T-Mobile.com Plays the Super Blitz to Perfection

Is it really possible to play offense and defense simultaneously in the biggest game of the year?

T-Mobile did.

Super Bowl Sunday, February 7, 2016:

Digital technology teams from The Garrigan Lyman Group (GLG) assembled at points around the country and stood ready as our client T-Mobile aired TV commercials beautifully created by brand agency Publicis, featuring rapper Drake and TV host Steve Harvey during the largest television event of the year. When the first ad hit in front of some 100 million US viewers, our defense was set to take on an almost immediate spike in user traffic on T-Mobile.com. But our offense was ready, too, with a simultaneous launch of a corresponding digital experience on the Un-carrier’s site.

While monitoring site activity non-stop through the broadcast, we compared baseline resource consumption with the super surge in traffic, and the user experience was flawless throughout. But how did we deliver T-Mobile’s huge digital Super Bowl experience without compromising site performance? Practice and preparation, which began five days before the big game.

Tuesday, February 2, 2016:

The team used this day to test the runbook and deployment procedure used during the big game as the ads hit the airwaves. Remembering that screens of every size will be hitting this instantly as soon as the URL is shown, along with a host of other traffic sources and social feeds, we were now able to use this critical step to load-test the site and validate assumptions for game day.

Everything from the code to the infrastructure was optimized to make sure that there were no limiting factors, no bottlenecks, and no slowdowns. The site performance needed to deliver the creative experience from end to end.

Wednesday, February 3, 2016:

GLG ran overnight load tests on the teaser page and pushed the page to its absolute limit. This gave us an idea of where the breaking points were and how to plan for them. We tried this two ways: First, a steady escalation of traffic to see what and when resources started to drain out. Second, an all-out blitz of simulated traffic that took us to a perceived breaking point immediately. This was a more likely traffic scenario—all of it hitting at once. In years past, we have seen this pattern of a sudden wall of traffic, followed by a steady ramp-down as people get back into the game, and then a return surge later at night as ad critics start to flood the ether.

Thursday–Saturday, February 4–6, 2016:

With the load-test data providing GLG the best insight possible, we made adjustments to the caching strategy and content delivery networks (where we offloaded more static assets to the cloud). Finally, the team assessed how the dynamic data affected the load time of the site. GLG worked the next three days enhancing and testing these optimizations.

The result speaks for itself. 

GLG

Daily Thread

We put all our social feeds in one place to give you a glimpse into GLG: what we’re working on, what we’re thinking about, and what inspires us.