Of Earthquakes, Clouds and Auld Lang Syne
Posted on Mon, Nov 21, 2011 @ 02:14 PM
By Dave Connolly, Marketing Communications at Sonus Networks
On the twenty-third day of August, a 5.8-magnitude earthquake rattled the Eastern Coast of the United States. I’m not sure what the first thing is you’re supposed to do in an earthquake—probably hide under a desk or grab the hand of your Earthquake Buddy—but here on the East Coast the first thing people did was reach for their phones. The result was a huge spike in call activity. Just one of our customers, for example, registered 2,000 call attempts on their East Coast-based GSX9000 gateway in a single second and nearly 1,500 Call Attempts Per Second (CAPS) over a five-minute period right after the earthquake.
Of course, it’s no surprise that natural disasters put a lot of stress on communications systems. In fact, we built our VoIP architecture around the need for geo-redundant telephony switches and centralized routing. The benefits of this approach became increasingly apparent in the wake of Hurricane Katrina. What is surprising is that the spikes are no longer isolated anomalies—or connected only to natural disasters—but are becoming part of the regular fabric of communications. Take a look at the two charts below:
The chart on the left depicts the post-earthquake call spike. The chart on the right depicts the calling patterns in the hours leading up to New Year’s Eve. Of course, they’re very different events: one is highly predictable, the other highly unpredictable. But to a communications network, they’re both a cause for alarm.
When Nature Calls: Session Management Gone Wild
As much as the communications industry could do with a little less turbulence, the spikes are here to stay. And it’s not just earthquakes or hurricanes or champagne-fueled brotherhood that service providers need to watch out for. We live in a hyper-connected world where, instead of mass communication, we have communications en masse: everyone talking about the same thing at once. These spikes in traffic place a tremendous strain on network session management, which means that border elements like gateways and SBCs need to become spikeproof or risk getting spiked.
What do I mean by spikeproof? Well, I don’t mean planning high capacity for the spikes and running at 1% capacity the rest of the time. During a spike, sessions will be rejected . . . sometimes a lot of them. But the network needs to keep processing calls normally before, during and after the spike in order to mitigate the spike’s effects. That’s what I mean by spikeproof. You can also think of it as uber-overload control: How does a border element behave when it’s overloaded with requests?
One solution is to separate the session management (in this case, the overload controls) and call processing planes so that the border element keeps working at full capacity even when it’s being flooded by requests. That’s what we’ve done with the NBS5200 and NBS9000 session border controllers and they get consistently high marks in performance under stress, which is a compelling argument if you’re a service provider responsible for giving your customers a reliable communications network experience. For enterprises, however, the solution may be even simpler: get as much (or as little) network capacity as you need from the cloud on a “pay-per-byte” basis. Our service provider customers have already started to sell this kind of flexible cloud-based capacity to their enterprise customers.
The idea is that enterprises have their own unique spikes which, taken together, can be managed more easily by a cloud service provider. For example, an online retailer might have their biggest call spikes on Cyber Monday and December 23rd. For an online flower delivery service, those spikes may occur on Valentine’s Day and Mother’s Day. In these cases, the service provider can shift their network capacity to the enterprise customers who need it most at a particular point in time and “flatten” the spikes from their own internal network perspective. This is a very attractive cost model for enterprises, who avoid paying for capacity they don’t need, and an attractive revenue model for service providers, who move from being “invisible” bit-pipe providers to mission-critical cloud providers.
If you attended Connexions2011, you heard Hewlett Packard’s Joe Weinman and Korea Telecom’s Hong-jin Kim talk about the cloud-based future of communications. It may well be that these spikes are what drive the adoption of cloud communications, much as the potential flood of online visitors has driven the adoption of highly scalable, cloud-based, Web 2.0 app hosting platforms. So where do you stand? Are your eyes on the clouds, or do you think cloud-based communications are just so much smoke?