Derive the RFC 6298 retransmission timeout and the total time TCP spends retrying before it gives up.
RFC 6298 sets the retransmission timeout from the smoothed RTT plus four times its variance, then doubles it after every failed attempt. Summing that geometric series gives the wall-clock time before the stack reports the connection dead. This number is your worst-case failure detection time: it decides how long a client hangs on a dead server, so it has to be shorter than the health check and failover budget above it.
Connection Timeout
RTO = max(floor, SRTT + 4 × RTTVAR), and the backoff sum is RTO × (2^retries − 1).
RTO = max(floor, SRTT + 4 × RTTVAR), and the backoff sum is RTO × (2^retries − 1). RFC 6298 sets the retransmission timeout from the smoothed RTT plus four times its variance, then doubles it after every failed attempt. Summing that geometric series gives the wall-clock time before the stack reports the connection dead.
This number is your worst-case failure detection time: it decides how long a client hangs on a dead server, so it has to be shorter than the health check and failover budget above it.
This calculator takes 4 inputs: Smoothed round-trip time (SRTT), RTT variance (RTTVAR), Retransmission attempts, Minimum RTO floor. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
The floor guards against a spuriously low RTT estimate causing needless retransmissions. In datacentre networks with sub-millisecond RTTs the floor dominates completely, which is why incast recovery is often measured in hundreds of milliseconds.