Delay-Based TCP Congestion Control David Hayes

advertisement
Delay-Based TCP
Congestion Control
David Hayes
dahayes@swin.edu.au
Centre for Advanced Internet Architectures (CAIA)
Swinburne University of Technology
Outline
CAIA
Background
TCP congestion control
Delay based congestion signals
Some key delay-based algorithms
LCN paper Improved coexistence and loss tolerance for
delay based TCP congestion control
Hamilton Institute’s Delay-based algorithm (HD)
Shortcomings of HD,
improved by CAIA HD algorithm (CHD)
back-off decision frequency and scaling
tolerance to non-congestion related loss
improvements when coexisting with NewReno type flows
Other TCP work
Delay-gradient based congestion control
Stateless TCP
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
2
CAIA –
Centre for Advanced Internet Architectures
We are at Swinburne University of Technology, about 5 km
east of the Melbourne CBD
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
3
CAIA –
Centre for Advanced Internet Architectures
We are at Swinburne University of Technology, about 5 km
east of the Melbourne CBD
CAIA is the research arm of the Telecommunications
Engineering Academic Group (or Department)
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
3
CAIA –
Centre for Advanced Internet Architectures
We are at Swinburne University of Technology, about 5 km
east of the Melbourne CBD
CAIA is the research arm of the Telecommunications
Engineering Academic Group (or Department)
Research spans:
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
3
CAIA –
Centre for Advanced Internet Architectures
We are at Swinburne University of Technology, about 5 km
east of the Melbourne CBD
CAIA is the research arm of the Telecommunications
Engineering Academic Group (or Department)
Research spans:
Congestion control (TCP, etc)
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
3
CAIA –
Centre for Advanced Internet Architectures
We are at Swinburne University of Technology, about 5 km
east of the Melbourne CBD
CAIA is the research arm of the Telecommunications
Engineering Academic Group (or Department)
Research spans:
Congestion control (TCP, etc)
Traffic classification
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
3
CAIA –
Centre for Advanced Internet Architectures
We are at Swinburne University of Technology, about 5 km
east of the Melbourne CBD
CAIA is the research arm of the Telecommunications
Engineering Academic Group (or Department)
Research spans:
Congestion control (TCP, etc)
Traffic classification
Wireless networks
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
3
CAIA –
Centre for Advanced Internet Architectures
We are at Swinburne University of Technology, about 5 km
east of the Melbourne CBD
CAIA is the research arm of the Telecommunications
Engineering Academic Group (or Department)
Research spans:
Congestion control (TCP, etc)
Traffic classification
Wireless networks
Covert channels and lawful interception
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
3
CAIA –
Centre for Advanced Internet Architectures
We are at Swinburne University of Technology, about 5 km
east of the Melbourne CBD
CAIA is the research arm of the Telecommunications
Engineering Academic Group (or Department)
Research spans:
Congestion control (TCP, etc)
Traffic classification
Wireless networks
Covert channels and lawful interception
Network monitoring and visualisation
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
3
CAIA –
Centre for Advanced Internet Architectures
We are at Swinburne University of Technology, about 5 km
east of the Melbourne CBD
CAIA is the research arm of the Telecommunications
Engineering Academic Group (or Department)
Research spans:
Congestion control (TCP, etc)
Traffic classification
Wireless networks
Covert channels and lawful interception
Network monitoring and visualisation
BGP
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
3
CAIA –
Centre for Advanced Internet Architectures
We are at Swinburne University of Technology, about 5 km
east of the Melbourne CBD
CAIA is the research arm of the Telecommunications
Engineering Academic Group (or Department)
Research spans:
Congestion control (TCP, etc)
Traffic classification
Wireless networks
Covert channels and lawful interception
Network monitoring and visualisation
BGP
Address space exploration
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
3
CAIA –
Centre for Advanced Internet Architectures
We are at Swinburne University of Technology, about 5 km
east of the Melbourne CBD
CAIA is the research arm of the Telecommunications
Engineering Academic Group (or Department)
Research spans:
Congestion control (TCP, etc)
Traffic classification
Wireless networks
Covert channels and lawful interception
Network monitoring and visualisation
BGP
Address space exploration
Game traffic analysis
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
3
CISCO supported work
Exploring the efficacy of distributed statistical traffic
classification using modified open source packet filters
(2010)
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
4
CISCO supported work
Exploring the efficacy of distributed statistical traffic
classification using modified open source packet filters
(2010)
Implementing and testing delay-based and rate-based
transport protocols in FreeBSD (2008)
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
4
CISCO supported work
Exploring the efficacy of distributed statistical traffic
classification using modified open source packet filters
(2010)
Implementing and testing delay-based and rate-based
transport protocols in FreeBSD (2008)
Heuristics to reduce BGP Update Noise (2007)
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
4
CISCO supported work
Exploring the efficacy of distributed statistical traffic
classification using modified open source packet filters
(2010)
Implementing and testing delay-based and rate-based
transport protocols in FreeBSD (2008)
Heuristics to reduce BGP Update Noise (2007)
FreeBSD Implementation of an SCTP friendly NAT (2007)
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
4
CISCO supported work
Exploring the efficacy of distributed statistical traffic
classification using modified open source packet filters
(2010)
Implementing and testing delay-based and rate-based
transport protocols in FreeBSD (2008)
Heuristics to reduce BGP Update Noise (2007)
FreeBSD Implementation of an SCTP friendly NAT (2007)
Anomalous Traffic Detection and Collaborative Network
Configuration Using 3D Multiplayer Game Engines (2006)
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
4
CISCO supported work
Exploring the efficacy of distributed statistical traffic
classification using modified open source packet filters
(2010)
Implementing and testing delay-based and rate-based
transport protocols in FreeBSD (2008)
Heuristics to reduce BGP Update Noise (2007)
FreeBSD Implementation of an SCTP friendly NAT (2007)
Anomalous Traffic Detection and Collaborative Network
Configuration Using 3D Multiplayer Game Engines (2006)
Public Implementation and Interoperability Testing of Next
Generation TCP Stack Under FreeBSD (2005)
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
4
CISCO supported work
Exploring the efficacy of distributed statistical traffic
classification using modified open source packet filters
(2010)
Implementing and testing delay-based and rate-based
transport protocols in FreeBSD (2008)
Heuristics to reduce BGP Update Noise (2007)
FreeBSD Implementation of an SCTP friendly NAT (2007)
Anomalous Traffic Detection and Collaborative Network
Configuration Using 3D Multiplayer Game Engines (2006)
Public Implementation and Interoperability Testing of Next
Generation TCP Stack Under FreeBSD (2005)
Dynamic Self-Learning Traffic Classification Based on Flow
Characteristics (2004)
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
4
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
Delay based intuition:
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
Delay based intuition:
delay↑ ≡ queue↑
1
=⇒ indicates congestion
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
Delay based intuition:
delay↑ ≡ queue↑
=⇒ indicates congestion
Rate based intuition:
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
Delay based intuition:
delay↑ ≡ queue↑
=⇒ indicates congestion
Rate based intuition:
Send rate > receive rate
1
=⇒ indicates congestion
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
Delay based intuition:
delay↑ ≡ queue↑
=⇒ indicates congestion
Rate based intuition:
Send rate > receive rate
=⇒ indicates congestion
Basic questions:
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
Delay based intuition:
delay↑ ≡ queue↑
=⇒ indicates congestion
Rate based intuition:
Send rate > receive rate
=⇒ indicates congestion
Basic questions:
How is congestion determined?
