The Next Generation Internet: A white paper ----- Begin cover page --------- Name & Affiliation of authors: 1. Dr. S. Sitharama Iyengar, IEEE Fellow Chairman, Dept. of Computer Science Louisiana State University iyengar@bit.csc.lsu.edu 2. Sundararajan Vedantham (author for contact) Dept. of Computer Science 105 Peabody Hall Louisiana State University Baton Rouge, LA 70803 sundar@bit.csc.lsu.edu Phone: (504) 388-5099 Fax : (504) 388-2267 3. Dr. Ramana Rao Los Alamos National Lab. rao@lanl.gov 4. Dr. Amit Anil Nanavati Netscape Communications Corp. amit@netscape.com (Ideas & opinions expressed here are of the authors and not of the Netscape Communications Corporation.) End of cover page. -------------------------------------------------------------------------- Abstract: We identify and briefly discuss key issues that must be addressed, at the design stage, for the next generation of the Internet. This list is not exhaustive; moreover, further sub-issues may arise during the resolution of these key ideas. 1. Introduction The primary goal of the NGI initiative is to connect educational and research institutions with high speed networks. If the current Internet is any indication, NGI networking will eventually trickle down to the household level. Therefore, the NGI design team should study the transition of the current Internet so that the NGI might make such a transition more gracefully. 2. Performance The NGI initiative must invest R&D effort in speeding up data communication channels both at the network as well as the peripheral equipment levels. Computer processor performance approximately doubles every year; however, the performances of memory and peripherals (such as buses) have not kept pace with this increase. Computer buses today typically achieve only about a 1 Gigabit/second transfer rate while the current internet works at a modest 10 Megabit/second transfer rate or less. If the communication cost is brought down considerably, the entire network can be operated as an enormous bus interconnecting a large number of processors. Such a model would have phenomenal impact on computation intensive processes like climate modeling and space/nuclear simulations. 3. The Utility Model: Design from a user's perspective The NGI should be configured much like the electrical power grid, or the water supply system. For example, a person moving from New York to San Jose today can plug-in a table lamp or a coffee maker into the electrical outlet in the new apartment without the power company being informed about the specifics of this appliance. Clearly, getting reconnected to the Internet is not such a simple process. Therefore, one of the fundamental design philosophies of the NGI initiative must be the seamless incorporation of internet novices and the elimination/cleaning up of the patch work of LANs, WANs and NAPs that makeup the internet today. In addition, a standardized, easy-to- connect port design (replacing telephone jacks, ethernet, token ring, optical cable connectors, etc..) that can help the plug-and-play style of operation is highly desirable. 4. Addressing Schemes The method of assigning addresses to machines on the Internet must be flexible enough to support the projected user base several years in the future. Consider the unix file system designed in the late '60s. At a time when hard disk space rarely exceeded a hundred megabytes on the average computer, the file system supported addressability for a file with a trillion bytes of data. The NGI initiative must use comparable foresight to avoid the address crunch that exists on the internet today. Ideally, of course, there should be no upper limit on the number of addresses that can exist simultaneously on the Internet. Asynchronous Transfer Mode (ATM) networks use 20 byte long addresses (as opposed to 4 bytes used in IP) for initial and end point identification and to establish a virtual circuit (VC) to carry traffic between these two points. Once the VC is established, intermediate routers maintain relevant path information for transmission, while the cells carry the actual traffic, using only a 4-byte header. Such an approach eliminates the need for transmitting a long address repeatedly. 5. Bandwidth Traffic on the Internet will soon include audio and video data on a much larger scale. Hence the system must support services such as high quality video transmission (e.g. HDTV) in multicast situations with specified Quality of Service (QoS) guarantees. The demand for voice bandwidth in the telecommunications industry increases at about 3% every year while the demand for data bandwidth is increasing at a whopping 20% annually. With the telephone industry switching from copper to fiber optic cabling, the requirements for voice transmission are not likely to fill the available bandwidth over the next several years. This surplus bandwidth can now be utilized to handle the anticipated increase in data bandwidth in the future. In fact, it isn't hard to imagine a situation in the not too distant future where telephones are made obsolete by intelligent computers using fast networks for ferrying audio and video data. But currently, the reliability of the data network is not comparable to that of telephone networks. To facilitate this integration among different networks, the NGI must increase the reliability of current data networks at least to the level of telephone networks. One of the challenges of the NGI initiative will be to standardize a networking protocol that is easy to implement and yet capable of supporting the increasing number of services future networks will provide. TCP/IP, one of many internetworking protocols (among others such as OSI and SNA), is perhaps the most widely used over the Internet today. However, TCP/IP makes no performance guarantees. Streaming audio and video data in the future would require much higher transfer rates, with definite QoS guarantees. Thus, any protocol adopted as the standard by the NGI must be able to make transfer rate guarantees. An example of a need for such a guarantee would be in a video stream of size 512 x 512 pixels, showing a 32 bit color movie, running at 30 frames per second. This would require a transfer rate of about 1/4 Gb/s. Failure to achieve this would result in poor quality video, unacceptable to a user used to cable television. ATM networks address these requirements. 6. Adaptability We currently have a telephone network, cable TV network, and a computer network that are mutually incompatible. A serious disadvantage of this scenario is that free bandwidth available on one network cannot be used to re-route traffic from another network. The NGI initiative should ensure that technological advancements in any one area can improve the overall effectiveness of the network. This goal can be achieved by adopting a technology that treats different types of net traffic uniformly both in concept and implementation. ATM technology, for example, packetizes all data (video/audio/file transfer) into "cells" before transmission, which are reassembled at the receiving end. This scheme has the advantage that technological advancements in any area (in a video compression algorithm, for instance) helps all kinds of traffic. 7. Conclusion We have identified five key issues -- performance, utility, addressing, bandwidth and adaptability -- which are most important from a design perspective. References [1] Craig Partridge, Gigabit Networking. Addison-Wesley 1994. [2] Doug Comer, Internetworking with TCP/IP. Prentice Hall 1991. [3] A.S. Tanenbaum, Computer Networks. Prentice Hall 1988.