request time is usually the sum of upstream time + kong proxy latency. It seems the upstream time is so large that it dwarfs the proxy latency.
If you want to measure the total request time for a request that initiates at kong and finishes, use request time. If you want to measure the performance of your upstream service from kong’s perspective, use upstream time.
the above graph has shows Request time and upstream time is same but upstream time should be less as per your comment above.
"request time is usually the sum of upstream time + kong proxy latency. "
as we observed it is not same for 95th and 99th percentile as per above graph.