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
Delay based intuition:
delay↑ ≡ queue↑
=⇒ indicates congestion
Rate based intuition:
Send rate > receive rate
=⇒ indicates congestion
Basic questions:
How is congestion determined?
and if congested, how should cwnd be adjusted
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
Delay based intuition:
delay↑ ≡ queue↑
=⇒ indicates congestion
Rate based intuition:
Send rate > receive rate
=⇒ indicates congestion
Basic questions:
How is congestion determined?
and if congested, how should cwnd be adjusted
Issues:
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
Delay based intuition:
delay↑ ≡ queue↑
=⇒ indicates congestion
Rate based intuition:
Send rate > receive rate
=⇒ indicates congestion
Basic questions:
How is congestion determined?
and if congested, how should cwnd be adjusted
Issues:
Noise of measurements
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
Delay based intuition:
delay↑ ≡ queue↑
=⇒ indicates congestion
Rate based intuition:
Send rate > receive rate
=⇒ indicates congestion
Basic questions:
How is congestion determined?
and if congested, how should cwnd be adjusted
Issues:
Noise of measurements
Correlation of measurements with congestion
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Introduction to delay and rate based TCP
Promise of low latency zero loss1 transmission
the congestion signal can be decoupled from packet loss
potential for efficient transmission on lossy paths.
Delay based intuition:
delay↑ ≡ queue↑
=⇒ indicates congestion
Rate based intuition:
Send rate > receive rate
=⇒ indicates congestion
Basic questions:
How is congestion determined?
and if congested, how should cwnd be adjusted
Issues:
Noise of measurements
Correlation of measurements with congestion
Compatibility with existing TCP algorithms
1
congestion related
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
5
Background – TCP’s Congestion Window (w)
In general loss-based TCP:
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
6
Background – TCP’s Congestion Window (w)
In general loss-based TCP:
increases w by the maximum segment size every RTT
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
6
Background – TCP’s Congestion Window (w)
In general loss-based TCP:
increases w by the maximum segment size every RTT
or 1/w for every ACK
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
6
Background – TCP’s Congestion Window (w)
In general loss-based TCP:
increases w by the maximum segment size every RTT
or 1/w for every ACK
and halves w when a packet has been lost.
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
6
Background – TCP’s Congestion Window (w)
In general loss-based TCP:
increases w by the maximum segment size every RTT
or 1/w for every ACK
and halves w when a packet has been lost.
(
wi
lost packet
wi+1 = 2
1
wi + wi otherwise
120
cwnd (packets)
100
80
60
40
20
NewReno
0
20
25
CISCO
30
35
time (s)
http://www.caia.swin.edu.au
40
45
dahayes@swin.edu.au
50
15 October, 2010
6
Background: Base timing measurements
S1
dsw
dS1
S2
rttmin
S3
rtt1
S4
rttmax
S5
drw
A1
A2
A3
A4
A5
daw ≈ drw
S7
S9 S8
CISCO
S6
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
7
Background: Base timing measurements
S1
dsw
dS1
S2
rttmin
S3
rtt1
S4
rttmax
S5
drw
A1
A2
A3
A4
A5
daw ≈ drw
S7
S9 S8
CISCO
S6
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
7
Background: Base timing measurements
S1
dsw
dS1
S2
rttmin
S3
rtt1
S4
rttmax
S5
drw
A1
A2
A3
A4
A5
daw ≈ drw
S7
S9 S8
CISCO
S6
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
7
Background: Base timing measurements
S1
dsw
dS1
S2
rttmin
S3
rtt1
S4
rttmax
S5
drw
A1
A2
A3
A4
A5
daw ≈ drw
S7
S9 S8
CISCO
S6
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
7
Background: Base timing measurements
S1
dsw
dS1
S2
rttmin
S3
rtt1
S4
rttmax
S5
drw
A1
A2
A3
A4
A5
daw ≈ drw
S7
S9 S8
CISCO
S6
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
7
Background: Base timing measurements
S1
dsw
dS1
S2
rttmin
S3
rtt1
S4
rttmax
S5
drw
A1
A2
A3
A4
A5
daw ≈ drw
S7
S9 S8
CISCO
S6
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
7
Background: Base timing measurements
S1
dsw
dS1
S2
rttmin
S3
rtt1
S4
rttmax
S5
drw
A1
A2
A3
A4
A5
daw ≈ drw
S7
S9 S8
CISCO
S6
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
7
Background: Base timing measurements
S
S
S
A
A
daw
S
S
A
0
daw
00
daw
S
A
A
S
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
8
Background: Base timing measurements
S
S
S
A
A
daw
S
S
A
0
daw
00
daw
S
A
A
S
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
8
Background: Base timing measurements
S
S
S
A
A
daw
S
S
A
0
daw
00
daw
S
A
A
S
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
8
Background: Base timing measurements
S
S
Note: Queueing at
FIFO network nodes
can increase or
decrease the
interpacket times
S
A
A
daw
S
S
A
0
daw
00
daw
S
A
A
S
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
8
Background: Base rate measurements
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
9
Background: Base rate measurements
S1
Tmax =
T1 =
Pw
S
S2
rttmin
S3
Pw
S4
S
rtt1
S5
A1
A2
Ra =
Pw−a
1
A3
A4
A5
Ai
daw
S7
S9 S8
CISCO
http://www.caia.swin.edu.au
S6
dahayes@swin.edu.au
15 October, 2010
9
Background: Base rate measurements
S1
Tmax =
T1 =
Pw
S
S2
rttmin
S3
Pw
S4
S
rtt1
S5
A1
A2
Ra =
Pw−a
1
A3
A4
A5
Ai
daw
S7
S9 S8
CISCO
http://www.caia.swin.edu.au
S6
dahayes@swin.edu.au
15 October, 2010
9
Background: Base rate measurements
S1
Tmax =
T1 =
Pw
S
S2
rttmin
S3
Pw
S4
S
rtt1
S5
A1
A2
Ra =
Pw−a
1
A3
A4
A5
Ai
daw
S7
S9 S8
CISCO
http://www.caia.swin.edu.au
S6
dahayes@swin.edu.au
15 October, 2010
9
Background: Base rate measurements
S1
Tmax =
T1 =
Pw
S
S2
rttmin
S3
Pw
S4
S
rtt1
S5
A1
A2
Ra =
Pw−a
1
A3
A4
A5
Ai
daw
S7
S9 S8
CISCO
http://www.caia.swin.edu.au
S6
dahayes@swin.edu.au
15 October, 2010
9
Background: Base rate measurements
S1
Tmax =
T1 =
Pw
S
S2
rttmin
S3
Pw
S4
S
rtt1
S5
A1
A2
Ra =
Pw−a
1
A3
A4
A5
Ai
daw
S7
S9 S8
CISCO
http://www.caia.swin.edu.au
S6
dahayes@swin.edu.au
15 October, 2010
9
Quick early work overview
[Clark et al., 1985]&[Clark et al., 1987] NETBLT RFCs 996&998
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
10
Quick early work overview
[Clark et al., 1985]&[Clark et al., 1987] NETBLT RFCs 996&998
[Jacobson, 1988]a – footnote on connectionless rate based AIMD.
a
V. Jacobson, “Congestion avoidance and control,” in SIGCOMM ’88: Symposium
proceedings on Communications architectures and protocols. New York, NY, USA:
ACM, 1988, pp. 314–329
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
10
Quick early work overview
[Clark et al., 1985]&[Clark et al., 1987] NETBLT RFCs 996&998
[Jacobson, 1988]a – footnote on connectionless rate based AIMD.
