CUSTOMER STORY | UBER

Uber Solves IPv4 Exhaustion at Scale

Overview

Mitigate IPv4 exhaustion of private addresses

Simplified service creation and troubleshooting across DevOps teams

Reduced costs with flexible consumption-based licensing

COMPANY
Uber

INDUSTRY
Technology

SOLUTIONS
A10 Networks Thunder CGN
A10 Networks FlexPool

INTRODUCTION
Uber was founded in 2008 to provide a way to tap a button and hail a ride. Now, Uber is a household name and people took 14 million rides in 2019. Uber has 91 million customers and 3.9 million drivers. Bookings with Uber Eats, which addresses another massive market opportunity, grew 72 percent in 2019, and Uber continues to expand with on-demand freight and health transportation.

Challenges & Approach

The magic of satisfying a French fry craving or getting a ride to work in the rain is powered by Uber’s cloud. The company’s rapid growth and expansion means that Uber is consuming more of its IPv4 subnet for internal usage. In addition, acquisitions of other companies can result in overlapping IP addressing schemes. Making the most of finite resources to avoid IPv4 exhaustion is critical, and large-scale network address translation (LSN) is the way forward.

“We needed a better way to operate CGNAT for the long haul. I was looking for three things: zero-touch provisioning, a good CLI, and not having to manage licensing features” says Jason Black, Head of Global Network Infrastructure.

quote

IPv4 addresses are exhausted. The secondary market for IP addresses isn’t cheap, and we wouldn’t be able to get as much address space as we needed anyway.

Jason Black
Head of Global Network Infrastructure, Uber

Solutions

Thunder CGN deployed for largescale network address translation

Uber operates a hybrid multi cloud, using a mix of on-premises infrastructure, private cloud services, and public cloud providers and data centers. Uber deployed Thunder CGN for largescale network address translation in its Phoenix, AZ and Ashburn, VA data centers. There, Thunder CGN provides high-performance, always-on transparent CGNAT to extend finite IPv4 addressing resources to support Uber’s applications and services.

Uber Solutions
quote

A10 Thunder CGN delivers the most value for CGNAT products.

quote

Deployment of large-scale network address translation was methodically planned and executed. Thunder CGN fit neatly into Uber’s zero-touch provisioning mindset and automated tooling. A10 Networks’ open RESTful API and Thunder CGN’s powerful CLI provided full control and automation so Thunder CGN could be easily integrated into Uber’s Metal team processes. The team migrated the Phoenix data center to Thunder CGN first, moving traffic loads over several weeks. Once it knew the solution was successful, it did the same in Ashburn.

Key Results

Magenta Checkmark
Mitigate exhaustion of private IPv4 addresses
Magenta Checkmark
Simplified service creation and troubleshooting across DevOps teams
Magenta Checkmark
Centrally configured and manage policies for large-scale network address translation
Magenta Checkmark
Reduced costs with flexible consumption-based licensing
Magenta Checkmark
Simplified IP address capacity planning for innovation with self-driving cars

“With Thunder CGN, the infrastructure team can control our own destiny. We don’t have to scale across four different teams to manage the hosts.”

With Thunder CGN managing network address and protocol translation, the team can configure and manage network addressing policies across Uber’s on-premises and cloud environments.“With Thunder CGN, we have a standard monitoring pipeline,” says Black. Automated tooling manages and monitors network configurations and large-scale network address translation. That improves service reliability, operational efficiency, and simplifies future capacity planning.

A10 Networks FlexPool® licensing was another must-have. With FlexPool, Uber has consumption flexibility to distribute CGN services wherever and whenever they are needed across its cloud. CGN services are dynamically allocated to ensure application availability and over-provisioning is eliminated to reduce unused resources. “With A10, we can make sure we’re not over-licensed,” says Black