various computers
connected to it. These standards need to cover not only the packet
transport capability but must address all the services required for
networking in a scientific domain, including but not limited to file
transfer, remote login, and electronic mail. Ultimately it is
desirable to move towards a single set of standards for the various
common services, and the logical choice for those standards are
those being developed in the international commercial community
(i.e. the ISO standards). However, many of the scientific networks
today use one or more of a small number of different standards; in
particular the TCP/IP protocol suite mentioned above, the MFEnet
protocols, and DECNET. As the international standards mature, it is
expected that the number of communities using the same protocol
suite will grow [5] [6]. Even today, several of the
agencies/communities are using a common protocol suite, namely the
TCP/IP suite. All the users connected to those computers and
networks are able to have the full functions of an interoperable
networking capability. And therefore the ability of the users to
share resources and results will increase.
User Services
In order that scientists can effectively use the network, there
needs to be a user support organization. To maximize the cost
effectiveness of the overall IRI, the local user support personnel
must be used effectively. In particular, it is anticipated that
direct support of users/researchers would be provided by local
support personnel. The IRI user support organization would provide
support to those local support personnel in areas where nationwide
common service is cost effective.
In particular, the this organization has several functions: assist
the local support personnel in the installation of facilities
compatible with the IRI, provide references to standard facilities
(e.g. networking interfaces, mail software) to the local support
personnel, answer questions that local personnel are not able to
answer, aid in the provision of specific user community services,
e.g. database of relevance to specific scientific domain.
Internet Research Coordination
To evolve internet to satisfy new scientific requirements and make
use of new technology, research is required in several areas. These
include high speed networking, type of service routing, new end to
end protocols, and congestion control. The IRI organizational
structure can assist in identifying areas of research where the
various agencies have a common interest in supporting in order to
evolve the network, and then assist in the coordination of that
research.
MANAGEMENT APPROACH
A management approach is required that will allow each agency to
retain control of its own networking assets while sharing certain
resources with users sponsored by other agencies. To accomplish
this, the following principles and constraints need to be followed.
IRI consists of the infrastructure to connect agency networks and
the user services required for effective use of the combined
networks and resources.
An organization must be identified to be responsible for the
engineering, operation, and maintenance of both the interconnecting
infrastructure and the user services support.
While some agencies may choose to make use of IRI facilities and
contractors to manage their individual agency networks, this would
not be required and is not anticipated to be the normal situation.
Any such arrangement would have to be negotiated individually and
directly between the agency and the IRI operations organization.
Normally, the IRI organization would neither manage the individual
agency networks nor have any jurisdiction within such networks.
Gateways that interconnect the agency networks as well as any long-
haul networks put in place specifically as jointly supported transit
networks (if any such networks are required) will be managed and
operated under the IRI organization.
A support organization for common IRI services is required. The
principal clients for these services would be the local support
personnel.
The IRI structure should support the coordination of the individual
research activities required for evolution and enhancement of the
IRI.
General Management Structure
Figure 2 shows the basic management structure for the IRI. It is
based on the use of a non-profit organization (call it the
Interagency Research Internet Organization, IRIO) to manage both the
communications infrastructure and user support. The IRIO contracts
for the engineering, development, operations, and maintenance of
those services with various commercial and other organizations. It
would be responsible for providing technical and administrative
management of the contractors providing these functions. Having the
IRI operational management provided by an independent non-profit
organization skilled in the area of computer networking will permit
the flexibility required to deal with the evolving and changing
demands of scientific networking in a cost-effective manner.
Direction and guidance for the IRIO will be provided by a Policy
Board consisting of representatives from the Government agencies who
are funding the IRI. The Chairman of the Board will be selected from
the agency representatives on a rotating basis. The Board will also
have an Executive Director to provide administrative and other
support. To provide effective support for the IRI Policy Board as
well as assure appropriate coordination with the IRIO, the Executive
Director shall be the Director of the IRIO.
