I was reading about data-center-tcp DCTCP and the paper was talking about the low retransmission timer (RTT) of about 250µs, but didn't state how that is being done.
Somewhere else I read that DCTCP is using explicit congestion notification, which reduces the congestion control window. Does this lead to shorter RTTs?
So, which method does actually guarantee or achieves the low RTT compared to lets say TCP Tahoe or Reno?
