To invoke the problem case, the session must have expirienced loss (to
pull down ssthresh to a low value). Thereafter, provided there is no
additional data loss but the session runs application limited for longer
periods (sender only sends a small amount of data, waits for less than
RTO interval, and sends only a few kB every 10s of milliseconds),
the cubic-calculated cwnd may end up negative or beyond INT_MAX.
Some overflows (w_tf) are apparent during low-latency sessions immediately,
other (ticks_since_cong) may only happen after very long periods of time (~3 weeks) and cubic_cwnd only for large K/max_cwnd values.
Once the sender application starts to provide large amounts of data,
a line-rate traffic burst is unavoidable as cwnd may have grown to many
MB or GB in size.