2. Loss-Based Congestion Control
? Most of today’s network traffic uses
an extremely simple algorithm that
avoids networks collapsing under
extreme overload:
? Continually increase the amount of
data being pushed into the network
? Until a network queue overloads and
the sender detects the consequent
packet loss
? Halve the sending rate and do it again!
3. Refinements
? Change the congestion avoidance inflation algorithm
? Try to detect the difference between isolated damage packet loss and
queue overload loss
? Better understand the relationship between network buffers and
protocol performance
4. Is there a “better” way?
Trigger the congestion response at the
onset of queue formation rather than
at the point of catastrophic queue
collapse
5. BBR
? Detect queue formation through
pulsed testing and delay
sensitivity
12. ECN Control Loop
? A router “marks” IP packets at the onset of queue formation with a
bit signal
? The Receiver echoes this bit up into the transport protocol reverse
flow
? The sender reduces its sending window size (and notifies the receiver
that it was performed this window reduction)
IP
TCP
13. ECN Control Loop
? A router “marks” IP packets at the onset of queue formation with a
bit signal
? The Receiver echoes this bit up into the transport protocol reverse
flow
? The sender reduces its sending window size (and notifies the receiver
that it was performed this window reduction)
IP
TCP
14. IP Header
Version IHL Total Length
Flags
Identification Fragment Offset
Time To Live
Source Address
Destination Address
Options Padding
Protocol Header Checksum
Type of Service
Precedence
0 1 2 3 4 5 6 7
D T R 0 0
Version Traffic Class Flow Label
Payload Length Next Header Hop Limit
Source Address
Destination Address
ECN Bits
0 0 – Non-ECN Capable Transport
0 1 – ECN Capable TransporT
1 0 - ECN Capable Transport
1 1 – Congestion Experienced
15. TCP
TCP
Destination Port
Source Port
Sequence Number
Acknowledgment Number
Data
offset
F
I
N
S
Y
N
U
R
G
A
C
K
P
S
H
R
S
T
Window
Checksum Urgent Pointer
Padding
TCP Options
TCP Data
E
C
E
C
W
R
ECE – receiver back to sender – CE received
CWR – sender to receiver – Congestion Window Reduced
SYN+ECE+CWR – ECN capable on session start
SYN+ACK+ECE – ECN capable response
16. ECN Measures
Packet Count (by remote IP addresses) for Feb/March 2025
0. IP Sources 303,545,388
1. IP ECT 6,815,753 (2.45% of sources)
2. IP CE 1,098,965 (16.12% of ECT sources)
17. ECN Measures
Packet Count (by remote IP addresses)
0. IP Sources 303,545,388
1. IP ECT 6,815,753 (2.45% of sources)
2. IP CE 1,098,965 (16.12% of ECT sources)
3. TCP ECN Opt 7,478,207 (2.46% of sources)
4. TCP ECE (Rec’d) 20,862 (0.27% of TCP sources)
5. TCP CWR (Rec’d) 335,209 (4.48% of TCP sources)