a hybrid systems modeling framework for data communication networks ph.d dissertation proposal...
TRANSCRIPT
A Hybrid Systems Modeling Framework for Data Communication
Networks
Ph.D Dissertation Proposal
Junsoo Lee
9/5/2003
2
Studying Networks…
• Study of networks and network protocols have used:– Analytical models.
– Simulation tools.
• Limitations:– Analytical models
• Significant accuracy loss
• Only applicable to limited application
– Simulation tools• Long simulation time
• Large memory overhead
3
Motivation
• Simulation speed up– Faster than packet level simulation– More accurate than fluid simulation
• Validate designs through simulation– Scalability, performance
• Analyze and design protocols– Throughput, fairness, security
• Tune network parameters– Queue size, bandwidth
4
Expected contribution
• Provide a scalable framework for the design, analysis, and evaluation of large-scale computer networks and their protocol
• Contribute to the networking research and industry communities by allowing efficient and accurate simulation of large-scale network
• Provide tools to generate hybrid model without programming by generating automatic simulation code from a given network topology.
• Provide test environment of the network protocols on networks with large delay bandwidth product
5
Talk Outline
• Related work• Simplified Hybrid Model of TCP • Generalized Hybrid Model Framework• Validation • Simulation Complexity• Contributions, Proposed Work &
Schedule• Conclusion
6
Related Work: Packet model
• Track individual data packets
• Computationally intensive
• Complexity depends on the number of events
• Does not scale to large bandwidth and complex topology
• NS-2 (NS00)
• Pdns (Riley99)
• QualNet
• Opnet (Desbrandes93)
• SSFNET
7
Related Work: Fluid Model
• Track time/ensemble-average packet rates
• Computationally efficient
• Complexity depends on the rate changes
• Only suitable to model many flows
• Does not explicitly model individual event
• ATM (Kesidis96)
• Time driven (Yan99)• Stochastic Differential Equation
(Misra99,20)
• Time-Stepped Hybrid Simulation (Guo00)
• Fluid-Simulation using SSF (Nicol98)
• More efficient and larger scale (Liu03)
8
Related Work: Hybrid model
• Discrete Event + analytical technique
• Packet (foreground) + fluid model (back-ground)
• Packet (edge) + fluid mode (backbone)
• Abstract technique
• Computer systems (Schwetman78)
• Fluid model extension to QualNet (Tak01)
• HDCF-NS (Melamed01)• HDCF-NS + PDNS (Riley02)
• Hybrid mode buffer (cameron03)
• Abstract technique (Huang99)
9
Our Approach: Hybrid model• Track packet rates for each flow averaged over small time
scales
• explicitly models some discrete events (drops, queues becoming empty, etc.)
• time accuracy of a few milliseconds (round-trip time)
• Key idea presented at SIGMETRIC 2003
10
Talk Outline
• Related work• Simplified Hybrid Model of TCP • Generalized Hybrid Model Framework• Validation • Simulation Complexity• Contributions, Proposed Work &
Schedule• Conclusion
11
Simple Hybrid Model Example
State 1
State 2
Cx
),( yxfy
Dx
),( yxgy
?by
?ax
0:x
transition enabling condition
state reset
[Shaft00]
16
Dumbbell topology
When i ri exceeds B the queue fills and data is lost (drops)
rate = B bps
drop (discrete event)
r1 bps
r2 bps
r3 bpsq( t ) = queue size
queue (temporary storage for data)
f1
f2
f3
f1
f2
f3
17
Window-based rate adjustmentwf (window size) = number of packets that can remain
unacknowledged for by the destination
1st packet sent
e.g., wf = 3
t
2nd packet sent3rd packet sent 1st packet received & ack. sent
2nd packet received & ack. sent3rd packet received & ack. sent1st ack.
