The Case for IVR Hosting

We're going to do some arithmetic here to demonstrate how an infrastructure can be built robust enough to guarantee 99.999% uptime. Before we proceed, we'll have to make several assumptions which are as follows:

• 10% of your components will fail any given year. This includes IVRs, servers, and network connections.
• 10% of your telecom circuits will also fail any given year.
• If a backup exists for a component or circuit, the failure rate drops to 0.008%.
• Most carriers will experience some sort of major service disruption once every ten years or so.

Low volume IVR applications rarely require more than a T1 worth of telecom capacity. This would indicate a need for only a single IVR server to which a single T1 is connected. This IVR, in turn, must communicate with an application server that then communicates with a database server. The illustration below shows all of the parts of this non-redundant system. When a customer buys an onsite server from Plum, this is the typically the type of infrastructure into which the IVR is integrated.

Any of the components in this diagram, from the T1 to the database, has approximately a 1 in 10 chance of failing for the year. Because the failure of any of these components will result in the failure of the entire system (until the component is fixed or replaced), the odds of a total system failure occurring in a given year is 52%. On average, a little less than once every two years, there will be an outage.

Let's try a simple tweak to improve our odds. Let's add a second T1 from the carrier. Each T1 individually has a 90% chance of staying up all year long. The odds of both of them failing over the course of a single year drops to 0.008%.

Now there's a 46% chance that some piece of the infrastructure will fail this year. The T1s are no longer a concern, but all of the servers are still fault-susceptible single-points-of-failure. On average, a little more than once every two years, there will still be a system outage.

One of the great things about VoiceXML system architectures is the fact that IVR capacity can be scaled linearly. If you need more IVR capacity, buy another IVR server and provision more T1s. By merely adding a redundant IVR server, we not only increase capacity, we also dramatically improve fault tolerance of the system as a whole.

The odds of infrastructure failure drop to 34% for any given year. You can thus expect a failure once every three years. The infrastructure still has exposed single-points-of-failure across the application and database servers. And it should be noted that if you build out such an infrastructure from scratch, you'll be adding a significant amount of idle capacity which your company may or may not grow into. Such is the cost of building a robust, dedicated infrastructure. By choosing to host with Plum Voice, you'll be able to take advantage of a robust shared infrastructure where the infrastructure and operational investments are borne by Plum so you focus on your business rather than worrying about operations, infrastructure, and capital expenditures.

Back office web infrastructure is a little trickier to beef up. Load balancers are required to properly distribute traffic across multiple application servers. Carefully planned-out replication strategies are required to build fault-tolerant database systems.

However, if you are willing to invest in that level of IT infrastructure as we have here at Plum Voice, you will find the odds of system failure for a given year drop to 10%. A complete system failure will occur only once every 10 years.

Finally, in order to remove the final site-wide single-point-of-failure, you will need more than one carrier from which to receive call traffic. Here at Plum Voice we receive traffic from four different carriers. However, by adding even a second carrier (much less a third or fourth,) you will have eliminated the final single-point-of-failure from your infrastructure:

A failure of such an infrastructure would occur once every 1,700 years.