Protocol Interoperability and VoIP Performance
Like this article? We recommend
Like this article? We recommend
Like this article? We recommend
In my recently released book Putting VoIP to Work: Softswitch Network Design and Testing (Prentice Hall PTR, 2001), I discuss detailed interoperability call flows with the current attributes and properties of the VoIP signaling protocols. However, it is a major task for a softswitch platform to define the types of calls that will be used in setting the performance expectation of the platform in calls per second. Several factors come into play when undertaking such an effort:
There are basic calls and feature calls in telephony. Basic calls are nothing more than a station-to-station call without operator assistance. You dial the number on your phone, the far ends rings, someone answers, you stay on the phone for some time, and then you terminate the call. The performance parameters in this exchange are simply the post-dial delay, the call capacity of the switch (which might be affected by the call hold time), and the number of calls that the switch can handle in a time interval (per second) of this call type.
There is another type of voiceband call, the often-misunderstood fax or modem call, which might (or might not) have been implemented as a straight voice call between machines, as is the case with the TDM voice network (as in a G711 call). For example, if a protocol such as T.38 has been used to transport fax signaling and data, then this is no longer a basic voice call; its performance impact on the softswitch platform must be assessed. In other words, the fraction of fax calls as a percentage of the total calls through the system will move the calls-per-second operating range of the switch in one way,
Firewalls tend to put a damper on things, depending on the sophistication of the platform. For example, a platform that does 100 calls per second and that uses a firewall capable of only 30 calls per second in the signaling path will be dragged down to the lowest parameter—the firewall. It doesn't matter how sophisticated the switch itself is. This part of the design requires special attention.
QoS considerations during the setup of the call might affect performance in the time that it takes to complete a call. Scalability could be affected as well if resources are tied up to service the call for its duration.
Finally, the types of signaling protocols that need to be invoked in the end-to-end path of a call will have a profound impact on softswitch platform performance.
The last issue is the focal point of this article for further discussion.
Engineers have often asked, "Do we really need all these VoIP signaling protocols to replace the loop-start/ground-start signaling mechanism of the PSTN?" The major ones that have survived and that are finding acceptance in implementations are MGCP, SIP, H.323, and Megaco/H.245. Another question that is often asked is, "How can we pin down the performance of a platform in the design phase if we don't know the call mix and the protocols that will be involved in the signaling process?"
First, let's address the question about the plethora of VoIP protocols. Each protocol offers the promise of integrated voice, video, and data (such as 24 x 7 Internet access) over the same set of wires, and each one represents a school of thought in signaling architecture. SIP, H.323, MGCP, and Megaco all address these requirements from a different perspective. SIP and the streamlined H.323v4 are intelligent but have had to be modified since their earlier incarnations to accommodate the stringent requirements of replacing the basic telephony service and feature support from local service providers. The term basic telephone service is key because it implies conformance with the federal and local regulations that govern the behavior of the service that we receive from the PSTN LECs. MGCP and Megaco address the same requirements, but at a much lower level; they utilize a "decomposed" gateway model, whereby the endpoints (telephones) are pretty dumb and all the intelligence is located in the media gateway controllers in the service provider's domain. As such, the number of signaling exchanges to complete a basic call with MGCP and H.248 is greater than the number in SIP and the streamlined H.323.