The translation has come to an end.
The translation has come to an end.
Exit Zen
186
../../principles/sharing.rst:1041
English
A congestion control scheme for our simple transport protocol could be implemented as follows.
0/940 · 94
187
../../principles/sharing.rst:1060
English
In the above pseudocode, `cwin` contains the congestion window stored as a real number of segments. This congestion window is updated upon the arrival of each acknowledgment and when congestion is detected. For simplicity, we assume that `cwin` is stored as a floating point number but only full segments can be transmitted.
0/3240 · 324
188
../../principles/sharing.rst:1063
English
As an illustration, let us consider the network scenario above and assume that the router implements the DECBit binary feedback scheme [RJ1995]_. This scheme uses a form of Forward Explicit Congestion Notification and a router marks the congestion bit in arriving packets when its buffer contains one or more packets. In the figure below, we use a `*` to indicate a marked packet.
0/3800 · 380
189
../../principles/sharing.rst:1218
English
When the connection starts, its congestion window is set to one segment. Segment `S0` is sent an acknowledgment at roughly :math:`t_{0}+4`. The congestion window is increased by one segment and `S1` and `S2` are transmitted at time :math:`t_{0}+4` and :math:`t_{0}+5`. The corresponding acknowledgments are received at times :math:`t_{0}+8` and :math:`t_{0}+10`. Upon reception of this last acknowledgment, the congestion window reaches `3` and segments can be sent (`S4` and `S5`). When segment `S6` reaches router `R1`, its buffers already contain `S5`. The packet containing `S6` is thus marked to inform the sender of the congestion. Note that the sender will only notice the congestion once it receives the corresponding acknowledgment at :math:`t_{0}+18`. In the meantime, the congestion window continues to increase. At :math:`t_{0}+16`, upon reception of the acknowledgment for `S5`, it reaches `4`. When congestion is detected, the congestion window is decreased down to `2`. This explains the idle time between the reception of the acknowledgment for `S*6` and the transmission of `S10`.
0/10970 · 1097
190
../../principles/sharing.rst:1220
English
In practice, a router is connected to multiple input links. The figure below shows an example with two hosts.
0/1090 · 109
191
../../principles/sharing.rst:1279
English
In general, the links have a non-zero delay. This is illustrated in the figure below where a delay has been added on the link between `R` and `C`.
0/1460 · 146
192
../../principles/sharing.rst:1326
English
Footnotes
0/100 · 9
195
../../principles/sharing.rst:1331
English
Some networking technologies allow to adjust dynamically the bandwidth of links. For example, some devices can reduce their bandwidth to preserve energy. We ignore these technologies in this basic course and assume that all links used inside the network have a fixed bandwidth.
0/2770 · 277
196
../../principles/sharing.rst:1333
English
In this section, we focus on congestion control mechanisms that regulate the transmission rate of the hosts. Other types of mechanisms have been proposed in the literature. For example, `credit-based` flow-control has been proposed to avoid congestion in ATM networks [KR1995]_. With a credit-based mechanism, hosts can only send packets once they have received credits from the routers and the credits depend on the occupancy of the router's buffers.
0/4510 · 451
197
../../principles/sharing.rst:1335
English
For example, the measurements performed in the Sprint network in 2004 reported more than 10k active TCP connections on a link, see https://research.sprintlabs.com/packstat/packetoverview.php. More recent information about backbone links may be obtained from caida_ 's real-time measurements, see e.g. http://www.caida.org/data/realtime/passive/
0/3450 · 345
The translation has come to an end.