The best of both worlds

Defining the role of TCP/IP in network architectures has been a challenge for many vendors and carriers during the last three years. But whether the debate is connection vs. connectionless, TCP/IP vs. asynchronous transfer mode or packet vs. circuit, a successful competitive local exchange carrier will blend the benefits of all these concepts by successfully integrating the open systems interconnection and TCP/IP models. The fundamental question is, can there be a seamless integration of OSI and TCP/IP that is economically justifiable and technically capable of providing future-proof communications services?

Figure 1 shows how the OSI and TCP/IP models have been used over the years. Traditional cable networks have had a huge advantage because they generally represent one or more LANs that cover whole cities. Within this network, hybrid fiber/coax (HFC) plant can easily transmit data as RF-modulated TCP/IP in a broadband LAN. Because the RF modem generally uses an Ethernet protocol, common interface equipment is economical and widely available. For this reason, cable operators can provide the template for the future - they only have to consider interfacing with the OSI model when accessing the public network, usually no more than once per cable franchise.

Data-centric CLECs, on the other hand, do not have to interface with the OSI public network model. Instead, they can use private fiber trunking to create a TCP/IP network, eliminating the need to convert to OSI. This inherently creates a "connectionless" network. Further strengthening the rationale for this model is the development of Internet protocol (IP) Version 6 - with a 128-header addressing scheme and 15 priority levels - which significantly improves network reliability. Some would argue that this connectionless network can perform as reliably as a connection-based network. The problem with this model is that it assumes that the physical network will remain reliable, which makes it vulnerable to physical-layer problems. On the other hand, the advantage to TCP/IP is simplicity and cost savings.

Traditional telephony had two agendas that were both embraced by the OSI model: the absolute reliability found in a connection-based network (emphasizing physical layer protocols and digital transmission integrity) and the existence of billions of dollars in legacy equipment. Yet the idea that OSI and TCP/IP can be interconnected has recently emerged, bringing together the best of both worlds. Many technologists and vendors on both philosophical sides are building "OSI gateways" to bridge these dissimilar models.

This concept of OSI and TCP/IP integration is the basis of planning at most communications companies. This model provides the best of both worlds: using ATM to seamlessly mesh together synchronous voice, OSI protocols and TCP/IP protocols. This becomes essential in the development of network reliability and network features. Because today's circuit-switched network - either cable or local telephone infrastructure - passes 100 million households, the ability to build upon this strength rather than abandon it is perhaps the most fundamental strategy.

The beginnings of OSI

While X.25 was gaining widespread acceptance in the late 1970s, the International Standards Organization crafted a seven-layer framework of protocols for data communications. This was called the OSI network model, and it defined the transition from circuit-based to packet-based communications in a public network environment.

The OSI network has long been the public network standard. As the label indicates, it represents "open" systems that allow equipment manufacturers and communications providers to develop applications that all can be organized in a common network. In the seven-layer OSI concept, each equal layer has a defined set of protocols to interface with the adjacent layer. From the bottom up, they include: - The application layer, which supports applications developed for open system communication and their semantic exchanges with lower-level layers in the OSI network.

- The presentation layer, which supports processing of the data sent and received from the applications layer, including data stream support, data conversions and code translations.

- The session layer, which supports the organization and processing of interactions between applications and devices.

- The transport layer, which supports the transfer of data, including management of data sequencing, control of data flow, error detection and application flow processing.

- The network layer, which supports the connection between networks. This layer includes all operational control procedures for internetwork communications, as well as routing information through multiple networks.

- The data link layer, which supports the connection between all network resources. It provides the functionality of the overall network and defines the protocols to transfer data. It also detects errors that may occur between network components.

- The physical layer, which defines all mechanical, electrical, optical, functional and procedural standards associated with the public network and physical transmission of data over the communications medium.

This organization in the OSI model, although complex, offers the ability to create different communication implementations by adhering to the standard interfaces between the layers. The International Standards Organization has long grappled with defining the OSI and the precise specifications that craft this and much of the public network model.

The transition from OSI to TCP/IP

The TCP/IP model was developed by a group of researchers at the U.S. Department of Defense. Their goal was to build a network model that supported a specific network designed for data communications. This was derived from the need to link virtual interconnection addresses to physical hardware link addresses.

Although Figure 2 depicts the commonality of the OSI and TCP/IP models, they are vastly different in origin and functionality. This is fundamental to the dilemma associated with carrying data, video and voice over a connectionless (TCP/IP) or connection-based (OSI) network. How can a TCP/IP network without reference to the physical or data link network layers be considered part of the public network? How can the OSI network integrate TCP/IP without embedding the cost of OSI's extensive Layer 1 and 2 protocols?

