Scaling to 700,000 Concurrent Connections

Web publishers were looking for alternatives to ads, and browser-based crypto mining was emerging as a viable option. The problem wasn't the idea, it was scale.

Browser mining meant handling massive, spiky, unpredictable traffic while staying profitable on razor-thin margins. At peak usage, even small inefficiencies could wipe out earnings or take the entire system down.

What started as a personal learning project around mining and protocol mechanics suddenly became production infrastructure under real-world load.

In 2017, Crypto-Loot started as a hobby. I originally built it to better understand how mining worked at a protocol level and to make participation accessible to non-technical users: no sketchy downloads, no complex configs, just open a browser and learn.

Then Coinhive took off.

I already had the core tech, but they had scale. Instead of competing on raw efficiency, I competed on economics, launching with a 12% fee versus their 30%, and opening the platform to webmasters.

Traffic exploded almost overnight.

The initial system wasn't built for this. Servers went down constantly. Early on, I made every classic mistake: vertical scaling, rushed patches, and firefighting. It became clear the architecture needed a ground-up redesign focused entirely on throughput and cost efficiency.

Over the next month, I rebuilt the platform while it was live:

• •Broke the system into a microservices architecture
• •Rewrote most backend services in Golang for lower latency and better concurrency
• •Optimized stratum proxy behavior for low-difficulty browser shares
• •Designed a tiered share-validation pipeline that selectively verified mining submissions based on historical accuracy, submission patterns, and cost thresholds - reducing verification overhead while preventing invalid or malicious share amplification attacks
• •Introduced Anycast routing so miners connected to the nearest edge proxy and pool region

Because there was no real documentation on scaling protocol-embedded infrastructure at the time, I leaned heavily into the Monero ecosystem: reading source code, joining IRC channels, and collaborating directly with contributors and pool operators. Several of the scaling strategies we landed on came out of real-world experimentation and failed PoCs.

The business stayed lean and profitable the entire time: no funding, no burn, just engineering and iteration.