Addressing Next-Generation Internet (NGI) Technology Issues Today Prepared by: Michael P. Jones Engineering Manager Houston Associates, Inc. 4601 North Fairfax Drive Arlington, VA 22203 703-284-8850 703-812-5099 mjones@hai-net.com Introduction/ Overview While the Internet has revolutionized to the way many businesses, organizations and individuals communicate and share information, there remains a great deal more potential for distributed networking technology to allow people to share information. Much of the new development on the Internet has involved multimedia applications that integrate voice, video, imagery and data. We are beginning to see this technology demonstrated through cybercasting of live events via RealAudio, or intercasting of news feeds and stock quotes via PointCast. While interesting and entertaining capabilities, these technologies are limited in there ability to support very numbers of large multicast group because of limitations with unicast IP protocols, insufficient bandwidth and lack of "smart" applications. This brief white paper discusses some of the capabilities that the Next- Generation Internet (NGI) offers, some enabling technology under research and development by the the DARPA/DISA Joint Program Office (DD JPO) that could help make this a reality, and issues related to commercial utility of these technologies. NGI Enabling Technology If the internetworking technology is an become integral to the competiveness of business and industry, then there are at least three areas of "enabling" technology that must be addressed. They are: · Multicasting · Resource Reservation (RSVP) · Quality of Service for IP and ATM Multicasting Filtering As multicast traffic increases on LANs for distributed simulation, conferencing and collaboration across the Internet, there is a need for multicast filtering of local traffic received by individual hosts as they participate in these activities. This presents a problem for the increasingly popular switched media hubs. Most switched media hubs operate at layer two and do not "listen" to Internet Group Management Protocol (IGMP) messages. They typically cannot discern which multicast groups their host computers belong, therefore these switches flood multicast traffic out all ports. This effectively defeats the purpose of multicasting. Two of the commercial solutions to address this problem are: 1) multicast filtering at the network interface card (NIC) or 2) filtering on the switched hub. With high performance host network interface cards, filtering of unwanted LAN multicast traffic would be performed in the interface card hardware without imposing a processing burden on the host processor. Today, Ethernet interface cards provide hardware filtering facilities on the order of 64 groups. This is probably sufficient today, however future requirements will be driven by the degree to which high-end multicast applications proliferate in the Next-Generation Internet. Alternatively, multicast-aware hubs ensure that attached hosts would receive only network traffic destined for either the hosts' individual (unicast) network addresses or for IP multicast addresses. These multicast aware hubs either listen to IGMP directly or get IGMP-like messages from a collocated IGMP router. Resource Reservation (RSVP) Resource reservation refers to the ability for end users or applications to request membership in a multicast group with a prescribed set of flow specifications (latency, bandwidth and jitter) from the network. The flow specification in effect provides best-effort "guarantee" for a certain quality of service. Nonetheless, despite the flow specification, by definition connectionless UDP/IP does not guarantee delivery and connection- oriented TCP/IP does not guarantee throughput within the packet- switching network. The Real-time Transfer Protocol addresses the issue of guaranteed delivery over TCP. While RSVP can support unicast and multicast traffic, the greatest efficiencies are achieved for multicast traffic. However, the NGI must be able to scale as multicast traffic increases. Multicast routing must support both a very large number of small multicast groups, as we see currently with chat and whiteboard sessions, and a smaller number of large multicast groups, as would be required for collaborative modeling and simulations or cybercasting. Quality of Service (QoS) As ATM is deployed as a backbone technology, IP resource reservations, join/leave multicast groups requests and IP QoS requirements must map to ATM signaling for multicast SVC/PVCs and QoS parameters to carry traffic through the ATM cloud. Unfortunately, while the RSVP specification allows reservation to be changed during the multicast session, the latest ATM standard, UNI 3.1, does not support this capability. The pending UNI 4.0 ATM specification will support modification of reservations but only on point-to-point virtual circuits(VCs), not for the point-to-multipoint VCs required. Finally, ATM requires the receivers to conform to a homogenous flow specification while RSVP supports best effort and varied specifications for members of the same multicast group. On the ATM side, all point-to-multipoint VCs share identical QoS parameters. Smart Applications The majority of the Internet applications used today require standard IP unicast transmissions. In the future, multimedia applications will not only require resource reservation, but these applications must dynamically assess the quality of service provided by the network. Applications must be "smart" enough to modify reservation and QoS requirements as quality parameters such as latency, bandwidth or jitter are compromised due to aggregate internetwork traffic. Depending upon the activity, applications will statically remain a member of a multicast group until explicitly removed by the user (voice/ video conference) or dynamically join and leave multicast groups using "smart" API hooks into the underlying IP protocol stack. Applications must also be instrumented to provide management and control information so that SNMP based tools can determine the state of the applications and the degree to which they are "well-behaved" on the Internet. Summary The issues of scaleability, multicasting, resource reservation and IP-to- ATM QoS are being analyzed, tested, and prototyped extensively by the DD JPO to support the modeling and simulation community on the DSI and STOW programs. Just as the original Internet evolved from research and development programs sponsored by the DARPA, many of the technology issues associated with the Next-Generation Internet are also being funded and by DARPA and DISA. Many of these technologies may have commercial viability as the standards become better defined and advanced technologies are implemented in commercial vendor products. Research and development in these areas should once again prove viable to commercial businesses, industry and government as we migrate toward the Next- Generation Internet.