[Jain, 1989]b normalised delay gradient.
a
V. Jacobson, “Congestion avoidance and control,” in SIGCOMM ’88: Symposium
proceedings on Communications architectures and protocols. New York, NY, USA:
ACM, 1988, pp. 314–329
b
R. Jain, “A delay-based approach for congestion avoidance in interconnected
heterogeneous computer networks,” SIGCOMM Comput. Commun. Rev., vol. 19, no. 5,
pp. 56–71, 1989
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
10
Quick early work overview
[Clark et al., 1985]&[Clark et al., 1987] NETBLT RFCs 996&998
[Jacobson, 1988]a – footnote on connectionless rate based AIMD.
[Jain, 1989]b normalised delay gradient.
[Wang and Crowcroft, 1992]c DUAL algorithm.
a
V. Jacobson, “Congestion avoidance and control,” in SIGCOMM ’88: Symposium
proceedings on Communications architectures and protocols. New York, NY, USA:
ACM, 1988, pp. 314–329
b
R. Jain, “A delay-based approach for congestion avoidance in interconnected
heterogeneous computer networks,” SIGCOMM Comput. Commun. Rev., vol. 19, no. 5,
pp. 56–71, 1989
c
Z. Wang and J. Crowcroft, “Eliminating periodic packet losses in the 4.3-Tahoe
BSD TCP congestion control algorithm,” SIGCOMM Comput. Commun. Rev., vol. 22,
no. 2, pp. 9–16, Apr. 1992
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
10
Quick early work overview
[Clark et al., 1985]&[Clark et al., 1987] NETBLT RFCs 996&998
[Jacobson, 1988]a – footnote on connectionless rate based AIMD.
[Jain, 1989]b normalised delay gradient.
[Wang and Crowcroft, 1992]c DUAL algorithm.
[Brakmo and Peterson, 1995]d TCP Vegas.
a
V. Jacobson, “Congestion avoidance and control,” in SIGCOMM ’88: Symposium
proceedings on Communications architectures and protocols. New York, NY, USA:
ACM, 1988, pp. 314–329
b
R. Jain, “A delay-based approach for congestion avoidance in interconnected
heterogeneous computer networks,” SIGCOMM Comput. Commun. Rev., vol. 19, no. 5,
pp. 56–71, 1989
c
Z. Wang and J. Crowcroft, “Eliminating periodic packet losses in the 4.3-Tahoe
BSD TCP congestion control algorithm,” SIGCOMM Comput. Commun. Rev., vol. 22,
no. 2, pp. 9–16, Apr. 1992
d
L. S. Brakmo and L. L. Peterson, “TCP Vegas: end to end congestion avoidance on
a global internet,” IEEE J. Sel. Areas Commun., vol. 13, no. 8, pp. 1465–1480, Oct.
1995
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
10
A quick look at some
key/interesting algorithms in
the literature
Centre for Advanced Internet Architectures (CAIA)
Swinburne University of Technology
Algorithms: Pkt pair flow control [Keshav, 1994]
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
12
Algorithms: Pkt pair flow control [Keshav, 1994]
All data is sent as
back-to-back pairs
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
12
Algorithms: Pkt pair flow control [Keshav, 1994]
All data is sent as
back-to-back pairs
p1
p2
t
RTT
pair
disperion
pair
disperion
estimate
SOURCE
CISCO
http://www.caia.swin.edu.au
network node BOTTLENECK
dahayes@swin.edu.au
SINK
15 October, 2010
12
Algorithms: Pkt pair flow control [Keshav, 1994]
All data is sent as
back-to-back pairs
p1
p2
t
RTT
pair
disperion
pair
disperion
estimate
SOURCE
CISCO
http://www.caia.swin.edu.au
network node BOTTLENECK
dahayes@swin.edu.au
SINK
15 October, 2010
12
Algorithms: Pkt pair flow control [Keshav, 1994]
All data is sent as
back-to-back pairs
p1
p2
t
RTT
pair
disperion
pair
disperion
estimate
SOURCE
CISCO
http://www.caia.swin.edu.au
network node BOTTLENECK
dahayes@swin.edu.au
SINK
15 October, 2010
12
Algorithms: Pkt pair flow control [Keshav, 1994]
All data is sent as
back-to-back pairs
p1
p2
t
RTT
pair
disperion
pair
disperion
estimate
SOURCE
CISCO
http://www.caia.swin.edu.au
network node BOTTLENECK
dahayes@swin.edu.au
SINK
15 October, 2010
12
Algorithms: Pkt pair flow control [Keshav, 1994]
All data is sent as
back-to-back pairs
p1
p2
t
RTT
pair
disperion
pair
disperion
estimate
SOURCE
CISCO
http://www.caia.swin.edu.au
network node BOTTLENECK
dahayes@swin.edu.au
SINK
15 October, 2010
12
Algorithms: Pkt pair flow control [Keshav, 1994]
All data is sent as
back-to-back pairs
p1
p2
t
RTT
pair
disperion
pair
disperion
estimate
SOURCE
CISCO
http://www.caia.swin.edu.au
network node BOTTLENECK
dahayes@swin.edu.au
SINK
15 October, 2010
12
Algorithms: Pkt pair flow control [Keshav, 1994]
All data is sent as
back-to-back pairs
p1
p2
t
RTT
pair
disperion
pair
disperion
estimate
SOURCE
CISCO
http://www.caia.swin.edu.au
network node BOTTLENECK
dahayes@swin.edu.au
SINK
15 October, 2010
12
Algorithms: Pkt pair flow control [Keshav, 1994]
All data is sent as
back-to-back pairs
p1
p2
t
RTT
pair
disperion
pair
disperion
estimate
SOURCE
CISCO
http://www.caia.swin.edu.au
network node BOTTLENECK
dahayes@swin.edu.au
SINK
15 October, 2010
12
Algorithms: Pkt pair flow control [Keshav, 1994]
All data is sent as
back-to-back pairs
p1
p2
t
Available send rate is:
T =
size(p2 )
pair dispersion
RTT
pair
disperion
pair
disperion
estimate
SOURCE
CISCO
http://www.caia.swin.edu.au
network node BOTTLENECK
dahayes@swin.edu.au
SINK
15 October, 2010
12
Algorithms: Pkt pair flow control [Keshav, 1994]
All data is sent as
back-to-back pairs
p1
p2
t
Available send rate is:
T =
size(p2 )
pair dispersion
RTT
pair
disperion
Presumes routers use
round robin scheduling
pair
disperion
estimate
SOURCE
CISCO
http://www.caia.swin.edu.au
network node BOTTLENECK
dahayes@swin.edu.au
SINK
15 October, 2010
12
Algorithms: Vegas [Brakmo and Peterson, 1995]
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
13
Algorithms: Vegas [Brakmo and Peterson, 1995]
Iconic rate based TCP
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
13
Algorithms: Vegas [Brakmo and Peterson, 1995]
Iconic rate based TCP
Defines two rates:
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
13
Algorithms: Vegas [Brakmo and Peterson, 1995]
Iconic rate based TCP
P
Defines two rates:
P
actual =
S
rtt
S1
S
S2
rttmin
S3
S4
rtt1
S5
A1
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
13
Algorithms: Vegas [Brakmo and Peterson, 1995]
Iconic rate based TCP
P
Defines two rates:
P
actual =
S
rtt
S1
S
S2
S3
rttmin
S4
rtt1
S5
w
expected =
rttmin
CISCO
A1
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
13
Algorithms: Vegas [Brakmo and Peterson, 1995]
Iconic rate based TCP
P
Defines two rates:
P
actual =
S1
S
S2
S3
rttmin
S
rtt
S4
rtt1
S5
w
expected =
rttmin
and
A1
diff = expected − actual
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
13
Algorithms: Vegas [Brakmo and Peterson, 1995]
Iconic rate based TCP
P
Defines two rates:
P
actual =
S1
S
S2
S3
rttmin
S
