Transcript TCP w/o Congestion Control

TCP/IP Over Lossy Links
TCP SACK without Congestion Control
Organization
1. The History of TCP
2. Current TCP Congestion Control
3. Design Ideas: no congestion control at all
4. Measurement Results
5. Future TCP Congestion Control?
6. Conclusion
1. The History of TCP (incomplete)
„Old Tahoe“:
slow start and congestion avoidance after a lost packet, wait
for a timeout, then perform slowstart to recover
„Tahoe“ (`88):
also contains fast recovery. After the reception of a triple
duplicate ACKs, performs fast retransmit followed by a
slowstart [Jaco88].
„Reno“ (`90):
fast retransmit extended by fast recovery.
But
in case multiple packets are lost from a window of data,
very likely, TCP „Reno“ has to wait for a timeout followed
by slow start to recover [Jaco90].
[Jaco88] Van Jacobson, “Congestion Avoidance and Control”, ACM SIGCOMM '88
[Jaco90] Van Jacobson, “Modified TCP Congestion Avoidance Algorithm”, email to [email protected], April 1990
1. The History of TCP (incomplete)
Vegas (`94, never actually implemented):
modified congestion behavior. By measuring the output queue size,
equilibrium is detemined (Little‘s Law of queuing theory)
[BraMalPet94].
„New Reno“ (~`94):
small optimization of TCP Reno, uses partial ACKs as an indication that
the following packet got lost and immediately retransmits it without
leaving fast recovery. [Hoe96].
TCP SACK(~`95):
added the SACK option within ACKs. Allows receiver to specify the
range of packets that were received out of order. [MatmahFlRo96].
In contrast to all previous flavors, more than
one packet per RTT during fast recovery can be send
[BraMalPet94]
Lawrence S. Brakmo, Sean W. O'Malley, Larry L. Peterson, „TCP Vegas: New Techniques for
Congestion Detection and Avoidance“, Sigcomm 1994
[MatMahFlRo96] M. Mathis, J. Mahdavi, S. Floyd, A. Romanow, „TCP Selective Acknowledgement Options“, RFC
2018, April 1996
[Hoe96]
Janey C. Hoe, “Improving the start-up behavior of a Congestion Control Scheme for TCP, Sigcomm
1996
1. The History of TCP (incomplete)
FACK (`96):
In fast recovery, congestion window is fixed. TCP FACK uses „pipe“
variable to estimate the data in the network by taking the
transmitted out-of-order segments into account, thus preventing
premature timeouts [MatMah96].
[MatMah96] M. Mathis and J. Mahdavi, "Forward acknowledgement: Refining TCP congestion
control“, ACM Computer Communication Review, Oct 1996.
2. Congestion Control: slow start
Slow Start: with every received
ACK, double the number of packets
sent.
Slow start adds a window to the
sender's TCP: the congestion
window, called cwnd as well as a
variable called ssthres
exponential growth of
the Congestion Window
up to ssthres, then
linear growth
Figure taken from [Jaco88]
The congestion window is flow control imposed by the sender.
It is based on the sender's educated guess of perceived network
congestion. Congestion Control assumes that packets are only lost
due to overfull queues.
2. Congestion Control: congestion avoidance in TCP Reno
window
SS
time
CA
SS: Slow Start
CA: Congestion Avoidance
Fast retransmission/fast recovery
2. TCP Congestion Control
TCP send rate is determined by three windows:
win=min(snd_cwnd,snd_wnd,snd_bwnd)
Congestion window
assumed bottlenecks: queue sizes
in the network
Advertised window
assumed bottleneck: receiver’s buffer
Bandwidth window, “ACK clock”
assumed bottleneck: link capacity
3. Design Idea: no congestion control at all
The sending rate is given by:
Now:
win=min(snd_cwnd,snd_wnd,snd_bwnd)
win=min(snd_bwnd,snd_wnd)
Without SACK, this flavor of TCP will perform poorly
(waste of bandwidth on duplicate ACKs that can lead to timeouts)
SACK gives us control over the now “static” window
UDP?
In contrast to UDP, the protocol will still guarantee for
in-order delivery and will adopt to the link capacity.
4. Measurements: the emulation environment
Node 1
Node 2
Sender
with modified TCP
100Mbps
Receiver
with modified TCP
10Mbps
Router 1
Sender
with original TCP
Node 0, „base“
Router 2
100Mbps
Receiver
with original TCP
Node 3, „base“
delays for each link:
On
all100ms
links,anddelay=10ms.
