White Paper for Next Generation Internet Workshop ATM: The Design Tool for Networking and Distributed Systems Research Jonathan Turner Washington University in St. Louis Department of Computer Science Campus Box 1045 One Brookings Drive St. Louis, Missouri 63130-4899 email: jst@cs.wustl.edu Phone: 314-935-6132 Fax: 314-935-7302 Javad Boroumand University of Southern California Information Sciences Institute 4350 North Fairfax Drive Suite 620 Arlington, Virginia 22203 email: javad@isi.edu Phone: 703-812-3711 Fax: 703-812-3712 Allison Mankin University of Southern California Information Sciences Institute 4350 North Fairfax Drive Suite 620 Arlington, Virginia 22203 email: mankin@isi.edu Phone: 703-807-0132 Fax: 703-812-3712 ATM: The Design Tool for Networking and Distributed Systems Research The NGI initiative offers a rare opportunity to dramatically upgrade the national networking infrastructure in support of the total research and development enterprise in the US. By building upon recent advances in networking technology it can offer unprecedented levels of performance, a variety of new services and a more flexible infrastructure that can accommodate the incremental introduction of additional capabilities. In this white paper we argue that the NGI should offer both production networking services to the general academic research and educational enterprise and experimental facilities for use by networking and distributed systems researchers. We show how such facilities can be provided without putting production capabilities at risk, and how the facilities can offer great experimental flexibility without perturbing the management of the production infrastructure. The networking research that will be possible by use of the NGI and Internet-2 will directly benefit the enhancement of this infrastructure and its applications. Research areas may include new signaling and control paradigms, advanced congestion control and packet scheduling mechanisms, scaleable multicast routing and new control paradigms for distributed metacomputing. The research will produce new technologies and services in support of new NGI and Internet-2 applications with various QoS, scalability and reliability requirements. ATM VIRTUAL NETWORKS While there is considerable debate as to which end-to-end network services are most important, the merits of ATM as a flexible infrastructure for data and multimedia communication are widely recognized and the technology is growing rapidly in both campus and wide-area network settings. One key capability that ATM offers is the ability to construct virtual networks on a shared infrastructure. This is an extremely powerful tool that enables the kind of flexible allocation of resources that is needed to create an NGI that can evolve and grow in response to changing needs and new technological developments. The principal tool used to create virtual networks is the Virtual Path (VP). VPs offer a second level of multiplexing that supplements the more familiar Virtual Circuit (VC) multiplexing. Using VPs, collections of VCs can be routed through a series of intermediate switches in a transparent fashion. Signaling and other control messages pass directly between the switches at the endpoints of the VP and are not seen by intermediate switches. This makes it possible to have multiple virtual networks operating over a common ATM infrastructure. At each campus a Gateway Switch provides the interface to the carrier network. Gateway switches on different campuses can be linked via Permanent Virtual Paths (PVPs) into a virtual network. This virtual network can be used to carry both Permanent Virtual Circuits (PVCs) and Switched Virtual Circuits (SVC), supporting either IP-based or native ATM applications. All that is required for this is for the gateway switches to have signaling enabled on the internal campus links and on the PVPs linking them to other campuses. In a similar way, the campus gateway router can be connected to gateway routers on other campuses via its own set of virtual links. In this case, PVCs are sufficient. The gateway switch passes another set of PVPs to an experimental network that may include multiple switches, routers and other experimental devices. Within the experimental network, the PVPs may be routed and terminated as needed to support various networking and distributed systems research activities. (While ATM is used as an underlying mechanism for configuring the facilities, the experimental possibilities are by no means limited to ATM; indeed this infrastructure is eminently suitable for research on IP and on higher-level distributed systems; CAIRN and vBNS are existing proofs.) While experimenters on different campuses will have to coordinate their activities, there is no need for coordination with either the carrier providing the backbone network service or the campus organization providing production network services to others within the campus. This ability to decouple the experimental activities from the production services while sharing the same core infrastructure is key to enabling ongoing innovation while protecting users requiring a stable networking platform. To isolate experimental and production traffic from one another, each PVP should have a bandwidth allocation that is reserved on an end-to-end basis. The appropriate service is a variable bit rate service with equal values for the peak cell rate and sustainable cell rate parameters. This allows the virtual networks using the VPs to treat them as physical links. Traffic shapers at the boundaries between the carrier network and the campus, and between the experimental and production networks on campus, can monitor traffic usage at the boundary and buffer cells temporarily (on a VP basis) to limit the traffic flow to the configured rate. The traffic shaping function can be implemented within a switch or in a separate device. INTERCARRIER ATM INTERCONNECTIONS Although ATM has been widely deployed by the carriers within their core networks, there is still no intercarrier ATM interconnections. This a key issue in building the NGI infrastructure and its virtual networks. Currently intercarrier PVPs can be established only via a customer site that has multiple carrier ATM service (a couple of CAIRN sites are passing PVPs between MCI and Sprint ATM networks). Such interconnections should be done within the carriers facilities with the cost distributed among all the interconnect users. Scalability of PVP management across these interconnects can be an issue and therefore new ATM management tools are required. GIGAPOPS The virtual network concept can be readily extended to the gigapop setting. A switch at the gigapop would simply propagate the PVPs between the campuses and the carrier network. Alternatively, the gigapop switch may terminate some or all of the production PVPs and provide SVC service among all the PVPs it terminates. This enables more flexible and efficient sharing of the carrier facilities, but does require that the gigapop switch support the SVC service. Note that this approach is not applicable to the experimental PVPs, since this would restrict the ability of experimenters to configure their virtual networks appropriately. Gigapops could also be the interim locations for intercarrier ATM interconnections.