The paper formally analyzes the definitions used for Retransmission Time Out (RTO) in the standard for TCP, and shows that it allows for spurious timeouts, that is, for events where the RTO is too low and congestion control algorithms switch too fast into a mode where transmission is slowed down almost to a halt. Code inspection reveals that kernels do not follow the standard to the letter, and create wraparounds. This paper shows that the wraparounds do not scale and a simpler change is more effective. The analysis is done using standard methods of probability theory.