rtt
S4
rtt1
S5
w
expected =
rttmin
and
A1
diff = expected − actual
window adjustment:


w − 1 diff > β
w ← w + 1 diff < α


w
otherwise
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
13
Algorithms: Vegas [Brakmo and Peterson, 1995]
Iconic rate based TCP
P
Defines two rates:
P
actual =
S1
S
S2
S3
rttmin
S
rtt
S4
rtt1
S5
w
expected =
rttmin
and
A1
diff = expected − actual
window adjustment:


w − 1 diff > β
w ← w + 1 diff < α


w
otherwise
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
13
Algorithms: Vegas [Brakmo and Peterson, 1995]
Iconic rate based TCP
P
Defines two rates:
P
actual =
S1
S
S2
S3
rttmin
S
rtt
S4
rtt1
S5
w
expected =
rttmin
and
A1
diff = expected − actual
window adjustment:


w − 1 diff > β
w ← w + 1 diff < α


w
otherwise
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
13
Algorithms: Vegas [Brakmo and Peterson, 1995]
Iconic rate based TCP
P
Defines two rates:
P
actual =
S1
S
S2
S3
rttmin
S
rtt
S4
rtt1
S5
w
expected =
rttmin
and
A1
diff = expected − actual
window adjustment:


w − 1 diff > β
w ← w + 1 diff < α


w
otherwise
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
13
Algorithms: FAST [Wei et al., 2006]
Enhanced Vegas type algorithm
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
14
Algorithms: FAST [Wei et al., 2006]
Enhanced Vegas type algorithm
MIMD — AIMD to slow for high BDP networks
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
14
Algorithms: FAST [Wei et al., 2006]
Enhanced Vegas type algorithm
MIMD — AIMD to slow for high BDP networks
Uses delay as a rich (non binary) congestion indicator
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
14
Algorithms: FAST [Wei et al., 2006]
Enhanced Vegas type algorithm
MIMD — AIMD to slow for high BDP networks
Uses delay as a rich (non binary) congestion indicator
Cwnd is updated at regular time intervals (∆t):
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
14
Algorithms: FAST [Wei et al., 2006]
Enhanced Vegas type algorithm
MIMD — AIMD to slow for high BDP networks
Uses delay as a rich (non binary) congestion indicator
Cwnd is updated at regular time intervals (∆t):
rttmin,i
wt + α + (1 − γ)wt
wt+∆t = min 2wt , γ
rtti
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
14
Algorithms: FAST [Wei et al., 2006]
Enhanced Vegas type algorithm
MIMD — AIMD to slow for high BDP networks
Uses delay as a rich (non binary) congestion indicator
Cwnd is updated at regular time intervals (∆t):
rttmin,i
wt + α + (1 − γ)wt
wt+∆t = min 2wt , γ
rtti
Smoothed window increase
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
14
Algorithms: FAST [Wei et al., 2006]
Enhanced Vegas type algorithm
MIMD — AIMD to slow for high BDP networks
Uses delay as a rich (non binary) congestion indicator
Cwnd is updated at regular time intervals (∆t):
rttmin,i
wt + α + (1 − γ)wt
wt+∆t = min 2wt , γ
rtti
When congested, decreases in proportion to queueing
delay
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
14
Algorithms: FAST [Wei et al., 2006]
Enhanced Vegas type algorithm
MIMD — AIMD to slow for high BDP networks
Uses delay as a rich (non binary) congestion indicator
Cwnd is updated at regular time intervals (∆t):
rttmin,i
wt + α + (1 − γ)wt
wt+∆t = min 2wt , γ
rtti
increase is limited to 2w per ∆t
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
14
Algorithms: FAST [Wei et al., 2006]
Enhanced Vegas type algorithm
MIMD — AIMD to slow for high BDP networks
Uses delay as a rich (non binary) congestion indicator
Cwnd is updated at regular time intervals (∆t):
rttmin,i
wt + α + (1 − γ)wt
wt+∆t = min 2wt , γ
rtti
For MIMD, α(wt , qi )
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
14
Algorithms: FAST [Wei et al., 2006]
Enhanced Vegas type algorithm
MIMD — AIMD to slow for high BDP networks
Uses delay as a rich (non binary) congestion indicator
Cwnd is updated at regular time intervals (∆t):
rttmin,i
wt + α + (1 − γ)wt
wt+∆t = min 2wt , γ
rtti
For MIMD, α(wt , qi )
increase is proportional to the size of cwnd and the network
queueing delay.
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
14
Algorithms: Compound TCP [Tan et al., 2006]
In high speed high BDP networks aims to increase:
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
15
Algorithms: Compound TCP [Tan et al., 2006]
In high speed high BDP networks aims to increase:
efficiency
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
15
Algorithms: Compound TCP [Tan et al., 2006]
In high speed high BDP networks aims to increase:
efficiency
RTT fairness and TCP fairness
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
15
Algorithms: Compound TCP [Tan et al., 2006]
In high speed high BDP networks aims to increase:
efficiency
RTT fairness and TCP fairness
In MSW Vista and 7
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
15
Algorithms: Compound TCP [Tan et al., 2006]
In high speed high BDP networks aims to increase:
efficiency
RTT fairness and TCP fairness
In MSW Vista and 7
Uses Vegas’ rates: diff = (expected − actual)rttmin
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
15
Algorithms: Compound TCP [Tan et al., 2006]
In high speed high BDP networks aims to increase:
efficiency
RTT fairness and TCP fairness
In MSW Vista and 7
Uses Vegas’ rates: diff = (expected − actual)rttmin
Provides NewReno+ performance throughput
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
15
Algorithms: Compound TCP [Tan et al., 2006]
In high speed high BDP networks aims to increase:
efficiency
RTT fairness and TCP fairness
In MSW Vista and 7
Uses Vegas’ rates: diff = (expected − actual)rttmin
Provides NewReno+ performance throughput
The send window, winj , is calculated as:
winj = min(wj + dwndj , awndj )
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
15
Algorithms: Compound TCP [Tan et al., 2006]
In high speed high BDP networks aims to increase:
efficiency
RTT fairness and TCP fairness
In MSW Vista and 7
Uses Vegas’ rates: diff = (expected − actual)rttmin
Provides NewReno+ performance throughput
The send window, winj , is calculated as:
winj = min(wj + dwndj , awndj )
where wj is NewReno’s cwnd
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
15
Algorithms: Compound TCP [Tan et al., 2006]
In high speed high BDP networks aims to increase:
efficiency
RTT fairness and TCP fairness
In MSW Vista and 7
Uses Vegas’ rates: diff = (expected − actual)rttmin
Provides NewReno+ performance throughput
The send window, winj , is calculated as:
winj = min(wj + dwndj , awndj )
where wj is NewReno’s cwnd
and dwndj is the delay based window.
