Putting data in the cloud may be an all-or-nothing proposition

Teradata CTO Stephen Brobst sees a hybrid future for most parts of information technology, but not for data stores.

There's no shortage of technology and applications now moving into hybrid cloud situations, but enterprises need to think twice before splitting their data stores between clouds or on-premises environments. That's because network latency can still slow things down.

data-big-data-2-photo-by-joe-mckendrick.jpg
Photo: Joe McKendrick

That's the word from Stephen Brobst, chief technology officer with Teradata. I had the opportunity to sit down with Brobst at the latest Teradata Partners confab, where he laid out the state of cloud adoption.

Some companies are lifting and shifting everything and anything right to the cloud, he relates. For others, the move is more gradual, usually starting off with functions such as dev/test and disaster recovery. Plus, cloud resources come in handy for bursting situations, such as seasonal surges in user demand.

But there's one area that needs to be all or nothing when it comes to cloud -- that's data stores, Brobst says. In particular large enterprise stores -- such as data warehouses -- that are employed for right-time analytics -- need to be cohesive. "If you separate data between the cloud and on-premise, the cost of integration is quite high," he says. "You're joining data across a wide area network, which is not really a great idea."

Managing data stores in a separate location, such as cloud, works if it's for a function "isolated" from the mainstream enterprise, he continues. For example, one of his customers, an auto manufacturer, launched a data repository in the cloud for an experimental vehicle loaded with sensors. "It's a lot of data, and they really didn't know what they were going to do with it," he relates. Ultimately, they replicated portions of the cloud data set and combined the replicated data with their on-premises data. "Once the experiment was successful, then they had motivation to consolidate the data," he says.

Ultimately, cloud may be the best hedge against technology obsolescence. Brobst believes things are moving too fast to settle on a particular set of solutions, or even a particular architectures. "In this space, anything that has been commercially packaged is already obsolete," he says, adding the "the open source community is setting the pace."

Even the new technologies flooding into the corporate space -- algorithms, artificial intelligence, and deep learning -- may be obsolete before they can be of value to enterprises. Brobst cautions. "The whole infrastructure around deploying deep learning is very top of mind for customers -- there's so much hype right now around deep learning. The reality is it's not always the answer." For instance, he says,while deep learning works well for "very high dimensionality, dirty and incomplete data and so on," in many cases, simple and cheaper linear approaches will do the job.

"It's a bad idea to get locked into an particular algorithm or software packaging of technology.. that does not allow you to move fast, because this space is moving very fast," he says."Whatever you put in today, two years from now there's going to be a much better answer -- and a lot of technical debt you can't get rid of."

(Disclosure: Teradata assisted with my travel expenses to the Partners event.)

Newsletters

You have been successfully signed up. To sign up for more newsletters or to manage your account, visit the Newsletter Subscription Center.
See All
See All