In the TCP/IP model, there is no protocol organization or consideration for the physical layer. Historically, the network was not public, and therefore the need to define it never materialized. As it can be critically viewed today, there are at least three different views of how this affects public network policy:

- Cable operators have the greatest advantage because the HFC network, with an RF-based information stream, is not a public network but similar to a LAN/WAN network. The TCP/IP model can be implemented without much concern, as it favors this type of network.

- Data-centric CLECs use the public network for information flow. To base communications on this network without seeing the physical network layer is a significant risk. For voice over IP and similar real-time applications, carriers cannot rely solely on TCP/IP networking in a connectionless network.

- Contemporary CLECs, which deploy OSI network organization, have the ability to bridge the TCP/IP traffic onto the OSI platform. These opportunities represent the best of both worlds because the CLEC experiences the reliability of the OSI network model and the flexibility of the TCP/IP model.

CLECs and telephone companies can infinitely improve the reliability of TCP/IP communications by using the OSI model on the edges of each network connection. This allows for the successful integration of IP and the establishment of virtual circuits through the TCP layer for dedicated two-way data transmission, voice or multimedia.

The disadvantage of ATM management of TCP/IP is cost. The network planner must accommodate Layer 1 and Layer 2 costs in the network model. The solution is to place ATM on the edge of the public cloud, where the TCP/IP network interface or OSI gateway exists. This ensures public network reliability but does not enter the ATM cost into the TCP/IP local network. There are far too many technologists that are ignoring Layer 1 and Layer 2 aspects of the OSI architectural model. There is an overall assumption that TCP/IP can be directly ported to the public network and that the other guy can deal with the reliability, costs and physical layer of the network. This is a dangerous assumption.

The advantage of ATM management of the TCP/IP network is reliability. ATM has its own quality-of-service (QOS) standards. ATM's adaptation layer scheme segregates types of information for more efficient handling, and it has superior hierarchical routing and addressing schematics that allow for complete network reliability. Today's communications users demand these advantages.

Without ATM, TCP/IP is a routing protocol that cannot access the public network with the predictable reliability needed in the public network. Carriers have created specialized TCP/IP networks that provide users with OSI-level reliability by staying off the public network platform. But any national network provider must access the public network at some point.

Thus, CLECs should consider the following in handling TCP/IP network design:

- TCP/IP can and must be planned and engineered with the OSI network gateways at all points entering the public network.

- The most important interface to provide universal QOS is an ATM interface and the ATM "classic" IP-over-ATM protocol defined in ATM AAL5. - Without integration into the OSI model, the TCP/IP network must stay on dedicated TCP/IP trunks to yield the same results.

So where does that leave us?

As with most concepts, there is no one right answer. However, there are two predominant models that reflect the integration of OSI and TCP/IP networks. Figure 3 represents the most efficient way. This is particularly justified if QOS is essential or local telephone service is being offered over a facilities-based solution or circuit-based switch.

In this scenario, the TCP/IP network directs local traffic into an ATM or packet-based switching system. This TCP/IP network can bridge the circuit- to packet-switch world to allow IP originated voice calls to terminate on a circuit switch. It can also bridge point-topoint protocol traffic to the circuit-based world. The TCP/IP network can terminate all sorts of router originated broadband traffic. This is all accomplished with OSI gateway equipment.

Figure 4 represents an alternative methodology that has been accepted by some carriers and data-centric CLECs. Although this does not fully embrace the OSI model, it does provide equivalent functionality of the OSI model and eliminates the need for a circuit-based switch for voice over IP. In this scenario, the TCP/IP carrier provides a redundant (not always self-healing) IP backbone. The local CLEC network can be pointed directly at the backbone for outbound traffic. The packet switch has an OSI local loop interface to work with the public network. However, the circuit switch and much of the OSI provisioning is eliminated. All traffic, whether in the LAN or hosted over a leased public local loop, is assigned IP addressing to handle all information flow.

Fundamentally, however, all information streams are virtual circuits routed over a "best effort" networking protocol. The risk is that this network continues to send traffic without prior verification that the network is intact. Assuming thousands of simultaneous TCP sockets are in use, this represents a potentially large-scale data loss.

Regardless, both network architectures have their particular strengths. The consideration of OSI network architecture must be incorporated into the TCP/IP model if carriers want to make a conscious effort at meeting QOS standards.