On-Demand Outages the 'New Normal'?
Earlier this week, Salesforce.com migrated a subset of its overall user base to its Spring '08 release. In the process, Salesforce.com experienced an outage to its NA5 server Feb. 11 that lasted, intermittently, for about 7 hours and affected a small set of customers.
On the same day, BlackBerry users found only a black hole lasting for several hours when they tried to reach e-mail and calendar services. While it's not clear exactly what caused either outage, the bottom line is that on-demand services, like the phone company, are subject to some unexpected downtime.
Are intermittent system outages the end of the world? No, for most users (though it may seem like it at the time). But small outages may well be the new norm for users of on-demand services, at least for the time being.
Charlie Wood, developer of Spanning Sync, a service that synchronizes Google Calendar and Apple iCal, sends about 10 million calls a day to the Google Calendar Data API.
"Most people would be surprised how often Google Apps are not available," Wood said. "It's not these major outages like Salesforce.com or BlackBerry, it's microusers-a small subset of users that would be affected for 15 minutes. It's very distributed. We see it all day long. It's constant that they have these small, typically short-lived outages that affect small groups of users. We're syncing our users on the back end and if one fails, 10 minutes later another one will succeed."
Wood said Google is aware of the issue and working "really, really hard" to make Google applications more available. (A Google Gears developer advocate did not return calls from eWEEK.)
In fact, most SAAS (software-as-a-service) providers are acutely aware of any downtime issues and spend a good deal of capital ensuring what's typically referred to as the "five nines" or 99.999 percent uptime for customers.
In 2005, Salesforce.com spent $50 million to build additional data centers and get as close to 100 percent uptime as possible by replicating customer information in real time. But even that effort was not without issues. Salesforce.com customers experienced a string of hours-long outages in 2006 that CEO Marc Benioff said were a result of the data center upgrades.
The bottom line for Wood, who formerly had a similar syncing program for Salesforce.com, is not that the APIs don't work sometimes, but that the outages are small and, in the long run, workable.
"It's the new normal," Wood said. "I think what people are going to have to realize is it's the nature of the Web, the nature of the Internet-delivery of information is not guaranteed. [Data] is very likely to get there, but it's not guaranteed. If you go to a Web page [and it's not available], you hit refresh and it reloads. People will have to get used to that with online applications. There [have] to be some shifts in expectations."
So what should customers of on-demand applications do? Wood said users need to have a backup plan when their data is sitting in the cloud. "Using traditional on-premises applications is like owning a car, whereas on-demand is like taking a taxi," he said. "With a taxi you pay for the ride, but you don't pay for things like insurance and maintenance. But you can't always be sure you're going to find a ride. But if you have a phone you can make other arrangements. I would say the same for apps."
When asked what a reasonable uptime expectation should be for users of on-demand services, Beagle Research analyst Dennis Pombriant said "expectation" is the key word. "The expectation that everybody has going forward is that the on-demand technology world will be as reliable as the phone system," Pombriant said. "The phone system overall doesn't usually go down. It may have an outage in a small area-a line that's down somewhere-but the phone system doesn't go down."
But in contrast to the 125-year old phone system, the SAAS industry is still in its infancy, and should continue to improve, Pombriant said.
"You should expect uptime to get better over time, but also expect that there are going to be continuing minor problems simply because we're dealing with a moving target," he said. "It's not just SAAS, it's SAAS at a multilevel with additional partners building applications and hosting applications. It's a global business. There are a lot of parameters ... the best expectation is no outages. The next best is small outages."