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
15
Algorithms: Compound TCP [Tan et al., 2006]
In high speed high BDP networks aims to increase:
efficiency
RTT fairness and TCP fairness
In MSW Vista and 7
Uses Vegas’ rates: diff = (expected − actual)rttmin
Provides NewReno+ performance throughput
The send window, winj , is calculated as:
winj = min(wj + dwndj , awndj )
where wj is NewReno’s cwnd
and dwndj is the delay based window.
and awndj is the receivers advertised window.
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
15
Algorithms: DUAL [Wang and Crowcroft, 1992]
Designed to supplement loss based congestion control
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
16
Algorithms: DUAL [Wang and Crowcroft, 1992]
Designed to supplement loss based congestion control
Delay based measurements provide “slow tuning” of cwnd
every 2nd RTT
(
max )
βw rtt > (rttmin +rtt
2
w←
w
otherwise
where β =
CISCO
7
8
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
16
Algorithms: DUAL [Wang and Crowcroft, 1992]
Designed to supplement loss based congestion control
Delay based measurements provide “slow tuning” of cwnd
every 2nd RTT
(
max )
βw rtt > (rttmin +rtt
2
w←
w
otherwise
where β =
7
8
Attempts to keep network buffers half full
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
16
Algorithms: DUAL [Wang and Crowcroft, 1992]
Designed to supplement loss based congestion control
Delay based measurements provide “slow tuning” of cwnd
every 2nd RTT
(
max )
βw rtt > (rttmin +rtt
2
w←
w
otherwise
where β =
7
8
Attempts to keep network buffers half full
Smaller multiplicative decrease
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
16
Algorithms: DUAL [Wang and Crowcroft, 1992]
Designed to supplement loss based congestion control
Delay based measurements provide “slow tuning” of cwnd
every 2nd RTT
(
max )
βw rtt > (rttmin +rtt
2
w←
w
otherwise
where β =
7
8
Attempts to keep network buffers half full
Smaller multiplicative decrease
Relies on accurate estimates of rttmin and rttmax
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
16
Algorithms: Others of Interest
[King et al., 2005] — TCP-Africa
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
17
Algorithms: Others of Interest
[King et al., 2005] — TCP-Africa
Two modes: Fast delay based, and slow NewReno based.
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
17
Algorithms: Others of Interest
[King et al., 2005] — TCP-Africa
Two modes: Fast delay based, and slow NewReno based.
Compound TCP is based on some of Africa’s ideas
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
17
Algorithms: Others of Interest
[King et al., 2005] — TCP-Africa
Two modes: Fast delay based, and slow NewReno based.
Compound TCP is based on some of Africa’s ideas
[Baiocchi et al., 2007] — YeAH-TCP
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
17
Algorithms: Others of Interest
[King et al., 2005] — TCP-Africa
Two modes: Fast delay based, and slow NewReno based.
Compound TCP is based on some of Africa’s ideas
[Baiocchi et al., 2007] — YeAH-TCP
Yet Another Highspeed TCP
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
17
Algorithms: Others of Interest
[King et al., 2005] — TCP-Africa
Two modes: Fast delay based, and slow NewReno based.
Compound TCP is based on some of Africa’s ideas
[Baiocchi et al., 2007] — YeAH-TCP
Yet Another Highspeed TCP
Two modes like Africa
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
17
Algorithms: Others of Interest
[King et al., 2005] — TCP-Africa
Two modes: Fast delay based, and slow NewReno based.
Compound TCP is based on some of Africa’s ideas
[Baiocchi et al., 2007] — YeAH-TCP
Yet Another Highspeed TCP
Two modes like Africa
Provides performance improvements on lossy paths.
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
17
Algorithms: Others of Interest
[King et al., 2005] — TCP-Africa
Two modes: Fast delay based, and slow NewReno based.
Compound TCP is based on some of Africa’s ideas
[Baiocchi et al., 2007] — YeAH-TCP
Yet Another Highspeed TCP
Two modes like Africa
Provides performance improvements on lossy paths.
A number of schemes propose traffic shaping TCP’s send
rate
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
17
Algorithms: Others of Interest
[King et al., 2005] — TCP-Africa
Two modes: Fast delay based, and slow NewReno based.
Compound TCP is based on some of Africa’s ideas
[Baiocchi et al., 2007] — YeAH-TCP
Yet Another Highspeed TCP
Two modes like Africa
Provides performance improvements on lossy paths.
A number of schemes propose traffic shaping TCP’s send
rate
[Karandikar et al., 2000] – ABR like
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
17
Algorithms: Others of Interest
[King et al., 2005] — TCP-Africa
Two modes: Fast delay based, and slow NewReno based.
Compound TCP is based on some of Africa’s ideas
[Baiocchi et al., 2007] — YeAH-TCP
Yet Another Highspeed TCP
Two modes like Africa
Provides performance improvements on lossy paths.
A number of schemes propose traffic shaping TCP’s send
rate
[Karandikar et al., 2000] – ABR like
[Wu et al., 2002] – leaky bucket
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
17
Algorithms: Others of Interest
[King et al., 2005] — TCP-Africa
Two modes: Fast delay based, and slow NewReno based.
Compound TCP is based on some of Africa’s ideas
[Baiocchi et al., 2007] — YeAH-TCP
Yet Another Highspeed TCP
Two modes like Africa
Provides performance improvements on lossy paths.
A number of schemes propose traffic shaping TCP’s send
rate
[Karandikar et al., 2000] – ABR like
[Wu et al., 2002] – leaky bucket
[Abendroth et al., 2002] – improved leaky bucket for network
burstiness.
CISCO
http://www.caia.swin.edu.au
dahayes@swin.edu.au
15 October, 2010
17
Improved coexistence and
loss tolerance for
delay based
TCP congestion control
Best Paper Award LCN 2010
David Hayes and Grenville Armitage
{dahayes,garmitage}@swin.edu.au
Centre for Advanced Internet Architectures (CAIA)
Swinburne University of Technology
Introduction
Delay-based congestion control can potentially provide:
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
19
Introduction
Delay-based congestion control can potentially provide:
low latency transmission no congestion induced packet loss.
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
19
Introduction
Delay-based congestion control can potentially provide:
low latency transmission no congestion induced packet loss.
efficient TCP over lossy paths (wireless links).
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
19
Introduction
Delay-based congestion control can potentially provide:
low latency transmission no congestion induced packet loss.
efficient TCP over lossy paths (wireless links).
Issues:
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
19
Introduction
Delay-based congestion control can potentially provide:
low latency transmission no congestion induced packet loss.
efficient TCP over lossy paths (wireless links).
Issues:
Measuring delay to infer congestion
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
19
Introduction
Delay-based congestion control can potentially provide:
low latency transmission no congestion induced packet loss.
efficient TCP over lossy paths (wireless links).