received )4th packet can be sent
t
source f destination f
wf effectively determines the sending rate rf :
t0
t1
t2
t3
0
1
2
propagation delay
time in queueuntil transmission
round-trip time
18
TCP Sack Congestion Control
B
qTRTT p
RTTw f
1
1. While there are no drops, increase wf by 1 on each RTT2. When a drop occurs, divide wi by 2
(congestion controller constantly probe the network for more bandwidth)
Queuing model TCP controllers
drop
RTT
rf
RTT
wr f
f
f
f Brqq max, 2
w
w f
Consider only CA for now for the simplicity
20
Talk Outline
• Related work• Simplified hybrid model of TCP • Generalized Hybrid Model Framework• Validation • Simulation Complexity• Contributions, Proposed Work &
Schedule• Conclusion
21
General Topology f1
f2
f1
f2
F := { f1, f2, … } : set of end2end flows
N := { n1, n2, … } : set of nodes
L := { 1, 2, … } : set of links
n1
n2
n3
n4
n5
n6
1
2
3
4
5
B = bandwidth of link T = prop. delay of link
22
Queue Dynamics
total queue size queue size due to flow f
the packets of each flow are assumed uniformly distributed in the queue
in-queue rates out-queue rates
…drop rates
Queue dynamics:
23
Queue Dynamics
queue not empty/full
queue full
queue empty
same in and out-queue rates
out-queue rates proportional to fraction of packets in the queue
no drops
drops proportional to fraction in-queue rates
in-queue rates out-queue rates
…drop rates
24
Hybrid Queue Model-queue-not-full
-queue-full
transition enabling condition
exporteddiscrete event
discrete modes
25
TCP: AIMD
congestion-avoidance
set of links transversed by flow f
propagation delays
1. While there are no drops, increase wf by 1 on each RTT (additive increase)
2. When a drop occurs, divide wf by 2 (multiplicative decrease)
(congestion controller constantly probe the network for more bandwidth)
importeddiscrete event
26
TCP: Slow Start
3. Until a drop occurs (or a threshold ssthf is reached), double wf on each RTT4. When a drop occurs, divide wf and the threshold ssthf by 2
cong.-avoid.slow-start
27
TCP: Timeout, Fast Recovery
dropf
f
dropf
ntw
tw
Floorntwn
2
)(1
2
)(1
log:)),((0
0
20
}42,2max{ dropdropf nnw
2/)( 0 twn fdrop
12/)( 0 twn fdrop dropdropf nFloorntwn 20 log1:)),((
6. During fast recovery, data is sent at a rate consistent with a window size of wf /2
7. Duration of fast recovery (RTT) for Tcp-sack
5. Timeout occurs when
29
Congestion Controlrouting
queue dynamics
sendingrates
drops
out-queuerates
in-queue rates
TCP
RTTs
30
Talk Outline
• Related work• Simplified Hybrid Model of TCP• Generalized Hybrid Model Framework• Validation • Simulation Complexity• Contributions, Proposed Work &
Schedule• Conclusion
31
Comparison of Hybrid Model Simulation Environments
Simulator SHIFT DYMOLA
Language SHIFT MODELICA
Institution Berkeley Dynasim
Solver Fixed Fixed/Variable
Analysis Tool No Yes
Object Oriented Yes Yes
Speed Slow Fast
Platform Linux/win32 Redhat/win32
Public Yes No
Dymola has variety of solvers and efficient methods for determining when discrete events occur
32
Validation MethodologyCompared simulation results from• ns-2 packet-level simulator• hybrid models implemented in Modelica and Shift
Plots in the following slides refer to two test topologies
• 10ms propagation delay• drop-tail queuing• 5-500Mbps bottleneck throughput
• 45,90,135,180ms propagation delays• drop-tail queuing• 5-500Mbps bottleneck throughput• 0-10% UDP on/off background traffic
Y-topologydumbbell
34
4 flow : Dumbbell• four competing TCP flow• 5Mbps bottleneck throughput• no background traffic
hybrid model ns-2
the hybrid model accurately captures flow synchronization
35
4 flows with BG:Y-shape
hybrid model
• four competing TCP flow• 5Mbps bottleneck throughput• 10% UDP background traffic
(exponentially distributedon-off times)
ns-2
36
Average throughput and RTT