To assure that the IRI provides the best support possible to the
scientific research community, the Policy Board will be advised by a
Technical Advisory Board (TAB) consisting of representatives from
the network research and engineering community, the various networks
being interconnected with the IRI, and the scientific user
community. Members of the TAB will be selected by the Policy Board.
The TAB will review the operational support of science being
provided by the IRI and suggest directions for improvement. The TAB
will interface directly with the IRIO to review the operational
status and plans for the future, and recommend to the Policy Board
any changes in priorities or directions.
Research activities related to the use and evolution of the internet
system will be coordinated by the Internet Research Activities Board
(IRAB). The IRAB consists of the chairmen of the research task
forces (see below) and has as ex-officio members technical r
representatives from the funding agencies and the IRIO. The
charter of the IRAB is to identify required directions for research
to improve the IRI, and recommend such directions to the funding
agencies. In addition, the IRAB will continually review ongoing
research activities and identify how they can be exploited to
improve the IRI.
The Research Task Forces will each be concerned with a particular
area/emphasis of research (e.g. end-to-end protocols, gateway
architectures, etc.). Members will be active researchers in the
field and the chairman an expert in the area with a broad
understanding of research both in that area and the general internet
(and its use for scientific research). The chairmen of the task
forces will be selected by IRAB, and thus the IRAB will be a self-
elected and governing organization representing the networking
research community. The chairmen will solicit the members of the
task force as volunteers.
+------+ +------+ +------+ +------+ .... +------+
DARPA NSF DOE NASA Others
+--+---+ +--+---+ +--+---+ +--+---+ +--+---+
+--+--------+-----------+----+------+-------------------+
+------------+
Funding Representatives Scientific
Research
V V Community
+-------------+ +-------------+ +----------+-+
Selecting Policy Policy
Contracting<-------------+ Board Advice
Agency +-> <------------+
+-----+-------+ +------+------+
Funding Management +------+<--+
Advice and Plans TAB <-------+
+---------------+ V +------+<---- +
+------------+ ^ ^
+---------------------> Interaction
IRIO <-----------+
+-------------> <-----------+
Interaction +-----+------+ Interaction
V
+-----------+----------+ +------------+
Management Funding Constituent
Networks
V V V V V +------------+
+-------+ +--------+ +--------+ +-----------+
IRAB Network User Other
+-------+ O&M Services Contractors
+----+---+ +---+----+ +-----+-----+
+---------+-------------+----------------+
+-----------------+--------------------+
Chair Chair Chair
V V V
+----------+ +----------+ +----------+
TASK FORCE TASK FORCE .... TASK FORCE
+----------+ +----------+ +----------+
^ ^ ^
V V V
+--------------------------------------+
Network Research Community ------------------+
+--------------------------------------+
Figure 2: IRI Management Structure
Funding
In this section, the funding of the IRI is described. Recall that
the IRI consists of the infrastructure to connect the agency
networks and the services required for users to make effective use
of such an infrastructure. These costs are divided into two
categories; operations costs and research costs. The operations
costs are those to operate and maintain both the communications
infrastructure and the user services. These costs must be shared
between the various agencies and channeled to the IRIO to operate
the IRI. The research costs are those used to carry out the needed
research to evolve the IRI. These costs are handled within the
various agency budgets and used to support research in each agency
with coordination between the agencies.
Operations Cost
Each participating agency will contribute a share of operations cost
of IRI. Initially, each agency will contribute an equal share.
Later, perhaps, the agency contributions will be adjusted according
to a number of factors such as number of users, amount of traffic,
type of support required (high bandwidth real time versus low
bandwidth mail for example).
To facilitate the funding and administration of the IRI, one agency
will be selected to manage the contract with IRIO. All funds will
flow through that agency to the IRIO via interagency transfer. The
role of the selected agency would be to provide the needed
contractual activities and adminstrative management. Technical
guidance and monitoring of IRIO activities would be provided by the
IRI Policy Board.