Issues:
Measuring delay to infer congestion
Coexistence with current loss-based TCP
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
19
Introduction
Delay-based congestion control can potentially provide:
low latency transmission no congestion induced packet loss.
efficient TCP over lossy paths (wireless links).
Issues:
Measuring delay to infer congestion
Coexistence with current loss-based TCP
We propose a delay based algorithm which:
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
19
Introduction
Delay-based congestion control can potentially provide:
low latency transmission no congestion induced packet loss.
efficient TCP over lossy paths (wireless links).
Issues:
Measuring delay to infer congestion
Coexistence with current loss-based TCP
We propose a delay based algorithm which:
improves TCP efficiency over lossy paths
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
19
Introduction
Delay-based congestion control can potentially provide:
low latency transmission no congestion induced packet loss.
efficient TCP over lossy paths (wireless links).
Issues:
Measuring delay to infer congestion
Coexistence with current loss-based TCP
We propose a delay based algorithm which:
improves TCP efficiency over lossy paths
improves coexistence with loss-based TCP
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
19
Introduction
Delay-based congestion control can potentially provide:
low latency transmission no congestion induced packet loss.
efficient TCP over lossy paths (wireless links).
Issues:
Measuring delay to infer congestion
Coexistence with current loss-based TCP
We propose a delay based algorithm which:
improves TCP efficiency over lossy paths
improves coexistence with loss-based TCP
implement algorithms in the FreeBSD kernel
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
19
Hamilton Delay (HD) based window updates
[Budzisz et al., 2009]
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
20
Hamilton Delay (HD) based window updates
[Budzisz et al., 2009]
Probabilistic delay-based backoff (coexistence)
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
20
Hamilton Delay (HD) based window updates
[Budzisz et al., 2009]
Probabilistic delay-based backoff (coexistence)
Per−packet
backoff
probability
g (q )
backoff
probability
pmax
B
A
qmin qth
CISCO
qmax
http://www.caia.swin.edu.au
Queuing
delay
{dahayes,garmitage}@swin.edu.au
15 October, 2010
20
Hamilton Delay (HD) based window updates
[Budzisz et al., 2009]
Probabilistic delay-based backoff (coexistence)
Per−packet
backoff
probability
g (q )
backoff
probability
pmax
(
B
A
qmin qth
CISCO
wi+1 =
qmax
http://www.caia.swin.edu.au
wi
2
wi +
1
wi
X < g(qi )
otherwise
Queuing
delay
{dahayes,garmitage}@swin.edu.au
15 October, 2010
20
Hamilton Delay (HD) based window updates
[Budzisz et al., 2009]
Probabilistic delay-based backoff (coexistence)
Per−packet
backoff
probability
g (q )
backoff
probability
pmax
(
B
A
qmin qth
CISCO
wi+1 =
qmax
http://www.caia.swin.edu.au
wi
2
wi +
1
wi
X < g(qi )
otherwise
Queuing
delay
{dahayes,garmitage}@swin.edu.au
15 October, 2010
20
Hamilton Delay (HD) based window updates
[Budzisz et al., 2009]
Probabilistic delay-based backoff (coexistence)
Per−packet
backoff
probability
g (q )
backoff
probability
pmax
(
B
A
qmin qth
wi+1 =
qmax
wi
2
wi +
1
wi
X < g(qi )
otherwise
Queuing
delay
Region A stable when queueing delay is low
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
20
Hamilton Delay (HD) based window updates
[Budzisz et al., 2009]
Probabilistic delay-based backoff (coexistence)
Per−packet
backoff
probability
g (q )
backoff
probability
pmax
(
B
A
qmin qth
wi+1 =
qmax
wi
2
wi +
1
wi
X < g(qi )
otherwise
Queuing
delay
Region A stable when queueing delay is low
Region B only stable when queueing delay is high
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
20
RTT – loss-based & delay-based congestion control
140
RTT (s)
120
100
80
60
NewReno
40
20
25
30
35
Time (s)
40
45
50
140
Hamilton
RTT (s)
120
100
80
60
40
20
25
CISCO
30
35
Time (s)
http://www.caia.swin.edu.au
40
45
{dahayes,garmitage}@swin.edu.au
50
15 October, 2010
21
Delay-based back-off decision frequency
Per−packet
backoff
probability
g (q )
backoff
probability
pmax
A
B
qmax
qmin qth
CISCO
http://www.caia.swin.edu.au
Queuing
delay
{dahayes,garmitage}@swin.edu.au
15 October, 2010
22
Delay-based back-off decision frequency
Per−packet
backoff
probability
g (q )
backoff
probability
pmax
A
B
qmax
qmin qth
Queuing
delay
HD decision per packet:
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
22
Delay-based back-off decision frequency
Per−packet
backoff
probability
g (q )
backoff
probability
pmax
A
B
qmax
qmin qth
Queuing
delay
HD decision per packet:
Doesn’t scale well:
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
22
Delay-based back-off decision frequency
Per−packet
backoff
probability
g (q )
backoff
probability
pmax
B
A
qmax
qmin qth
Queuing
delay
HD decision per packet:
Doesn’t scale well:
P[backoff] increases with w
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
22
Delay-based back-off decision frequency
Per−packet
backoff
probability
g (q )
backoff
probability
pmax
B
A
qmax
qmin qth
Queuing
delay
HD decision per packet:
Doesn’t scale well:
P[backoff] increases with w
CPU
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
22
Delay-based back-off decision frequency
Per−RTT
backoff
probability
g (hr )
backoff
probability
pmax
B
A
qmax
qmin qth
HD decision per packet:
Queuing
delay
CHD decision once per RTT
Doesn’t scale well:
P[backoff] increases with w
CPU
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
22
Delay-based back-off decision frequency
Per−RTT
backoff
probability
g (hr )
backoff
probability
pmax
B
A
qmax
qmin qth
HD decision per packet:
Queuing
delay
CHD decision once per RTT
Doesn’t scale well:
Uses hr = maxr (qi )
P[backoff] increases with w
CPU
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
22
Delay-based back-off decision frequency
Per−RTT
backoff
probability
g (hr )
backoff
probability
pmax
B
A
qmax
qmin qth
HD decision per packet:
Queuing
delay
CHD decision once per RTT
Doesn’t scale well:
Uses hr = maxr (qi )
P[backoff] increases with w
Scales with w
CPU
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
22
Delay-based back-off decision frequency
Per−RTT
backoff
probability
g (hr )
backoff
probability
pmax
B
A
qmax
qmin qth
HD decision per packet:
Queuing
delay
CHD decision once per RTT
Doesn’t scale well:
Uses hr = maxr (qi )
P[backoff] increases with w
CPU
Scales with w
less CPU
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
22
Tolerance to non-congestion related packet loss
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
23
Tolerance to non-congestion related packet loss
HD (and NewReno) do not tolerate packet loss
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
23
Tolerance to non-congestion related packet loss
HD (and NewReno) do not tolerate packet loss
w = w/2 when a packet is lost
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
23
Tolerance to non-congestion related packet loss
HD (and NewReno) do not tolerate packet loss
w = w/2 when a packet is lost
CHD tolerates low level packet loss well by:
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
23
Tolerance to non-congestion related packet loss
HD (and NewReno) do not tolerate packet loss
w = w/2 when a packet is lost
CHD tolerates low level packet loss well by:
Ignoring packet loss when queueing delays are small
(region A)
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
23
Tolerance to non-congestion related packet loss
HD (and NewReno) do not tolerate packet loss
w = w/2 when a packet is lost
CHD tolerates low level packet loss well by:
Ignoring packet loss when queueing delays are small
(region A)
Per−RTT
backoff
probability
g (hr )
backoff
probability
pmax
A
B
qmax
qmin qth
CISCO
http://www.caia.swin.edu.au
Queuing
delay
{dahayes,garmitage}@swin.edu.au
15 October, 2010
23
Tolerance to non-congestion related losses
6
10
x 10
NewReno
HD
CHD
1/sqrt(p)
Goodput (bps)
8
6
4
2
0
0
0.01
0.02
0.03
0.04
Probability of non−congestion related loss
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
0.05
15 October, 2010
24
Improving coexistence with loss-based TCP
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
25
Improving coexistence with loss-based TCP
To improve CHD’s coexistence ability
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
25
Improving coexistence with loss-based TCP
To improve CHD’s coexistence ability
React only to packet loss in region B
Per−RTT
backoff
probability
g (hr )
backoff
probability
pmax
A
B
qmax
qmin qth
CISCO
http://www.caia.swin.edu.au
Queuing
delay
{dahayes,garmitage}@swin.edu.au
15 October, 2010
25
Improving coexistence with loss-based TCP