Thru. 1 Thru. 2 Thru. 3 Thru. 4 RTT1 RTT2 RTT3 RTT4
ns-2 1.873 1.184 .836 .673 .0969 .141 .184 .227
hybrid model 1.824 1.091 .823 .669 .0879 .132 .180 .223
relative error 2.6% 7.9% 1.5% .7% 9.3% 5.9% 3.6% 2.1%
• four competing TCP flow• 5Mbps bottleneck throughput• 20 trials with 10 minutes
simulation time• 10% UDP background traffic
(exponentially distributedon-off times)
the hybrid model accurately captures TCP unfairness in 10% relative error for different propagation delays
• 45,90,135,180ms propagation delays• drop-tail queuing• 5Mbps bottleneck throughput• 10% UDP on/off background traffic
37
Empirical Distribution
hybrid model ns-2
the hybrid model captures the whole distribution of congestion windows and queue size
L-1 difference
cwnd1
cwnd2
cwnd3
cwnd4
bottleneck queue
dumbbell .71% .67% .71% .66% 1.1%
Y-shape .34% .44% .25% .33% .54%
38
Talk Outline
• Related work• Simplified Hybrid Model of TCP • Generalized Hybrid Model Framework• Validation • Simulation Complexity• Contributions, Proposed Work &
Schedule• Conclusion
39
Execution Time-1
0.1
1
10
100
1000
10000
1 10 100 1000
bottleneck bandwidth [Mbps]
execution tim
e for
10m
in
of sim
ula
tion tim
e
[sec]
ns-2
hybrid model
1 flow
3 flows
• ns-2 complexity approximately scales with
• hybrid simulator complexity approximately scales with
number of flows
per-flow throughput
(# packets)
5Mbps
50Mbps
500Mbps
hybrid models are particularly suitable for large, high-bandwidth simulations (satellite, fiber optics, backbone)
40
Execution Time-2
• dumbbell topology with 100ms propagation delay
Exe
cutio
n tim
e fo
r 10
min
Of s
imul
atio
n tim
e [s
ec]
The hybrid model is hundred times faster than ns-2 when bandwidth 1Gbps and there is 30 flows
sec
0.01
0.1
1
10
100
1000
10000
10 100 1000
Bottleneck Bandwidth [Mbps]
1 flow-hybrid
10 flows-hybrid
20 flows-hybrid
30 flows-hybrid
1 flow-ns
10 flows-ns
20 flows-ns
30 flows-ns
41
1
10
100
1000
10000
5 50 500
Bottleneck Bandwidth [Mbps]
sec hybrid
ns
Execution Time-3
• Execution time for 200 seconds of simulation time
• 4 TCP and 10 UDP flows with Y-Shape topology
Exe
cutio
n tim
e fo
r 20
0 se
cO
f sim
ulat
ion
time
[sec
]
The hybrid model is 50 times faster than ns-2 with Y-shape topology
42
Talk Outline
• Related work• Simplified Hybrid Model of TCP • Generalized Hybrid Model Framework• Validation • Simulation Complexity• Contributions, Proposed Work &
Schedule• Conclusion
43
Contribution (so far)
• Apply hybrid systems to model communication network for the first time
• Develop hybrid framework for TCP congestion control and validate it by comparing to packet-level simulations
• Implement network model using SHIFT and Modelica hybrid model language
• Simulation speed up to few hundred times compare to packet model
• Simple automatic hybrid model generator from network topology
• Develop On-off TCP flows characterizes on period using some file size and off period using some distribution
44
Proposed work
1. Tools to generate simulation code from a given topology
2. Improve scalability of simulator by extending hybrid technique
(e.g. prediction of drop, aggregation of flows, skip multiple drop transition, removing fast recovery)
3. Extension to other forms of congestion control, queuing policies, and drop models
(e.g. priority queuing, TCP-vegas, wireless, HTTP)
4. Illustrate and verify protocol for high delay and bandwidth product
(e.g. FAST TCP)
45
Expected contribution
• Provide a scalable framework for the design, analysis, and evaluation of large-scale computer networks and their protocol
• Contribute to the networking research and industry communities by allowing efficient and accurate simulation of large-scale network
• Provide tools to generate hybrid model without programming by generating automatic simulation code from a given network topology.
