Load Testing an NFT Burn for the Super Bowl: Part 1 — The Mission
This is Part 1 of a four-part series on load testing and scaling the backend behind Limit Break’s Super Bowl NFT token burn. Part 2 covers our load testing strategy, Part 3 covers the infrastructure journey, and Part 4 covers what actually happened on game day.
On February 12, 2023, during Super Bowl LVII, Limit Break — the Web3 gaming studio behind DigiDaigaku, led by Gabriel Leydon — aired a reported $6.5 million national TV spot. Front and center was a QR code, sending millions of viewers straight into an NFT claim flow backed by a token burn on our end.
If you want to see the ad that started all of this, here it is:
That single QR code was the whole problem. A Super Bowl ad doesn’t ramp traffic up gradually — it fires a wall of concurrent users at your backend the instant it airs, and that wall arrives whether or not your infrastructure is ready for it. There’s no autoscaling policy in the world that reacts fast enough to a spike that starts at zero and hits its peak within seconds of the spot going live.
Our job was to make sure the NFT token burn backend could take that hit — not gracefully degrade, not queue people for five minutes, but actually serve the request volume of a Super Bowl audience acting on a call to action all at once.
What we were actually building for
The collection was 10,000 free Dragon Egg NFTs in total, and roughly 4,000 of those were specifically allocated to whoever scanned the on-screen QR code during the broadcast and claimed fast enough. The service in front of that QR code needed to:
- Authenticate and process NFT token burns at a volume nobody could give us a confident number for in advance — “roughly 4,000, claimed by whoever’s fastest, out of a Super Bowl audience” is not a number you can plan capacity around in the normal sense
- Survive a traffic curve that looks nothing like normal product usage — a near-vertical spike, not a diurnal curve
- Stay correct under load — this is a blockchain-adjacent transaction path, not a page view, so failure modes matter more than they would for, say, a marketing landing page
We didn’t have the luxury of “let’s see how it performs in production and iterate.” There is no second Super Bowl ad next week if the first one falls over. Whatever confidence we had going into game day had to come from testing beforehand, at a scale that actually resembled the real thing.
That meant leaning heavily on the AWS Distributed Load Testing on AWS solution, and eventually stretching it well past its default assumptions to generate enough load to be believable.
QR code on screen"] --> Viewers["Millions of viewers,
same ~45s window"] Viewers --> Edge["Edge / load balancing layer"] Edge --> Backend["NFT token burn backend"] Backend --> Chain["Blockchain settlement"] subgraph Pre["Weeks before game day"] DLT["AWS Distributed Load Testing
(6 accounts x 2 regions, JMeter)"] -.simulates.-> Edge end
Everything downstream of the QR code had to survive a spike that looked like the diagram above — not a ramp, a step function. Parts 2 through 4 cover how we got there.
The shape of the rest of this series
The rest of this series walks through how we got there:
- Part 2 covers the load testing setup itself — why a single AWS account and region couldn’t generate enough load, and how spreading JMeter across six AWS accounts and two regions got us to a validated 1.2M-request test scale.
- Part 3 covers the backend and network journey behind three ALBs — walking away from third-party scaling vendors that capped out at 10k requests per second, an NLB detour on AWS support’s advice, a CloudFront-to-Cloudflare load balancer swap, running out of IPs in an undersized VPC subnet, the pre-scaling strategy that got us to 300 pods across 60 nodes, and why the burn counter ended up living in ElastiCache instead of the blockchain data pipeline.
- Part 4 covers what actually happened when the ad aired — the real peak, the 45-second sellout, and why we ended up routing people away from Twitter and back to our own domain mid-event.
None of this was theoretical. Every number in this series came from either a load test we ran or the traffic we actually saw during the Super Bowl broadcast.
Sources for the public details referenced in this post: CoinDesk — DigiDaigaku NFTs’ Price Rises After $6.5M Super Bowl Ad, VentureBeat — Limit Break will do free NFT mint during Super Bowl commercial.