Section 2.2. Distributed Versus Mainframe

2.2. Distributed Versus Mainframe

In the world of traditional telephony, endpoints and PBXs interact in a manner similar to dumb terminals and mainframe computers. That is, the PBX (or mainframe) has all of the application functionality built in, and the user interface functions of the endpoints (or terminals) are dictated by the PBX.

With IP telephony, voice endpoints are far more programmable, lessening the requirement for centralization. VoIP endpoints don't always have their functions dictated by a particular VoIP server. In fact, VoIP endpoints may interact with many services on many different physical servers: DNS, LDAP, SIP, and RTP are all VoIP- related application protocols that may be facilitated by separate servers or by no servers at all (some operate between two endpoints and don't require a server in between). The IP-to-IP call placed in Project 2.2 is a good example of that.

Compared to a traditional telephone call, which must always be routed through a telephone switch such as a PBX, this is a significant difference. A traditional telephone call is set up, torn down, and accounted for using the same piece of hardwarethe PBX. Moreover, the sounds of the conversation are routed through the PBX, because the PBX is the circuit-switching mechanism that provides the voice loop between caller and receiver. This is illustrated in Figure 2-6.

But in a VoIP network, the call-management functions are separated much more from the voice transmission functions. This allows each function to be enabled through separate network resources, as shown in Figure 2-7. Call management could occur over a wide area link, while the voice transmission could occur directly between two endpoints on the same local area link, in order to preserve capacity on the wide area link. The net result is that a single, powerful call-management server could work on behalf of many remote sites, increasing the value of the WAN and possibly saving money that might ordinarily be spent on remote PBX systems to support each site.

Figure 2-6. With a traditional PBX, voice transmission and call management are dependent upon a route through the voice switch

Figure 2-7. In IP telephony, call management and signaling can be separated from voice transmission

The distributed nature of VoIP applications makes them preferable to traditional telephony on a wide area networkbut that's not the only advantage of VoIP on a WAN. The other great benefit of VoIPespecially in a bandwidth-conservative WANis compression.

2.2.1. The Core and the Edge

At the heart of a network resides the core , or network backbone. In modern IP networks, the core serves the purpose of transporting high levels of aggregate traffic between nodes that are probably not endpointsthat is, they aren't the hosts where the traffic originated or the hosts where it is headed, but rather hosts whose purpose is to forward that traffic along the core network until it needs an exit along the route to its destination.

The core is kind of like the 10-lane interstate highway : a lot of people drive on it, but nobody's driveway is an entrance ramp to it. So, while billions of hosts may send and receive data that crosses the Internet core (backbone), almost none of those hosts are directly connected to the core.

Instead, IP network endpoints connect to disparate network links that share high-capacity aggregate connections to the core. These links are collectively known as the edge . The edge is like the surface streets that surround the 10-lane interstate highway. Most traffic that ends up on the big highway originates from the surface streets .

A key difference between distributed and mainframe computing follows this analogy: in a mainframe environment, such as the PSTN, all the endpoints have a direct connection to a corethe central office switch. Likewise, in a PBX system, all the endpoints have a direct connection to a corethe PBX switch. So, all the driveways in a mainframe town are actually entrance ramps right onto the big highway.

VoIP facilitates the build-out of the networking smarts that normally exist at the traditional PSTN core, so that application functionality gets closer and closer to the edge of the network. This is similar to the way distributed PC applications have been displacing mainframe client/server applications over the past 20 years .

With VoIP, the core network is still there, and very much required, but it serves a different purpose than the core network of the PSTN. In a VoIP environment, the core is mainly used to move data back and forth, and the programmatic functionality of voice applications exists in a distributed model of peers: VoIP servers and endpoints. These peers can reside anywhere on the edge and offer new and changing features, without requiring changes at the core.

In traditional telephony, that isn't the case. The PSTN's core is itself responsible for all of the features available to you as a telephone company customer or enterprise PBX user, and offering new features can require the phone company or enterprise PBX administrator to alter the core network.

