Monitoring thread. I have a service that always responds in 100ms, except for 5 minutes of every hour where it stalls completely. If a serial client (i.e. it sends a request, waits for it to finish, sends next) is monitoring latency, what 99th percentile does it measure?

Same service, but with a client sending requests every 1s whether the last one has completed or not. What 99th percentile latency does it see?

Same service, serial client with a 1s request timeout. What 99th percentile latency does it measure?