IEC Proposal - Attachment 1

ATTACHMENT 1: Sample Initial Topic Areas

Infrastructure Support

Routing and Router Testing. The Internet currently supports only the simplest routing functionality, and the resulting inflexibility has been a problem for network administrators. How to best support a range of quality of service, if at all, is still controversial. A strong contingent holds that the benefit of policy routing functionality would not justify either the initial layout or the operating cost. Network engineers must get a feel for how router designs scale with load, for example, and for what failure modes look like. This section will also deal with ways to design unbiased experiments for comparing design features.

Caching. Caching popular documents can help reduce Web traffic, but as with multicasting, deploying a globally realistic caching system requires serious attention to addressing, routing, and server performance, not to mention changing traffic dynamics. Caching cuts across all layers of the Internet architecture, in addition to the Web, and will become important to many layers. The repository will include supporting statistics methodologies and mechanisms to encourage their efficient and ethical use.

Security. There is an egregious lack of basic education about security on the Internet, which is becoming even more unpardonable as the level of integrity violations rises. Security basics must be taught on multiple levels, right across the curriculum, for audiences ranging from the most casual level of user through network and computer engineers.

Workload Characterization. As mentioned earlier, the fragility of the current network seems to increase as not only the volume but also the range of workloads increases. Network operators and builders have remarkably little understanding of what the traffic carried actually represents, in computational terms, leaving them with only one "engineering" response: increase bandwidth as fast as possible. Materials on traffic characterization and performance modeling would target the concept of effective bandwidth estimation under dynamic loads.

System administration. Many students graduate from computer science programs and go on to become system administrators; yet there are surprisingly few schools giving courses in system administration. Ethical system administration should be a basic requirement for undergraduate computer science students, and it should be presented again in the graduate curriculum.

Protocols of the Future

TCP Problems in the face of Workload Changes. The transport control protocol (TCP) layer of the Internet protocols lies on top of the IP layer and is tuned to acknowledgments from across the net that packets have been received, after which both the acknowledgment and the packet can be discarded. This enables the layer to resend unacknowledged packets, which is very important for text transmission, but far less so for audio or video, where the odd dropped packet does not significantly affect the material received. To avoid suboptimalities in future application protocols, a clear understanding of TCP dynamics and philosophy is needed.

Bandwidth and Resource Reservation, Quality of Service. Closely related to the need for more sophisticated routing is the capability to reserve network resources in order to secure service of a certain quality for a certain time period. The repository will contain materials on motivations, design principles, and example implementations of resource reservation protocols.

The http Protocol and Successors. The hypertext protocol and markup language require development to make new capabilities available and to avoid multiple, inefficient schemes of storage and markup conversion. Ideal solutions will bring together the expertise developed in the realms of booksetting, word processing, and typographic animation to yield versatility with the least redundancy.

Multicast. The Internet multicast backbone (Mbone), much like the Internet itself, developed to a large degree without an architecture designed for scaling to a global operation. Designers did not foresee that the research experiment would so quickly become so globally useful and be in such demand. The resulting operation requires large-scale re-engineering to carry projected multimedia services and other high-volume data. We present materials on the design philosophy and implementations of the original Internet multicast protocols, and present day efforts such as PIM and CBT, hierarchical PIM and DVMRP.

DNS. The Domain Naming System (DNS) has also outgrown its initial design, with the explosion of the ".com" space rendering system scalability a dubious proposition. The path to take from here is the subject of considerable controversy, aggravated by political perceptions of the current fiscal arrangement for supporting the service. Although an outstanding initial design, the DNS has stagnated with respect to authentication, automatic identification of broken servers and clients, integration with PGP, geographical locator records, and other desirable dimensions of a naming system.

Applications

Multimedia. Optimal engineering for combination of text, pictures, sound, animation, video, and 3-D versions of pictures and animations (VRML, Java) involves debates over separate streaming of distinct modalities and a variety of recombination schemes. All of the possible modes create huge demands on bandwidth that cannot be accommodated yet even on the high-end, broadband service (vBNS), let alone throughout the commercial Internet.

MUDs and MOOs. The potential of these on-line meeting places as educational tools is clear. Nevertheless, under significant loads for considerable periods of time, multi-user spaces require increased sophistication and scalability in underlying architecture and software. Solutions in this area may have broad application to other networking issues.

Large Database Mining. Large databases on the Internet will be distributed across heterogeneous platforms and archives in geographically remote locations. Nonetheless, users will want to examine the data in one query or analysis pass. Facilitating such capabilities are the advent of "intelligent" search agents and the development (as at SDSC) of mass storage and highly parallel analysis tools for large data sets. Specialized switching and caching may be needed to respond to these stresses. High-capacity storage delivering massive bursts of information requires integration into an already "bursty" traffic pattern.

Social Issues

Students of network engineering need a fundamental understanding of the social issues of justice and equity that are part and parcel of the Internet considered as a new integument for broad-scale social interaction. Democratic values are strong on the Internet, and ways to scale and preserve those values must be consciously and conscientiously sought. Scale itself, indeed, is a similar metatopic, and such topics can easily be indexed to the rest of the curriculum given here as well as to sites where such topics may be explored in greater depth.

Code Development

Distributed Code Parallelization. Supporting applications on wide-area networks will require greater attention to basic applications program interfaces (APIs), ranging from socket services to wrappers for scientific applications. Scientists are already finding network-related difficulties in running high-performance applications across two or three supercomputing sites using PVM and MPI. Parallelism is not simply a means of speeding scientific code; indeed, its best application may well be in network-wide uses. Repository materials will also be solicited in parallel programming, object-oriented programming, and algorithms.

Imminent Technologies

Of course, today's imminent technologies are tomorrow's standards -- or tomorrow's relics. As topics become widely adopted or discarded, the coordinator would adjust the organization accordingly.

Cell switching and ATM. The latest switching technology has considerable impact on high-end applications, as well as on the higher-level protocols. Asynchronous transfer mode (ATM) switches are a proposed switching modality for optical signals, and the I-WAY demonstrated their advantages. As deployment of such switches expands, it will be important to understand the differences and similarities among ATM switches of differing design and target application and their ramifications for the higher-level protocols.

Pricing. As common carriers begin delivering network services, various schemes for pricing services have emerged. Criteria must be developed for acceptable tradeoffs between the load they impose on the system and the degree of cost recovery they support. Schemes should be equitable with respect to the prior investments of customers and the bandwidth allocated. The infrastructure needs policies to simplify and rationalize the pricing system. A rational system must also support pricing for multiple qualities of service from the same (multicasting) source. A rational scheme should be developed for the interaction of caches with charging models.

Wireless Technology. Demand is growing for smoother integration of fiber and wireless technologies and for solutions to all the problems attendant upon wireless transmission.