NGI Research and Development Priorities

        Ray Bair, Pacific Northwest National Laboratory
       Stu Loken, Lawrence Berkeley National Laboratory
           Rick Stevens, Argonne National Laboratory

The Internet is essential to the Department of Energy in all aspects
of its programs.  DOE's nationwide computer network, ESnet,
interconnects the Energy Research community and its partners,
facilitating collaborations and providing scientists across the nation
and around the world with access to DOE computing, information, and
experimental facilities. ESnet also provides key elements of the
advanced infrastructure for DOE research programs in networking,
distributed computing, and electronic collaboration.

As DOE use of internetworking has grown, it has become clear that
present capabilities and approaches cannot simply be scaled up to meet
the needs of the department and the nation.  The Next Generation
Internet initiative must provide a new generation of communication
capabilities that support much tighter integration of projects and
facilities, that enable diverse and geographically separated teams to
address complex issues, and that offer location-independent access to
advanced computational resources--including terabyte and petabyte data
archives.

The priorities for NGI research and development within DOE are driven
by three major issues: (1) the need to support large-scale and widely
distributed science and technology collaborations, (2) the need to
build very large, distributed applications that make use of
supercomputers, research facilities, and databases, and (3) the need
to use a shared network infrastructure to support its programs. These
requirements, in turn, define key elements of an NGI R&D agenda,
including work on a new generation of network protocols and services
and creation of the middleware to support distributed applications and
collaborative work.

Collaboratories

By exploiting the capabilities of the NGI, laboratories without walls
(or collaboratories) have the potential to revolutionize how resources
and expertise are combined and applied to research projects.
Collaboratory applications place heavy performance demands on
underlying network bandwidth, reliability, latency, number of
receivers, etc. To meet these demands will require (a) protocols for
high-bandwidth, low-latency connectivity, (b) a quality of service
(QoS) management infrastructure that allows application QoS
requirements to be translated into appropriate scheduling operations
at the network level, and (c) support for reliable multicast, so that
data generated by a single sender can be propagated reliably to
multiple receivers. Without these capabilities, emerging collaboratory
applications will not obtain acceptable performance from networks.

Protocols: The rapidly growing congestion from increasing Web use
confirms that a significant problem exists with the current protocol
model. This model has very poor scaling properties and will not meet
the needs of the NGI. What is needed is a new model that incorporates
caching and can provide data from whoever has it. An approach, based
on Application-Level Framing (ALF), will provide a robust distributed
memory that supports Web caching. Also, the concept of a Universal
Resource Locator (URL) will change Internet references from a location
to the name of the desired entity.

Quality of Service: Effective use of NGI resources requires multiple
levels of service and network-wide agreements on the use of shared
resources. Abstractions are needed for organizational priorities.  An
important component is the development of a bandwidth broker
application, which will function in an administrative hierarchy that
can directly manipulate the bandwidth allocation machinery to provide
sustained bandwidth to an application.

Reliable Multicast: The use of multicast has become a critical part of
the support for large-scale collaborative work.  In many applications,
however, multicast must be extended to ensure that all participants
have received the same data or views.  Multicast reliability must be
implemented over the full range of applications and should accommodate
delivery from multiple sources.

Facilities on Line

To manage shared resources like experimental facilities and computers,
as well as the network itself, a decentralized mechanism is needed for
user authentication to ensure that the use of resources conforms to
existing agreements. Therefore, a critical component of a NGI is the
development and deployment of a scalable, decentralized authority with
generalized public-key certificate security architecture for
authentication, authorization, integrity, privacy, remote instrument
resource access control, and security context establishment for
communication services for distributed computation. Other aspects
include IPv6 security and key management, certification and validation
of services through trusted service agents, and integration of
certification authority and public-key infrastructures.

High-Performance Transport

In remote experiments, metacomputing, and collaborative engineering,
datastreams produced by instruments must be transported rapidly and
reliably to local and remote storage systems. In metacomputing
applications, researchers must be able to communicate and synchronize
activities.  Most existing standards such as DCE and CORBA do not
focus on high-performance communication and do not support specialized
communication structures commonly used in scientific programs, such as
message passing, collective operations, and stream-based
transfers. Systems that do focus on high performance, such as MPI and
active messages, are not broadly available in heterogeneous
environments. The need for message security and integrity introduces
additional performance demands.

Resource Location and Allocation

Widespread deployment and use of the NGI will require tools for
dynamically locating, characterizing, scheduling, controlling, and
monitoring resources used by computing, information and collaboratory
applications. These resources will have widely varying characteristics
and interfaces. Currently, applications deal with heterogeneity
poorly---using a fixed set of resources, scheduling from a predefined
set (as in DCE), or operating on a best effort basis (e.g.,
networks). By defining and deploying standard mechanisms for accessing
information about resources and manipulating resources, we can enhance
both portability and performance.

New-Generation Middleware

Interoperability, portability, extensibility, and code reusability are
central concerns for NGI applications. A new generation of middleware
is needed.  For example, one DOE program is addressing these issues by
producing a collaboratory software bus based on CORBA and Java
technologies.  Existing CORBA and Java services are being extended to
incorporate new services required for collaboratory applications, such
as multicast, high-performance transport, and asynchronous
invocation. A streamlined public domain CORBA system will be deployed
to enable widespread access to CORBA technology, and a set of template
applications that use these services will guide application
developers.