10ms,
400ms
Loss rates varied from p=0, p=0.001, p=0.01, p=0.1 to p=0.2
resulting RTTs: 60ms, 600ms, 2.4 s
Packt
loss events are uniformly distributed.
Experiment has been set up in the emulab environment [emu].
[emu] www.emulab.net
4. Measurements: collection of Data
1.
Initialize tcpdump on the to-be-observed node:
sudo tcpdump -c num -w file -i if &
2.
Start ttcp on the receiving node
ttcp -r -s src
3.
Start ttcp on the sending node
ttcp -t -s -n num dst
num
file
if
src
dst
-
number of packets to be captured
name of the dump file
interface to be listened to
IP address of the sending node
IP address of the receiving node
Traces have been analyzed off-line with ethereal [eth].
[eth] www.etheral.com, packet sniffer and analyzer
4. Measurements: SACKBASE, lossless link
tcpdump started on the sender,
zoomed into connection „set up“ phase
size of the send window in case
of a link bottleneck: bandwidth-delay product
advertised receiver
Window+Seq # of last ACK
seq # ACKs received
4. Measurements: SACKEXP, lossless link
tcpdump started on the sender,
zoomed into connection „set up“ phase
4. Definitions: Throughput and Goodput
Throughput:
The number of packets per unit of time
transmitted between sender and receiver.
Goodput:
The number of packets per unit of time
forwarded from sender to receiver, minus any
packets lost or retransmitted, minus ACKs.
4. Measurements: summary of results, competing flows
[KemXinKas04] R. Kempter, B. Xin, S. Kumar Kasera, “Towards a Composable Transport Protocol: TCP without
Congestion Control”, SIGCOMM 2004 poster presentation
4. Measurements: summary of results, competing flows
[KemXinKas04] R. Kempter, B. Xin, S. Kumar Kasera, “Towards a Composable Transport Protocol: TCP without
Congestion Control”, SIGCOMM 2004 poster presentation
5. Future TCP Congestion Control? ECN bit
Another way to do congestion control: the ECN bit
Instead of dropping packets, a router sends a TCP an explicit
message stating that the network is becoming congested.
The network determines an explicit rate for a sender [RamFloyd99].
Hop-by-Hop vs. End-to-end congestion control
ECN
bit
IP Packet
0
IP Header
1
IP Header
ECN
Echo
TCP ACK
1
[RamFloyd99] Ramakrishnan, K.K., and Floyd, S., A Proposal to add Explicit Congestion Notification
(ECN) to IP. RFC 2481, January 1999
6. Conclusion
• Ambiguity in packet loss  current TCP congestion control
 low throughput over lossy links
• TCP w/o congestion control and without SACK is inefficient
• p=0.1%, SACKEXP = 91% of goodput of lossless link,
TCP SACK = 65% (at identical efficiencies of 91%)
• As losses increase to 20%, SACKEXP = goodputs ~ 700%
at similar efficiencies compared to TCP SACK
• Plan: investigate performance of SACKEXP with
Congestion Control based on the ECN bit/ICMP Source Quench
• Performance of SACKEXP has to be compared to a TCP that can
resort to a Link Layer retransmission scheme.
• What about TCP fairness????
THE END
Questions are
welcome!
[Jaco88] Van Jacobson, “Congestion Avoidance and Control”, ACM SIGCOMM '88
[Jaco90] Van Jacobson, “Modified TCP Congestion Avoidance Algorithm”, email to [email protected], April 1990
[BraMalPet94] Lawrence S. Brakmo, Sean W. O'Malley, Larry L. Peterson, „TCP Vegas: New
Techniques for Congestion Detection and Avoidance“, Sigcomm 1994
[MatMahFlRo96] M. Mathis, J. Mahdavi, S. Floyd, A. Romanow, „TCP Selective Acknowledgement
Options“, RFC 2018, April 1996
[Hoe96]
Janey C. Hoe, “Improving the start-up behavior of a Congestion Control Scheme for TCP,
Sigcomm 1996
[MatMah96] M. Mathis and J. Mahdavi, "Forward acknowledgement: Refining TCP congestion
control“, ACM Computer Communication Review, Oct 1996.