Uber backs hybrid cloud as route to business and geographical expansion

Uber’s head of compute Dean Nelson has revealed how the firm is pursuing a hybrid cloud strategy to ensure its infrastructure has capacity to cope with its expanding business interests.

In the eight or so years Uber has been going, its app-based taxi-hailing service has rolled out to 600 cities worldwide, with its platform helping more than three million drivers around the world find fares, and 15 million customers a day get to where they want to go.

“We ignite opportunity by setting the world in motion. Everything is about moving something from A to B: it doesn’t matter if it is a package, a person or food. Anytime we need to move something, we want to be in the middle of that transaction,” said Nelson, during a keynote at the Datacenter Dynamics DCD>London conference in Old Billingsgate.

The company has also expanded into food delivery with Uber Eats, and is plotting a move into freight, drone-mediated food deliveries, healthcare services and aviation, with the first trials of its Uber Air taxi service set to begin in Dallas, Los Angeles and Paris in 2021.   

“The key [to remember] is we’re just getting started. It just blows my mind every day when I go to work, and I keep hearing about the new features we’re adding onto the platform, the new businesses we’re going into, but globally this is changing the way things work,” said Nelson.

As the range of services Uber has to offer continues to rise, so too does the amount of load its IT infrastructure has to handle. “They [the services] all require a transaction on our platform,” said Nelson.

To achieve this scale, Uber relies on a hybrid IT infrastructure setup, he said, which he refers to as its Tripod Strategy that combines the use of public cloud services with standardised on-premise server racks in its colocation facilities, all separately handling compute, storage, database and GPU workloads.

“The reason for that is, when you get to scale, you have volumes that you can get price/performance that rivals cloud, but it is extremely difficult to [get the] ability to move like a cloud by yourself,” he said.

“You need the ability to turn up for GDPR [General Data Protection Regulation] and data sovereignty in Europe tomorrow, so we have multiple cloud providers, and we also have on-premise to help us keep balance.”

The on-premise portion of the infrastructure is a reference architecture Nelson terms Uber Metal, which provides the “building blocks” the company needs to achieve the scale to not only expand the range of industries Uber crosses into, but also its geographical spread.

“What is Uber Metal? It is the infrastructure that runs our on-premise metal as a service,” he said, which is designed to run out of its colocation facilities.  

The aforementioned server racks are all organised into identical, standardised pods, containing 16 racks consisting of decreasing quantities of storage, database capacity, compute power and GPU processing capabilities, that are all hooked up to a 600G uplink.

“We have 16 racks that make up a pod. Then we take that one step further [to create a zone containing] 30 pods with 16 racks, which gives you reference architecture of 480 cabinets of usable IT capacity,” he said.

There are also some additional racks added to each one, bringing the total up to 576, to effectively act as a capacity buffer.

The idea being these datacentre zones will be dotted around the globe, for failover and latency reasons, while ensuring Uber works to ensure it has the IT capacity it needed to run its services all around the world.

The reason the firm is being so open about its Uber Metal vision, added Nelson, is because it is still under development and wants the industry’s help in refining it.

“I’m walking you through our reference architecture and the reason for that is I really want you guys to understand what it is we’re trying to build, because it takes all of us to get there and we need your help,” he added.