To improve CHD’s coexistence ability
React only to packet loss in region B
We use a shadow window (s) (shadows NewReno)
Per−RTT
backoff
probability
g (hr )
backoff
probability
pmax
A
B
qmax
qmin qth
CISCO
http://www.caia.swin.edu.au
Queuing
delay
{dahayes,garmitage}@swin.edu.au
15 October, 2010
25
Improving coexistence with loss-based TCP
To improve CHD’s coexistence ability
React only to packet loss in region B
We use a shadow window (s) (shadows NewReno)
On packet loss, wi+1 =
Per−RTT
backoff
probability
max(wi ,si )
2
g (hr )
backoff
probability
pmax
A
B
qmax
qmin qth
CISCO
http://www.caia.swin.edu.au
Queuing
delay
{dahayes,garmitage}@swin.edu.au
15 October, 2010
25
Improving coexistence with loss-based TCP
To improve CHD’s coexistence ability
React only to packet loss in region B
We use a shadow window (s) (shadows NewReno)
On packet loss, wi+1 =
Per−RTT
backoff
probability
max(wi ,si )
2
g (hr )
backoff
probability
pmax
A
B
qmax
qmin qth
Queuing
delay
Best explained with an example
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
25
Shadow window example
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
26
packets
Shadow window example
w
s=0
number of round trip times
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
26
packets
Shadow window example
w
delay based congestion
s=0
number of round trip times
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
26
Shadow window example
packets
s sync
s
w
delay based congestion
number of round trip times
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
26
Shadow window example
packets
s sync
s
w
delay based congestion
number of round trip times
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
26
Shadow window example
lost packet
packets
s sync
s
w
w recovery
delay based congestion
number of round trip times
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
26
Shadow window example
lost packet
s sync
packets
Region B
max(wi , si )
2
s
w
w recovery
delay based congestion
number of round trip times
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
26
Shadow window example
lost packet
packets
s sync
s
w
w recovery
delay based congestion
without w recovery
number of round trip times
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
26
Shadow window example
lost packet
packets
s sync
s
lost
transmission
opportunity
w
w recovery
delay based congestion
without w recovery
number of round trip times
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
26
Shadow window example
lost packet
packets
s sync
s
lost
transmission
opportunity
w
gained
transmission
opportunity
w recovery
delay based congestion
without w recovery
number of round trip times
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
26
Testbed for coexistence tests
Delay CC Sources
(FreeBSD)
Delay CC Sink
(FreeBSD)
20ms
Dummynet Router
(FreeBSD)
20ms
NewReno Sources
(FreeBSD)
NewReno Sink
(FreeBSD)
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
27
Testbed for coexistence tests
Delay CC Sources
(FreeBSD)
Delay CC Sink
(FreeBSD)
20ms
Dummynet Router
(FreeBSD)
20ms
NewReno Sources
(FreeBSD)
NewReno Sink
(FreeBSD)
We will look at HD and CHD coexisting with NewReno
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
27
Testbed for coexistence tests
Delay CC Sources
(FreeBSD)
Delay CC Sink
(FreeBSD)
20ms
Dummynet Router
(FreeBSD)
20ms
NewReno Sources
(FreeBSD)
NewReno Sink
(FreeBSD)
We will look at HD and CHD coexisting with NewReno
For 0 % and 1 % non-congestion related losses
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
27
HD coexisting with NewReno (1s Av, 0 % loss)
HD
6
x 10
NewReno
NewReno
HD
10
Goodput (bps)
8
6
4
2
0
0
20
40
60
80
100
120
140
Time (s)
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
28
HD coexisting with NewReno (1s Av, 0 % loss)
HD
6
x 10
NewReno
NewReno
HD
10
Goodput (bps)
8
6
4
2
0
0
20
40
60
80
100
120
140
Time (s)
HD does not compete well with NewReno
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
28
CHD coexisting with NewReno (1s Av, 0 % loss)
CHD
6
x 10
NewReno
NewReno
CHD
10
Goodput (bps)
8
6
4
2
0
0
20
40
60
80
100
120
140
Time (s)
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
28
CHD coexisting with NewReno (1s Av, 0 % loss)
CHD
6
x 10
NewReno
NewReno
CHD
10
Goodput (bps)
8
6
4
2
0
0
20
40
60
80
100
120
140
Time (s)
CHD reclaims some of the lost capacity from NewReno
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
28
Coexiting with NewReno on a lossy path
1 % loss
5 s averages
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
29
HD coexisting with NewReno (5s Av, 1 % loss)
6
x 10
10
HD
NewReno
NewReno
HD
Goodput (bps)
8
6
4
2
0
0
20
40
60
80
100
120
140
Time (s)
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
29
HD coexisting with NewReno (5s Av, 1 % loss)
6
x 10
10
HD
NewReno
NewReno
HD
Goodput (bps)
8
6
4
2
0
0
20
40
60
80
100
120
140
Time (s)
HD and NewReno cannot efficiently utilise the available
bandwidth
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
29
CHD coexisting with NewReno (5s Av, 1 % loss)
6
x 10
10
CHD
NewReno
NewReno
CHD
Goodput (bps)
8
6
4
2
0
0
20
40
60
80
100
120
140
Time (s)
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
29
CHD coexisting with NewReno (5s Av, 1 % loss)
6
x 10
10
CHD
NewReno
NewReno
CHD
Goodput (bps)
8
6
4
2
0
0
20
40
60
80
100
120
140
Time (s)
CHD is able to effectively use the available bandwidth
(including what NewReno is unable to use)
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
29
LCN Conclusions
CHD significantly enhances HD:
improves scalability with per RTT decisions
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
30
LCN Conclusions
CHD significantly enhances HD:
improves scalability with per RTT decisions
improves tolerance to non-congestion related packet losses
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
30
LCN Conclusions
CHD significantly enhances HD:
improves scalability with per RTT decisions
improves tolerance to non-congestion related packet losses
improves coexistence with loss based TCP (NewReno)
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
30
LCN Conclusions
CHD significantly enhances HD:
improves scalability with per RTT decisions
improves tolerance to non-congestion related packet losses
improves coexistence with loss based TCP (NewReno)
Shadow window
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
30
LCN Conclusions
CHD significantly enhances HD:
improves scalability with per RTT decisions
improves tolerance to non-congestion related packet losses
improves coexistence with loss based TCP (NewReno)
Shadow window
Lightly multiplexed environments
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
30
LCN Conclusions
CHD significantly enhances HD:
improves scalability with per RTT decisions
improves tolerance to non-congestion related packet losses
improves coexistence with loss based TCP (NewReno)
Shadow window
Lightly multiplexed environments
CHD and HD have been implemented in the FreeBSD kernel
(caia.swin.edu.au/urp/newtcp/tools.html)
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
30
LCN Conclusions
CHD significantly enhances HD:
improves scalability with per RTT decisions
improves tolerance to non-congestion related packet losses
improves coexistence with loss based TCP (NewReno)
Shadow window
Lightly multiplexed environments
CHD and HD have been implemented in the FreeBSD kernel
(caia.swin.edu.au/urp/newtcp/tools.html)
This work was made possible in part
by a grant from the Cisco University Research Program Fund
at Community Foundation Silicon Valley
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
30
A brief look at some other
TCP work at CAIA
David Hayes and Grenville Armitage
{dahayes,garmitage}@swin.edu.au
Centre for Advanced Internet Architectures (CAIA)
Swinburne University of Technology
Delay-gradient based TCP congestion control
We investigated a delay-gradient congestion signal
because:
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
32
Delay-gradient based TCP congestion control
We investigated a delay-gradient congestion signal
because:
it does not require an accurate estimate of base RTT
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
32
Delay-gradient based TCP congestion control
We investigated a delay-gradient congestion signal
because:
it does not require an accurate estimate of base RTT
delay thresholds are hard to set —
need to know path’s delay characteristics
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
32
Delay-gradient based TCP congestion control
We investigated a delay-gradient congestion signal
because:
it does not require an accurate estimate of base RTT
delay thresholds are hard to set —
need to know path’s delay characteristics
We have implemented it in FreeBSD, to be released soon.