2.2.2. VoIP in Enterprise Networks

VoIP can be used to connect IP phones on an Ethernet segment to a VoIP server that is used for call management, and that VoIP server can be used to provide a connection for those phones to the PSTN, as in Figure 2-8.

Figure 2-8. A VoIP server can be a PSTN gateway for IP phones connected via Ethernet

A single VoIP server can act as a PSTN gateway for IP phones on Ethernet segments located at remote offices, as long as WAN connectivity exists between them. This way, the IP phones at all the sites can call one another, and the VoIP server routes calls between the offices and to the PSTN, as in Figure 2-9.

Figure 2-9. A VoIP server can be a PSTN gateway for many IP phones on a wide area network

If a large company uses a conventional PBX at every site around the country, all can be linked together using VoIP over a WAN. This way, each PBX can connect calls within its local network of traditional phones, as well as calls between them and the PSTN, but calls placed between phones on opposing PBXs can be routed over the WAN using VoIP, as in Figure 2-10.

Figure 2-10. VoIP servers can use a WAN to connect calls between PBXs in different offices (switch-to-switch trunking)

At a minimum, at least two VoIP devices (such as an IP telephone and a VoIP server or two VoIP servers) and at least one form of connectivity are required by all VoIP solutions.

Like the network, VoIP is a conversation-oriented technology. Its protocols are simply rules that devices and software must follow in order to carry on the conversations required to make VoIP applications workthat is, to carry each human speech conversation. Each VoIP protocol set (H.323 and SIP are the two big sets) has its own rules that enforce proper conversation, just as Congress has a parliamentary procedure that enforces its debates. The biggest rule is the definition of VoIP's minimum requirements: two or more TCP/IP hosts using one common protocol and connected data links.

2.2.3. Network Convergence

When you support only one transport (in VoIP's case, TCP/IP) for all networked applications, including telecommunications, you've achieved complete convergence. The more you leverage your TCP/IP network to support voice and multimedia telecom apps, the more you converge. Theory tells us that convergence increases administrator productivity, and experience tells us that support costs drop the more voice and data networks are converged .

Convergence isn't something that has to happen overnight. There may be plenty of reasons you don't want total convergence: capital that is tied up in perfectly good legacy hardware is one; network readiness is another. As with many past paradigm shifts in networking, a migration path exists to get you from partial to total convergence. One of these paths is the "hybrid" voice switch.

2.2.4. Pure IP or IP Enabled

Pure IP voice switches can't make direct use of traditional circuit-switched telephones and trunks. Vendors that refer to the VoIP solutions as pure IP mean that the phones and trunks connected to their switch are totally packet-based. Connections to outside non-IP systems, like the PSTN, are accomplished through outboard hardware that facilitates transmission of call signals to the call-switching server using IP. In this fashion, vendors whose switching servers support only IP endpoints and not traditional endpoints tend to use the pure IP moniker. Cisco's CallManager 4.0 is a good example of what pure IP meansit's a completely software-based switch that requires outboard hardware, called a media gateway, in order to support non-IP endpoints. As you can see in Figure 2-11, any devices that communicate with a pure IP PBX do so using the TCP/IP Protocol trunked over Ethernet.

IP-enabled voice switches, unlike pure IP systems, offer support for all kinds of voice endpoints and make no bones about connecting to analog phones and trunks like those from the PSTN. Analog, digital, and IP devices can all connect, as shown in

Figure 2-11. A pure IP switch has only IP-based trunks; all trunks that feed the same switch are carried by TCP/IP

Figure 2-12. The media interfacing required to use traditional telephony devices with an IP-enabled switch is all contained within the switch chassis, often using a single digital bus and microprocessor, much like a conventional PBX. Good examples of software-based IP-enabled switches are Avaya's Communication Manager 2.0 and Digium's Asterisk (an open source solution), both of which run on Linux. Sometimes VoIP implementers refer to IP-enabled switches as hybrid switches.