Editing Round-trip delay (section)

==Protocol design==
RTT is a measure of the amount of time taken for an entire message to be sent to a destination and for a reply to be sent back to the sender. The time to send the message to the destination in its entirety is known as the [[network latency]], and thus RTT is twice the latency in the network plus a processing delay at the destination. The other sources of delay in a network that make up the network latency are processing delay in transmission, propagation time, transmission time and queueing time. Propagation time is dependent on distance. Transmission time for a message is proportional to the message size divided by the bandwidth. Thus higher bandwidth networks will have lower transmission time, but the propagation time will remain unchanged, and so RTT does fall with increased bandwidth, but the delay increasingly represents propagation time.<ref>{{cite book |last1=Forouzan |first1=Behrouz A. |last2=Fegan |first2=Sophia Chung |title=Data communications and networking |date=2007 |publisher=McGraw-Hill Higher Education |location=Boston |isbn=9780072967753 |edition=4th}}</ref>{{rp|90,91}}

Networks with both high bandwidth and a high RTT (and thus high [[bandwidth-delay product]]) can have large amounts of [[data in transit]] at any given time. Such ''long fat networks'' require a special protocol design.<ref>{{citation |url=http://www.networkworld.com/article/2176641/tech-primers/are-your-pipes-too-big-.html |archive-url=https://web.archive.org/web/20140605161549/http://www.networkworld.com/article/2176641/tech-primers/are-your-pipes-too-big-.html |url-status=dead |archive-date=June 5, 2014 |title=Are your pipes too big? |author=Brian Heder |work=[[Network World]] |date=May 6, 2014 |access-date=2016-01-09}}</ref> One example is the [[TCP window scale option]].

The RTT was originally estimated in TCP by:

:<math>\mathrm{RTT} = \alpha \cdot \mathrm{old\_RTT} + (1 - \alpha) \cdot \mathrm{new\_round\_trip\_sample}</math>

where <math>\alpha</math> is constant weighting factor (<math>0 \leq \alpha < 1</math>).<ref>{{cite book |author=Douglas E. Comer |author-link=Douglas Comer |year=2000 |title=Internetworking with TCP/IP - Principles, Protocols and Architecture |edition=4th |publisher=Prentice Hall |isbn=978-0-13-018380-4 |page=226}}</ref> Choosing a value for <math>\alpha</math> close to 1 makes the weighted average immune to changes that last a short time (e.g., a single segment that encounters long delay). Choosing a value for <math>\alpha</math> close to 0 makes the weighted average respond to changes in delay very quickly. This was improved by the [[Jacobson/Karels algorithm]], which takes standard deviation into account as well. Once a new RTT is calculated, it is entered into the equation above to obtain an average RTT for that connection, and the procedure continues for every new calculation.