T-Mobile
“Kim’s Data Stash”
Role
Genres
For the Super Bowl, our team designed and developed an interactive campaign starring Kim Kardashian, for T-Mobile. Given the popularity of not only the NFL’s biggest event of the year, but also all of the brands involved, we knew the pressure was on to craft something perfect for our audience, while scalable for the surge in traffic we were sure to witness.
When finished, the infrastructure and software were built to sustain 10 million page views over the course of 4 hours. Behind the scenes, we provisioned and utilized a fleet of servers (over 5100) spanning multiple regions, each containing several availability zones.
This massive infrastructure roll-out for the Super Bowl campaign included dynamic video generation, Node.js-based servers, web sockets (Socket.IO), virtual private clouds (VPCs), Load Balancers, and Nginx-based servers built on the Amazon Web Services (AWS) network.
I worked directly with teams from T-Mobile and AWS to architect a purpose-built solution to support this Super Bowl campaign and accomplished our objectives in a mere 2.5 weeks.
Combining the talents of T-Mobile, Amazon, Twitter, Kim, and our team we produced a highly successful marketing campaign with 6-digit (~35,000) real-time active users:
We used just about every AWS cloud product (e.g., EC2, VPA, Auto Scaling, S3, RDS, ElasticCache, CloudFront, Elastic Load Balancing, etc…) to construct a stable, traffic-tested, set of environments that could produce thousands of dynamic user-generated content (UGC) videos within the span of the event (4 hours).
Under the veil of this one-page scrolling microsite, was a powerhouse of video renders, web sockets, HAProxy servers, subnets and webservers that were loaded and stress tested for optimal performance.
To monitor the infrastructure and software components we employed Nagios while utilizing Elasticsearch, along with Logstash and Kibana (then ELK Stack and now an Elastic Stack) to monitor the entire volume of server logs. Furthermore, all Elastic Load Balancers were pre-warmed in preparation of the traffic demands and configured with Auto Scaling groups.