☕ Kev's Tech Notes

I Thought My Nginx Config Was Fine. Then Black Friday Hit.

Last year around this time, I was pretty cocky. Config looked clean, tests passed. Then traffic spiked 8x in 20 minutes and everything fell apart. Not because of CPU. Because of one stupid default I never changed: worker_connections. Here's what actually happened, and the 5 lines I changed that fixed it.

worker_processes auto;
events {
  worker_connections 4096; # default is 512. Yeah. 512.
  use epoll;
  multi_accept on;
}

If you're running a real site and you still have the default 512, go fix it right now. I'll wait.

BBR vs CUBIC: I Actually Tested It, So You Don't Have To

Everyone says "just enable BBR, bro." I did. On a server from Dallas to Singapore. And guess what? It was worse for the first two days. Turns out BBR is amazing for high-latency, lossy links, but if your RTT is already low, you won't see magic. I ran iperf3 for a week, here's the real numbers.

Dallas → Singapore, CUBIC: ~42 Mbps average, huge jitter. BBR: ~118 Mbps, way more stable. Dallas → Chicago? Basically the same. Don't blindly copy sysctl settings from Reddit.

Zero-Trust Sounds Cool Until You Have To Explain It To Your Boss

I spent 3 weeks setting up mTLS + identity-aware proxy for our internal tools. Technically perfect. Then my PM asked, "So... do we still need a VPN?" I realized I had built the perfect system that nobody understood. Here's how I finally explained zero-trust in plain English, and the simple setup I actually kept.