• Provide test environment of the network protocols on networks with large delay bandwidth product
46
Schedule
• Fall 2003– Develop tools to generate hybrid simulation code from a given
topology
• Fall 2003 – Winter 2003– Improve scalability by extending hybrid technique
• Spring 2004– Extend to other forms of congestion control, queuing policies, and
drop models– Study on network protocol for large delay bandwidth product
• Summer 2004– Dissertation writing– Ph. D Defense
47
Conclusion
• Hybrid Systems provide a promising approach to model network traffic– Retain the low-dimensionality of continuous
approximations to traffic flow– Represent event based control mechanisms with
high accuracy, even at small time-scales– Complexity scales inversely with throughput
and RTT– Amenable to formal analysis
49
Simple Hybrid Model Example
State 1
State 2
Cx
),( yxfy
Dx
),( yxgy
by
?ax
0:x
transition enabling condition
state reset
54
Hybrid system model of TCP
queue-not-full
queue-full
(drop occurs)
(drop detected)
transition enabling condition
state reset
55
Drop probability vs. fraction of arrival rate
Blue flow gets most of the drops, in spite of using a smaller fraction of bandwidth when synchronization occurs, with sufficient randomness drop probability
56
Dumbbell topology
Several flows follow the same path and compete for bandwidth in a single bottleneck link
Prototypical network to study congestion control
single queuerouting is trivial
q( t ) ´ queue size
r1 bps
r2 bps
r3 bps
rate · B bps
queue
f1
f2
f3
f1
f2
f3
57
TCP Sack congestion control (Slow-Start)
)()()(1
log
tmwtweRTTtW ff
dtRTT
m
ff
fRTTt
t f
f
ff RTT
wr
1. While there are no drops, increase wi exponentially, being multipliedby 2 on each RTT
ff
f wRTT
mw
log
2. For an appropriately define constant m. If was constantwe get
fRTT
3. Since wf packets are sent each round-trip time, sending rate is
58
Slow Start : Dumbbell
hybrid model ns-2
• single TCP flow• 5Mbps bottleneck throughput• no background traffic
59
Execution Time-2
• Execution time for 10 minutes of simulation time
• dumbbell topology with 20ms propagation delay
0
500
1000
1500
2000
2500
3000
10 100 1000
Bottleneck Bandwidth [Mbps ]
sec
1 flow- hybrid10 flows - hybrid20 flows - hybrid30 flows - hybrid1 flow- ns10 flows - ns20 flows - ns30 flows - nsE
xecu
tion
time
for
10 m
inO
f sim
ulat
ion
time
[sec
]
60
Execution Time-4
0
200
400
600
800
1000
1200
1400
1600
5 50 500
Bottleneck Bandwidth [Mbps]
sec hybrid
ns
• Execution time for 200 seconds of simulation time
• 4 TCP and 10 UDP flows with Y-Shape topology
Exe
cutio
n tim
e fo
r 20
0 se
cO
f sim
ulat
ion
time
[sec
]
The hybrid model is faster than ns-2 when topology is more general such as Y-shape
61
Hybrid Queue Model (RED)
Random Early Dropactive queuing
stochastic counter-queue-not-full
-queue-full
discrete modes
62
Window-based rate adjustment
i
ii RTT
twtr
)()(
wi (window size) = number of packets that can remain unacknowledged for by the destination sending rate
totalround-trip
time propagationdelay
per-packettransmission time
time in queueuntil transmission
queuegets full
longerRTT
ratedecreases
queuegets empty
negative feedback
)(1
)( tqB
TtRTT p
63
Related Work: Others
• Steady state (Sally96, Padhye99, Yang00, Bansal00)
• Dynamic (Low02, Paganini03)
• Stochastic (Ott96, Padhye99-tr, Bohacek03)
• Flowsim (Ahn96)
• Flow level (Hong03)
64
Related Work-1 (Packet Models)• NS-2 (NS00)
– Most widely used simulator
