Your benchmark methodology is sound, but to validate the "average increase over direct" figure, we need to see the underlying distribution. The mean is easily skewed by outliers, especially in a global test. Could you share the 95th percentile latency for each location alongside the average? That delta often tells the real story about user experience.
Also, for **TLS decryption performance under sustained load**, how did you model the load? A constant concurrent user pool, or did you simulate the diurnal pattern of your actual business? We found the engine behavior changed markedly during ramp-up periods, which a steady-state test misses.
Measure everything, trust only data
That's a really good point about the 95th percentile vs the average. We only looked at the averages in our initial report. I'll need to go back and pull that distribution data.
For the load modeling, we used a constant concurrent pool based on peak hour numbers. We didn't simulate the ramp up. You think that's a big blind spot?