Partie 9 Congestion Control in TCP Contrôle de

Congestion Control
TCP
Contrôle de congestion
Partie 9
Congestion Control in TCP
Principles
Phases: Slow Start, Congestion
Avoidance, Fast Recovery
●
TCP s’appuie sur IP (sans connexion)
●
TCP opère de bout-en-bout (détection de la
congestion peu fiable et indirecte)
●
Chaque source TCP agit «seule» (pas de
coopération entre sources)
●
Pertes de paquets interprétées comme un
signalement de la congestion
588
589
Fenêtre d'émission
Fenêtre d'émission
F octets
●
RTT
●
●
Fenêtre d'émission - nombre d'octets non-acquittés
– F = min (cwnd, W)
»cwnd - variable maintenu par la source
» W - fixé par la destination, champs fenêtre
annoncée
F - le nombre d'octets non acquittés
– débit = F/RTT (formule de Little)
Si congestion
– RTT augmente, réduction automatique
du débit de la source
– mécanisme de contrôle : diminuer F
590
591
TCP and Congestion Control
Phases de contrôle de congestion
●
●
Connexion TCP peut être dans une de trois
phases (point de vue du contrôle de congestion)
– démarrage (slow start), après une perte
détectée par une temporisation
– récupération (fast recovery), après une perte
détectée par retransmission rapide (fast
retransmit)
– évitement (congestion avoidance), tous les
autres cas
●
●
592
TCP is used to avoid congestion in the Internet
– a TCP source adjusts its window to the congestion
status of the Internet
(slow start, congestion avoidance)
– this avoids congestion collapse and ensures some
fairness
TCP sources interprets losses as a negative feedback
– use to reduce the sending rate
UDP sources are a problem for the Internet
– use for long lived sessions (ex: RealAudio) is a
threat: congestion collapse
– UDP sources should imitate TCP : “TCP friendly”
593
1
Congestion Control
TCP
Slow Start and Congestion Avoidance
Increase/decrease
connection opening: twnd = 65535 B
cwnd = 1 seg
●
retransmission
timeout:
Congestion
Avoidance
Slow Start
exponential
increase for cwnd
until cwnd = twnd
- multiplicative
decrease for twnd
- cwnd = 1 seg additive
●
increase for twnd,
cwnd = twnd
cwnd = twnd
retransmission
timeout:
notes
this shows only 2 states out of 3
twnd = target window
- multiplicative
decrease for twnd
- cwnd = 1 seg
Multiplicative decrease
– twnd = 0.5 × current_window
– twnd = max (twnd, 2 × MSS)
Additive increase
– for each ACK
»twnd = twnd + MSS × MSS / twnd
(w←w + 1)
»twnd = min (twnd, max-size)
(64KB)
594
595
Increase/decrease
●
twnd Additive Increase
twnd
=
1 seg
Exponential increase
– for each ACK
»cwnd = cwnd + MSS
»if (cwnd == twnd) go to avoidance phase
●
●
2
3
4
during one round trip + interval between packets:
increase by 1 packet (linear increase)
(equivalent to twnd = twnd + 1/twnd if TCP
would have all segments of length MSS)
596
597
Slow Start
cwnd = 1 seg
2
3 4
Slow Start
5 678
●
/ * exponential increase for cwnd */
non dupl. ack received during slow start ->
cwnd = cwnd + seg (in bytes)
if cwnd = twnd then transition to
congestion avoidance
●
purpose of this phase: avoid burst of data after a
retransmission timeout
●
598
window increases rapidly up to the stored value of twnd
(this stored value is called ssthresh in the literature)
599
2
Congestion Control
TCP
Slow Start and Congestion Avoidance
Example
Slow Start and Congestion Avoidance
Example
Bytes
twnd
B
A
C
60
cwnd
Bytes
twnd
B
A
C
30
60
0
cwnd
0
30
1
2
3
4
5
6
7
8
9
seconds
B slow C
congestion avoidance start congestion avoidance
0
0
1
2
3
4
5
6
7
8
9
seconds
created from data from: IEEE Transactions on
Networking, Oct. 95, “TCP Vegas”, L. Brakmo
and L. Petersen
600
created from data from: IEEE Transactions on
Networking, Oct. 95, “TCP Vegas”, L. Brakmo
and L. Petersen
601
Congestion Control States
Slow Start and Congestion Avoidance
Example
B slow
start
congestion avoidance
new connection:
C
congestion avoidance
cwnd = twnd:
Congestion
Avoidance
Slow Start
- exponential
increase
- additive
increase
retr. timeout:
fast
retransmit:
fast
retransmit:
retr. timeout:
Fast
Recovery
retr. timeout:
- exponential
increase
beyond twnd
expected ack received:
created from data from: IEEE Transactions on
Networking, Oct. 95, “TCP Vegas”, L. Brakmo
and L. Petersen
602
Fast Recovery Example
Fast Recovery
●
Bytes
A
B C D
F
E
30
Fast Recovery
0
– cwnd = twnd + 3 × MSS (exp. increase)
– cwnd = min (cwnd, 64K)
– retransmission of the missing segment (n)
●
twnd
60
Multiplicative decrease
– twnd = 0.5 × current-size
– twnd = max (twnd, 2 × MSS)
●
603
0
E
cwnd
1
2
3
4
5
6
seconds
F
For each duplicated ACK
A-B, E-F: fast recovery
C-D: slow start
– cwnd = cwnd + MSS (exp. increase)
– cwnd = min (cwnd, 64K)
– send following segments
604
605
3
Congestion Control
TCP
Congestion Control: Summary
●
Congestion Control: Summary
Congestion control aims at avoiding congestion collapse
in the network
With TCP/IP, performed in end-systems, mainly with
TCP
TCP Congestion control summary
Principle: sender increases its sending window until
losses occur, then decrease
●
●
additive increase
●
●
Target window
– additive increase (no loss)
– multiplicative decrease (loss)
3 phases
– slow start:
» starts with 1, exponential increase up to twnd
– congestion avoidance
» additive increase until loss or no increase
– fast recovery
» fast retransmission of one segment
multiplicative decrease
loss
additive increase
loss
slow start
606
607
Congestion Control: Summary
●
●
●
TCP Friendly Applications
Slow start entered at setup or after retransmission
timeout
Fast recovery entered at fast retransmit
Congestion avoidance entered when cwnd = twnd
608
●
●
All TCP/IP applications that generate long lived
flows should mimics the behavior of a TCP source
– RTP/UDP flow sending video/audio data
Adaptive algorithm
– application determines the sending rate
– feedback - amount of lost packets, loss ratio
– sending rate = rate of a TCP flow experiencing the
same loss ratio
609
TCP Loss - Throughput formulae
θ=L C
T q
●
●
TCP connection with
– RTT T
– packet size L
– average packet loss ratio q
– constant C = 1.22
Transmission time negligible compared to RTT,
losses are rare, time spent in Slow Start and Fast
Recovery negligible
610
4