(waiting on a paper submission)
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
32
Comparing NewReno and our Delay-Gradient
TCP RTT dynamics as 3 sources start
140
RTT (s)
120
100
80
flow 1
flow 2
flow 3
60
40
0
NewReno
CISCO
20
40
http://www.caia.swin.edu.au
60
Time (s)
80
{dahayes,garmitage}@swin.edu.au
100
15 October, 2010
33
Comparing NewReno and our Delay-Gradient
TCP RTT dynamics as 3 sources start
140
RTT (s)
120
100
80
flow 1
flow 2
flow 3
60
40
0
NewReno
20
40
60
Time (s)
80
100
140
flow 1
flow 2
flow 3
RTT (s)
120
Delay-Gradient
100
80
60
40
0
CISCO
20
40
http://www.caia.swin.edu.au
60
Time (s)
80
{dahayes,garmitage}@swin.edu.au
100
15 October, 2010
33
Stateless TCP
Proposed by Geoff Huston to mitigate a DNS server issue
http://www.potaroo.net/ispcol/2009-11/
stateless.pdf
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
34
Stateless TCP
Proposed by Geoff Huston to mitigate a DNS server issue
http://www.potaroo.net/ispcol/2009-11/
stateless.pdf
Funded by APNIC and Nominet
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
34
Stateless TCP
Proposed by Geoff Huston to mitigate a DNS server issue
http://www.potaroo.net/ispcol/2009-11/
stateless.pdf
Funded by APNIC and Nominet
DNSSEC may cause the answers to DNS queries to exceed
a single UDP packet
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
34
Stateless TCP
Proposed by Geoff Huston to mitigate a DNS server issue
http://www.potaroo.net/ispcol/2009-11/
stateless.pdf
Funded by APNIC and Nominet
DNSSEC may cause the answers to DNS queries to exceed
a single UDP packet
Clients using TCP may overload servers
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
34
Stateless TCP
Proposed by Geoff Huston to mitigate a DNS server issue
http://www.potaroo.net/ispcol/2009-11/
stateless.pdf
Funded by APNIC and Nominet
DNSSEC may cause the answers to DNS queries to exceed
a single UDP packet
Clients using TCP may overload servers
Geoff’s idea → stateless TCP
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
34
Stateless TCP
Proposed by Geoff Huston to mitigate a DNS server issue
http://www.potaroo.net/ispcol/2009-11/
stateless.pdf
Funded by APNIC and Nominet
DNSSEC may cause the answers to DNS queries to exceed
a single UDP packet
Clients using TCP may overload servers
Geoff’s idea → stateless TCP
Implemented in FreeBSD, will be released on the CAIA web
site in a few weeks.
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
34
Basic stateless TCP idea
DNS server
Listen
Data
send data up
through UDP
Data
UDP
statelessTCP
if matches hash
TCP
IP
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
35
CPU time versus DNS query arrival rate
CPU usage in 10s interval
2.5
2
udp
tcp
stateless
1.5
1
0.5
0
0
100
CISCO
200
300
400
Average requests/second
http://www.caia.swin.edu.au
500
{dahayes,garmitage}@swin.edu.au
15 October, 2010
36
Thoughts and conclusions
Delay-based TCP coexistence with loss-based TCP
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
37
Thoughts and conclusions
Delay-based TCP coexistence with loss-based TCP
current schemes coexist by behaving like NewReno
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
37
Thoughts and conclusions
Delay-based TCP coexistence with loss-based TCP
current schemes coexist by behaving like NewReno
Low latency with no congestion related loss
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
37
Thoughts and conclusions
Delay-based TCP coexistence with loss-based TCP
current schemes coexist by behaving like NewReno
Low latency with no congestion related loss
only when there are no loss-based flows sharing path
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
37
Thoughts and conclusions
Delay-based TCP coexistence with loss-based TCP
current schemes coexist by behaving like NewReno
Low latency with no congestion related loss
only when there are no loss-based flows sharing path
If switches and routers could differentiate between loss and
delay based TCP, benefits would be realised sooner.
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
37
Thoughts and conclusions
Delay-based TCP coexistence with loss-based TCP
current schemes coexist by behaving like NewReno
Low latency with no congestion related loss
only when there are no loss-based flows sharing path
If switches and routers could differentiate between loss and
delay based TCP, benefits would be realised sooner.
Delay-gradient as a congestion indication
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
37
Thoughts and conclusions
Delay-based TCP coexistence with loss-based TCP
current schemes coexist by behaving like NewReno
Low latency with no congestion related loss
only when there are no loss-based flows sharing path
If switches and routers could differentiate between loss and
delay based TCP, benefits would be realised sooner.
Delay-gradient as a congestion indication
works well
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
37
Thoughts and conclusions
Delay-based TCP coexistence with loss-based TCP
current schemes coexist by behaving like NewReno
Low latency with no congestion related loss
only when there are no loss-based flows sharing path
If switches and routers could differentiate between loss and
delay based TCP, benefits would be realised sooner.
Delay-gradient as a congestion indication
works well
a composite delay-based congestion indication may be better
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
37
Thank you!
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
38
Thank you!
Questions?
CISCO
http://www.caia.swin.edu.au
{dahayes,garmitage}@swin.edu.au
15 October, 2010
38
Download