The rapid growth of Internet ( 130% per year ) has caused a logistical and operational nightmare at the hub sites where scores of point to point links terminate, in many cases from the same phone company central office using replicated physical facilities and equipment. Because of the bursty nature of traffic on the Internet, much of this dedicated bandwidth is not in simultaneous use and lies idle. No network management information is available from the telephone company about these links. An upgrade from one speed to another necessitates the provisioning of a brand new link. This architecture does not scale well with continued growth and does not meet the requirements of Internet Service Providers.
The requirements seem to have found a good solution in series of offerings from the telephone companies dubbed Fast Packet Services -- SMDS, Frame Relay and ATM. Fast packet services shift some of the burden of switching packets from the Internet hub sites to switches provided by the telephone company. The actual technology used by these service is not the focus here, we concentrate how the the use of these services is operationally expedient.
Fast Packet Services permit the use of a single link of varying bandwidth to the Internet hub site, on which packets sourced from the remote LANs have been multiplexed by the switch. Remote LANs connect to the switch using links of lower bandwidth. Reducing the number of links addresses some of logistical and physical problem not only on the hub site as well as the telephone company central office as the overall number of loops decreases from 2N to N+1 for N LAN connections to the hub site through the central office. The bandwidth of this link can be chosen to provide the desired quality of service in terms of latency and throughput. This permits the efficient use of bandwidth. The shared switching fabric permits backup routing from the remote LAN to an alternate backbone hub almost seamless. The use of an intelligent switch in the phone company central office ensures that network management information can be collected and queried. These services blur the line between MANs and WANs as the same technology can be used to deploy both kinds of networks.
CERFnet has gained significant experience in the past 3 years in deploying all three of these services in a production Internet environment. CERFnet pioneered the large scale use of SMDS in Internet access applications. These services have helped improved the operational efficiency of Internet services and we estimate that we have reduced the cost of service provision and incidence of total service failure by 25% each. These services differ in deployment requirements, range of available bandwidth, and failure modes. We found the the connectionless nature of SMDS more suited to traditional operation of the TCP/IP protocol, while the Committed Information Rate of Frame Relay and ATM allowed for finer control on bandwidth allocations at the expense of increased configuration and tracking information required to provision permanent virtual circuits. This virtual circuit concept also made detection of failures more evident, while failure modes in SMDS resembled those of a connectionless LAN like Ethernet. Frame Relay was only available at speeds of less than 1.544 Mbps, ATM only higher than 10 Mbps, and SMDS from all speeds from 56 Kbps to 34 Mbps. Only SMDS offered a Multicast capability that benefited certain applications and routing protocol performance. The differences in the three services will be explored in detail in the conference.