Google infrastructure czar Urs Hölzle is focused on a cloud future where customers don’t think about the infrastructure underlying all of the workloads they’re running.
In his view, one of the key advantages of the cloud is that customers can get the benefits of new hardware without having to completely rework their software.
“So that means you can have a million customers who move to that new hardware platform, not knowing they did,” he said Tuesday at the Structure Conference in San Francisco. “Which means that you can really insert this new technology in a much faster cycle than you could if you did the same thing on-premises.”
That means companies can get quick, seamless improvements to performance, as opposed to an on-premises deployment. When operating their own data centers, companies must take the time to evaluate new hardware, and take the time to roll it out.
Use of the cloud also means providers like Google can experiment with the underlying hardware to improve performance, and then decide to go in a different direction, without requiring customers to rework what they need to do.
While Hölzle’s vision holds true to a point, no abstraction layer will be completely bulletproof in the future, IDC Program Director Al Hilwa said by email.
“Over time, architectures and methodologies change, and in software, that is happening faster than in many spaces,” he said. “Still, a good abstraction provides protection and better longevity.”
Furthermore, Hölzle sees a future where the overwhelming majority of customers won’t even worry about what type of compute instance their workloads are running on. Instead, platforms will be responsible for intelligently suggesting what sort of compute resources customers should use and keep pace with what they need.
“So I hope, five years from now, 1 percent of cloud customers know the word ‘machine type,’ and 99 percent say ‘I’ve never thought about it,”Hölzle said. “‘My stuff runs at great performance, low cost, I guess someone must pick the right machine type for you.'”