Content deleted Content added
Pansapiens (talk | contribs) m Link to definition of "goodput" |
m →Protocol: reference added |
||
Line 21:
FASP's flow control algorithm, unlike TCP's, completely ignores packet drops. Instead, it acts on changes in measured packet delivery time. When that is growing, queues are getting longer and channel bandwidth is exceeded; falling, queues are getting shorter. Acting on this information is complicated because the receiver has it and the sender needs it, but its lifetime is often less than the transmission delay; and measurements are noisy. Thus, the sender uses a predictive filter fed updates from the receiver.<ref>{{cite web | url=https://news.ycombinator.com/item?id=21909190 | title=Ex Aspera Dev here. I did the encryption and early parallel work. There is a lot... | Hacker News }}</ref>
The transmission rate is chosen to match and not exceed the available channel bandwidth, and trigger no drops, accounting for all traffic on the channel.<ref>{{Cite book |last=Klimek |first=Ivan |url=https://www.yumpu.com/en/document/view/53450087/technicka-univerzita-v-kosiciach-stargatecnlsk |title=Wide Area Network Traffic Optimization |publisher=Technical University of Košice |year=2011 |___location=Košice |pages=49 |language=en |department=Faculty of Electrical Engineering and Informatics}}</ref> By contrast, TCP slowly increases its rate until it sees a packet drop and falls back, interpreting any drop as congestion. On a channel with long delay and frequent packet loss, TCP never approaches the actual bandwidth available. FASP cooperates with TCP flows on the same channel, using up bandwidth TCP leaves unused.
==See also==
| |||