It is not yet clear which Federal agency is best for this role. The
requirements for such an agency include the ability to deal flexibly
with the evolving requirements of the IRI, to deal with funding
flowing from the various agencies, and to deal flexibly with the
various agency technical representatives and incorporate their
recommendations into the contract as required. One of the first
activities required for the Policy Board would be to select an
appropriate funding agency.
All operations and maintenance funding for the IRI will flow through
the IRIO to selected contractors. This allows centralized management
of the operation of the IRI.
There are two major assumptions underlying the budgetary estimates
to follow. First of all, the IRIO should maintain a fairly low
profile with respect to the end users (i.e. the scientists and
researchers). That is, the users will interact directly with their
local support personnel. The IRIO will act as facilitator and
coordinator, and provide facilities, information and help services
to the local sites. This will allow the IRIO to remain relatively
small, as it will not need to deal directly with the thousands of
scientists/users.
Second, it is assumed that the operations budget supports the
interconnection of agency networks as well as transit networking
where required, but does not include costs of the individual agency
networks.
Appendix A provides details of the budgetary estimate. Table 1 gives
a summary. Note that the initial year has a higher expenditure of
capital equipment, reflecting the need to purchase both the gateways
needed for initial interconnection and the needed facilities to
provide the operation of the gateways and the user services.
Operations costs are expected to grow by inflation while the capital
costs should remain constant (decrease when inflation is considered)
as the IRI is stabilized.
Research Costs
In addition to the costs of operating and maintaining the
communications infrastructure and user services, funding must be
allocated to support an ongoing program of research to improve and
evolve the IRI.
While each agency funds its own research program, the intent is that
the various programs are coordinated through the IRI Policy Board.
Likewise, while it is not intended that funds shall be combined or
joint funding of projects is required, such joint activity can be
done on an individual arrangement basis.
Each agency agrees, as part of the joint IRI activity, to fund an
appropriate level of networking research in areas applicable to IRI
evolution. The total funding required is currently estimated to be
four million dollars in FY87, growing by inflation in the outyears.
Details of this budgetary estimate are provided in Appendix A.
+--------------------------------------------------+
Table 1
Annual IRI Operations Budget
+----------+-------------+------------+------------+
Fiscal Capital O & M Total
Year Cost Cost
($M) ($M) ($M)
+----------+-------------+------------+------------+
1987 2 8 10
+----------+-------------+------------+------------+
1988 1 9 10
+----------+-------------+------------+------------+
1989 1 10 11
+--------------------------------------------------+
1990 1 11 12
+--------------------------------------------------+
1991 1 12 13
+--------------------------------------------------+
PHASED IMPLEMENTATION PLAN
The long-term goal of the IRI activity is to put in place a
functional high-performance network available to scientists across
the nation. To accomplish this goal, a steady evolution of
capability is envisioned. This phased approach involves both
technical and administrative ASPects.
Technical Phasing
Currently, networks are being supported by a number of agencies as
discussed in Section 2. Many are using the DoD protocol suite
(TCP/IP, etc.) and others have incorporated or are incorporating
mechanisms for interoperability with networks using the DoD protocol
suite (e.g. MFEnet). Most have discussed eventual evolution to ISO
protocols and beyond. By and large, most of these networks are
hooked together in some mainly ad hoc manner already, some by
pairwise arrangement and some through third party connections (e.g.
a university network connected to two agency networks).
There are two major shortcomings to this ad hoc connection, though.
Performance is not adequate for advanced scientific environments,
such as supercomputer usage, and community wide user support is not
generally available. The phased apprach described below will allow
these deficiencies to be overcome through coordinated action on the
part of the various funding agencies.
Phase I - Functional Interoperability
The initial stage of the IRI would provide for sharing of the
communications facilities (e.g. channels, satellites, etc.) by
interconnecting the networks using the Internet Protocol and IP
gateways. In addition, mechanisms will be installed (where required)
and maintained to allow interconnection of the common user services,
such as electronic mail. This will allow sharing of resources
attached to the network, such as supercomputers. [7] [8] Note:
actual use of facilities other than mail would require arrangements
with the various responsible parties for each host. For example, to
login to a host not only requires network access; it also requires a
login account on that host.