– TCP, routing, multicast protocols over wired and wireless
• Pdns (Riley99)
– Parallel/Distributed version of NS-2
• QualNet
– Evolved from GloMosim (Zeng98) and PARSEC (Bagrodia98)
– Efficient and scalable simulation of wireless network
• SSFNET
– Collection of Java based components for modeling and simulation of Internet protocols
• Opnet (Desbrandes93)
– Originally developed at MIT and first commercial network simulator at 1987
65
Execution Time-3
0
200
400
600
800
1000
1200
1400
1600
1800
2000
10 100 1000
Bottleneck Bandwidth [Mbps]
sec
1 flow- hybrid10 flows- hybrid20 flows- hybrid30 flows- hybrid1 flow- ns10 flows- ns20 flows- ns30 flows- ns
• Execution time for 10 minutes of simulation time
• dumbbell topology with 100ms propagation delay
Exe
cutio
n tim
e fo
r 10
min
Of s
imul
atio
n tim
e [s
ec]
The hybrid model is faster than ns-2 when bandwidth 1Gbps and there is 30 flows
66
On-Off CBR Model
This is example of on-off CBR model and but on off period can follow any distribution
67
Related Work-2 (Analytical models)
• TCP model (Sally Floyd97, Mathis97)
• TCP friendly equation (Padhye98)-TCP’s steady state Throughput as a function of loss rate and RTT
• General and Binomial AIMD (Yang00, Bansal00)– Adjust sending rate by changing additive and multiplicative
constant• Equation Based Congestion Control (Padhye00)
– TCP Friendly Rate Control (TFRC) protocol– Based on Padhye’s equation
• Dynamics of TCP/RED and scalable control (Low02)– TCP/RED becomes unstable when delay increases
pRTTT
23.1
68
Related Work-3 (Fluid Models)
• ATM model (Kesidis96)- Simulation speed up
• Time driven model (Yan99)• Stochastic Differential Equation (Misra99, Misra00)
– Sources receive Poisson loss rate• Time Stepped hybrid simulation (Guo00)• Comparison with packet model (Liu01)
– Ripple effect• More efficient and larger scale (Liu03)
– Solving previous model numerically
69
Related Work-4 (Hybrid Models)• Hybrid model for computer systems (Schwetman78)
– Discrete event + analytical technique
• Adding fluid model to QualNet (Tak01)
– Misra’s fluid model
– Design an interface between a packet & fluid simulator
• Hybrid Discrete-Continuous Flow Network Simulator (Riley02)
– Flows arrive as a messages with workload, priority, and itinerary
• Integrate packet and fluid model (Riley02)
– Fluid modeling for background traffic: HDCF-NS (Melamed01)
– Packet modeling for foreground traffic: PDNS (Riley99)
• Hybrid Packet/Fluid model (cameron03)
– Fluid, packet, and hybrid mode buffer
70
Related Work: Analytical Model
• Track time/ensemble-average packet rates
• Computationally efficient
• Complexity depends on the rate changes
• Only suitable to model many flows
• Does not explicitly model individual event
• ATM (Kesidis96)
• Time driven (Yan99)• Stochastic Differential Equation
(Misra99,20)
• Time-Stepped Hybrid Simulation (Guo00)
• Fluid-Simulation using SSF (Nicol98)
• More efficient and larger scale (Liu03)
71
Related work-5 (others)
• Abstraction technique (Huang98)
– Centralized computation, End-to-End, Packet Delivery, Algorithmic Routing, FSA Modeling
• Packet train (Ahn96)
– Coarsening the network traffic
72
Analytical model Simulations
Steady state
Dynamic Fluid Sthochastic Hybrid Fluid Hybrid Packet
Sally96, Padhye99,Yang00, Bansal00
Misra99, Low02, paganini
03
Misra00, Liu03
Bohacek03 Nicol98, Guo00,
Kesidis96, Ahn96
Gu00, Schwetma
n78, Tak01,
Melamed01, Riley02, cameron03,huang99
NS-2, PDNS,Qua
lNet, SSFNET, OPNET