High Availability — In the Cloud

“Highly Availability” has been a hallmark of the telecom industry for more than a century but the design parameters have changed and not everyone has noticed. With the advent of computers in telecom more than 40 years ago, software issues became part of the puzzle. Now with the advent of commodity computers and cloud computing services, economic considerations tip the balance almost entirely to software design and yet I still encounter system architects who remain focused on hardware.

If you’re designing a PSTN gateway, it’s true you have to terminate legacy trunks, e.g. N+1 redundant T1/E1 lines or 1+1 redundant optical links. And you’d like to minimize costs by favoring N+1 redundancy wherever possible — the larger one can make N, the smaller the relative cost of the 1 redundant (and thus normally idle) component. But the difficult issues in such a design are software ones. How does call control distribute traffic? How does the software continue to function is the presence of processor failures and software bugs? And how do you guarantee data integrity so call processing resumes correctly after failures?

But once you’re designing services on IP networks, everything changes. You can assume a reliable Internet backbone, accessible by connecting to two or more backbone providers at different locations. Multiple PSTN wholesalers, each with multiple PSTN gateways, make reliable PSTN connectivity a commodity. And now there are multiple sources of commodity computing available for rent by the hour, e.g. Amazon’s EC2 service, Rackspace’s Mosso and many others. This means there is no hardware design in the traditional sense — all design issues are software issues. More importantly, it changes the economics and thus the design criteria.

With commodity cloud computing, your costs depend on storage used and on hourly traffic levels (as bandwidth is billed by the gigabyte and compute capacity is billed by the CPU-hour). For reliability, identical functions have to be distributed across at least two different cloud vendors and two different access vendors. For management purposes and reliability, the best architecture is one that uses identical execution images distributed across multiple CPUs. This allows the number of CPUs to be dynamically adjusted to match current traffic plus a safety margin.

The real design issue is how to do load balancing, but now that we’re 100 percent software, there are a number of solutions already in existence and directly applicable, including open source solutions like Linux Virtual Server, Red Hat (News - Alert) Cluster Suite and Ultra Monkey. What’s more, they run on the same cloud infrastructure.

Finally, there’s no up front investment in equipment. Today’s highly available service can be deployed on a pay-as-you-go basis. This is an exciting new world. IT

Brough Turner is Chief Strategy Officer of Dialogic (News - Alert) (www.dialogic.com).

» Internet Telephony Magazine Table of Contents