Specific steps to be undertaken in Phase I are the following:
Gateways will be purchased and installed where needed to
interconnect the agency networks. The location and performance of
these gateways will be specified by the IRIO and approved by the
Policy Board. This engineering will take into account an estimate of
current and future traffic requirements as well as existing
interconnecting gateways. It may also result in a recommendation
that some or all existing gateways between agency networks be
replaced with common hardware so that adequate management of the
interconnection can be achieved.
An IRI operations and management center will be established for the
interconnecting gateways. [9] [10] This perhaps could be done in
conjunction with a network management center for another set of
gateways, e.g. those supported by DARPA or NSF.
The requirement for application gateways or other techniques to
interconnect communities using different protocols will be
investigated and a recommendation made by the IRIO in conjuction
with the IRAB. The appropriate mechanisms will be installed by the
IRIO at the direction of the Policy Board.
An initial user services facility will be established. This facility
will provide at a minimum such services as a white pages of users
(similar to the current Internet "whois" service) and a means for
making accessible standard networking software.
The IRAB, in coordination with the Policy Board, will draft a
coordinated research plan for the development of the new
technologies required for evolution of the IRI.
Phase II - Full IRI Capability
Phase II will make the IRI fully functional with enhanced
capabilities and performance.
High performance gateways with appropriate new capabilities and
functions will be installed, replacing and/or augmenting the
gateways in place from Phase I. The functionality and performance
of these gateways will be specified based on the experience from
Phase I use, the anticipated new uses of the network, and the state
of the art technologies available as a result of the ongoing
research.
The basic user services facility will be mature and support network
operation. New capabilities will be developed to support specific
scientific communities (such as a data base of software used by a
specific community and its availability over the network.)
A high performance backbone network wil be installed if needed to
connect high performance agency networks. [11] [12] This is
anticipated because of the move in several agencies to provide high
bandwidth networks in support of such activities as supercomputer
access.
The introduction and use of international standards will be
investigated and a plan developed for providing more services to the
broad scientific community through use of these standards.
Administrative Phasing
The goal of the IRI is to get to a fully cooperating and managed
interagency research internet involving most if not all of the
agencies supporting scientific research. Recognizing that currently,
the major research networking players (both networking for research
and research in networking) are DOE, NASA, DARPA, and NSF, the
following steps are recommended:
The first and critical step is to establish a four agency Memorandum
of Agreement (MOA) to interconnect the agency networks and to share
the costs of interconnection, transit networks, and an operations
center. A management structure should be agreed upon as outlined
above. Agreement must also be reached on the need to fund an
ongoing research and engineering activity to evolve the internet.
A Policy Board and Technical Advisory Board should be established as
quickly as possible to assure appropriate guidance and direction.
The Policy Board shall then select an agency to handle the
administrative and contractual actions with the IRIO.
A non-profit organization shall then be selected by that agency
through an appropriate procurement mechanism to be the IRIO. The
Policy Board of the IRI shall be the selection panel.
The initial four agencies shall transfer the agreed upon funds to
the selected contracting agency on equal basis to start.
These funds will then allow the contracting agency to establish a
contract for the IRIO with the selected non-profit organization.
The IRIO can then establish sub-contracts for engineering,
procurement, installation, and management of gateways and operation
of the user services center.
To initiate the research coordination, the following steps will be
accomplished.
The Internet Activities Board will evolve into the Internet Research
Activities Board, through added membership and charter revision.
Additional task forces will be formed as needed to reflect the
expanded areas of research interest.
Once the IRI is established and operating, the funding and use of
the IRI will be reviewed to determine if equal funding is equitable.
If not, the IRIO should be tasked to develop a recommendation for a
practical cost allocation scheme. In addition, once the IRI has
proved itself to be successful, other agencies will join the IRI
and provide additional funding.
