Fast, but not magic
I wired one of our busiest endpoints through SpeedX because latency was getting embarrassing. Setup was mostly copy-paste from their docs, with a couple gotchas that any grown-up engineer can handle. First run shaved a noticeable chunk off p95, not a fairy tale zero, but real. The status page is blunt about incidents, which I weirdly appreciate. Billing metrics are readable without a magnifying glass. I wish their SDK logs were a tad quieter out of the box. Still, the thing is snappy and it stays out of my way.
0 comments
Replying to Harrison Tang
The comment must be at least 10 characters.