INDUSTRY ROLE
This report has thus far addressed the interconnection of agency
supported networks and the use of such an internet by agency
supported researchers. However, industry also has a need for a
similar infrastructure to support its research activities. [13]
[14]. Note that this refers only to industrial research activities.
It is not envisioned, nor would it be appropriate, for the IRI to
provide a communications system for normal industrial activities.
Regulatory concerns make it difficult for industry to connect to a
network that is supported by a federal agency in pursuit of the
agency mission.
The IRI structure above, though, may permit the connection of
industrial research organizations. Since the IRIO is a non-profit
non-government organization, it would be able to accept funds from
industry as a fair share of the costs of using the IRI. These funds
in turn can be used to expand the networking resources so that no
degradation of service is felt by the users suppported by the
federal agencies. This topic would need to be discussed further by
the Policy Board and the organization selected as the IRIO.
SUMMARY AND CONCLUSIONS
The interconnection of the various agency networks supporting
scientific research into an overall infrastructure in support of
such research represents an exciting opportunity. This report
recommends an approach and a specific set of actions that can
achieve that goal. It is hoped that, regardless of the mechanism
used, that the Federal agencies involved recognize the importance of
providing an appropriate national infrastructure in support of
scientific research and take action to make such an infrastructure a
reality.
ACKNOWLEDGEMENT
This report was prepared with advice and comments from a large
number of people, including the members of the FCCSET Committee
Network Working Group and the Internet Activities Board. Their
input is greatly appreciated, and I hope that this report represents
a consensus on both the need for the IRI and the proposed approach.
APPENDIX A - FUNDING BREAKDOWN
This appendix provides the details for the budgetary estimates of
Table 1.
Gateways
Gateways will be required between the various agency (and perhaps
regional) networks. As an upper bound, assume one IRI gateway per
state times $40K per gateway, spread out over two years, for a
capital cost of $1M per year for first two years.
Operation Center
The IRI operations center will have to engineer the location and
capacity of the gateways, as well as install, operate and maintain
them. It also will need to coordinate support and maintenance of
end-to-end service, helping to identify and correct problems in the
interconnections. Costs are estimated as two people round the clock
to man the operations center and three full time people to
coordinate, operate, and engineer the IRI. Using an estimate of
$120K (including other direct costs (ODC)) per year for an operator
and $200K per year for other activities, and translating 2 people
round the clock into 9 people results in a total annual cost of
$1.7M. In addition, equipment costs of roughly $500K per year can be
expected.
Transit Networks
It is expected that support of at least one transit network will be
necessary. This may involve reimbursement to one of the agencies for
use of their network, or may involve operations and maintenance of
an IRI dedicated network. An estimate for these costs, based on
historical data for operating the Arpanet, is $4M per year.
User Support Organization
To provide effective support as discussed above will require a staff
available during working hours. A reasonable estimate for the costs
of such an organization is 5 people times $200K per year, or $1M per
year (including ODC). In addition, there will be capital equipment
costs in the first two years totalling roughly $2M.
REFERENCES
1. FCCSET Committee on Very High Performance Computing Network
Working Group, Report on Interagency Networking for Research
Programs, February 1986.
2. Cerf, V.G. and P. Kirstein, "Issues in packet-network
interconnection," Proceedings of the IEEE, pp. 1386-1408,
November 1978
3. Cerf, V.G. and E. Cain, "The Dod intenet architecture model,
"Computer Networks, pp. 307-318, July 1983.
4. Leiner, B.M., J. Postel, R. Cole, and D. Mills, "The DARPA
internet protocol suite," IEEE communications Magazine
March 1985.
5. Defense Advanced Research Projects Agency, A History of the
Arpanet: The First Decade, Defense Advanced Research Projects
Agency, April 1981. (Defense Tech. Info. Center AD A1 15440)
6. Jacobs, I.M. et. al., "General purpose satellite networks,"
Proceedings of the IEEE pp. 1448-1467, November 1978
7. Tobagi, F., R. Binder, and B.M. Leiner, "Packet radio and
satellite networks," IEEE Communications Magazine, November
