82
1 Colored GSPN Models Colored GSPN Models for the QoS Design for the QoS Design of Internet Subnets of Internet Subnets Marco Ajmone Marsan Marco Ajmone Marsan IEIIT-CNR and Politecnico di Torino - Italy Eindhoven – June 27, 2003 ICATPN 2003

Colored GSPN Models for the QoS Design of Internet Subnets

  • Upload
    arty

  • View
    40

  • Download
    0

Embed Size (px)

DESCRIPTION

Colored GSPN Models for the QoS Design of Internet Subnets. Marco Ajmone Marsan IEIIT-CNR and Politecnico di Torino - Italy. Eindhoven – June 27, 2003 ICATPN 2003. Venice 1988. M y previous invited talk at ICATPN. Goal : convince researchers to use GSPN models. Today. - PowerPoint PPT Presentation

Citation preview

Page 1: Colored GSPN Models  for the QoS Design  of Internet Subnets

1

Colored GSPN Models Colored GSPN Models for the QoS Design for the QoS Design of Internet Subnetsof Internet Subnets

Marco Ajmone MarsanMarco Ajmone MarsanIEIIT-CNR and Politecnico di Torino - Italy

Eindhoven – June 27, 2003

ICATPN 2003

Page 2: Colored GSPN Models  for the QoS Design  of Internet Subnets

3

Venice 1988

My previous invited talk at ICATPN

Goal: convince researchers to use GSPN models

Page 3: Colored GSPN Models  for the QoS Design  of Internet Subnets

4

Today

Original goal: publish a paper that I thought nobody would accept …

…but the paper was accepted!

Page 4: Colored GSPN Models  for the QoS Design  of Internet Subnets

5

Today

New goal: explain why (IMO) GSPN models (and discrete-state models in general) are becoming inadequate for Internet modeling

Page 5: Colored GSPN Models  for the QoS Design  of Internet Subnets

6

Colored GSPN Models Colored GSPN Models for the QoS Design for the QoS Design

of Internet Subnetsof Internet Subnets??Marco Ajmone MarsanMarco Ajmone Marsan

IEIIT-CNR and Politecnico di Torino - Italy

Eindhoven – June 27, 2003

Page 6: Colored GSPN Models  for the QoS Design  of Internet Subnets

7

Outline

The Internet today

Dimensioning IP networks

GSPN and Queuing network models

Fluid approaches

Conclusions

Page 7: Colored GSPN Models  for the QoS Design  of Internet Subnets

8

Outline

The Internet today

Dimensioning IP networks

GSPN and Queuing network models

Fluid approaches

Conclusions

Page 8: Colored GSPN Models  for the QoS Design  of Internet Subnets

9Source: Internet Software Consortium (http://www.isc.org/)

Page 9: Colored GSPN Models  for the QoS Design  of Internet Subnets

10

Source: Internet Traffic Report (http://www.internettrafficreport.com/)

Page 10: Colored GSPN Models  for the QoS Design  of Internet Subnets

11

Source: Internet Traffic Report (http://www.internettrafficreport.com/)

Page 11: Colored GSPN Models  for the QoS Design  of Internet Subnets

12

Source: Internet Traffic Report (http://www.internettrafficreport.com/)

Page 12: Colored GSPN Models  for the QoS Design  of Internet Subnets

13

Source: Internet Traffic Report (http://www.internettrafficreport.com/)

Page 13: Colored GSPN Models  for the QoS Design  of Internet Subnets

14Source: Sprint ATL (http://ipmon.sprint.com/packstat)April 7th 2003, 2.5 Gbps link

Page 14: Colored GSPN Models  for the QoS Design  of Internet Subnets

15Source: Sprint ATL (http://ipmon.sprint.com/packstat)April 7th 2003, 2.5 Gbps link

Page 15: Colored GSPN Models  for the QoS Design  of Internet Subnets

16Source: Sprint ATL (http://ipmon.sprint.com/packstat)April 7th 2003, 2.5 Gbps link

Page 16: Colored GSPN Models  for the QoS Design  of Internet Subnets

17Source: Sprint ATL (http://ipmon.sprint.com/packstat)April 7th 2003, 2.5 Gbps link

Page 17: Colored GSPN Models  for the QoS Design  of Internet Subnets

18Source: Sprint ATL (http://ipmon.sprint.com/packstat)April 7th 2003, 2.5 Gbps link

Page 18: Colored GSPN Models  for the QoS Design  of Internet Subnets

19Source: Sprint ATL (http://ipmon.sprint.com/packstat)April 7th 2003, 2.5 Gbps link

Page 19: Colored GSPN Models  for the QoS Design  of Internet Subnets

20Source: Sprint ATL (http://ipmon.sprint.com/packstat)April 7th 2003, 2.5 Gbps link

Page 20: Colored GSPN Models  for the QoS Design  of Internet Subnets

21Source: Sprint ATL (http://ipmon.sprint.com/packstat)April 7th 2003, 2.5 Gbps link

Page 21: Colored GSPN Models  for the QoS Design  of Internet Subnets

22

And still growing ...

Subject: [news] Internet still growing 70 to 150 per cent per yearDate: Mon, 23 Jun 2003 09:55:45 -0400 (EDT)From: [email protected]

...Andrew Odlyzko, director of the Digital Technology Center at the University of Minnesota, ... says Internet traffic is steadily growing about 70 percent to 150 percent per year. On a conference call yesterday to discuss the results, he said traffic growth slowed moderately over the last couple of years, but it had mostly remained constant for the past five years....

Page 22: Colored GSPN Models  for the QoS Design  of Internet Subnets

23

Outline

The Internet today

Dimensioning IP networks

GSPN and Queuing network models

Fluid approaches

Conclusions

Page 23: Colored GSPN Models  for the QoS Design  of Internet Subnets

24

Over 90 % of all Internet traffic is due to TCP connections

TCP drives both the network behavior and the performance perceived by end-users

Analytical models of TCP are a must for IP network design and planning

Consideration

Page 24: Colored GSPN Models  for the QoS Design  of Internet Subnets

25

A TCP Primer in 10 Slides• TCP is a reliable packet transfer protocol that

uses a variable window algorithm for:– Error control– Flow control– Congestion control

• Two main algorithms (and a number of gadgets):– Slow start– Congestion avoidance

Page 25: Colored GSPN Models  for the QoS Design  of Internet Subnets

26

Slow Start Algorithm

• Idea:– The new segment (packet) transmission rate

adapts to the ACK reception rate– The TCP transmitter “tests” the link capacity

• At connection setup, cwnd = 1 segment (actually, cwnd=MSS)

• At every received ACK, cwnd = cwnd + 1

• The resulting growth is exponential

Page 26: Colored GSPN Models  for the QoS Design  of Internet Subnets

27

Slow Start AlgorithmHost A

1 segment

RT

T

Host B

Time

2 segments

4 segments

Page 27: Colored GSPN Models  for the QoS Design  of Internet Subnets

28

Slow Start: Sample Trace

Page 28: Colored GSPN Models  for the QoS Design  of Internet Subnets

29

Congestion Avoidance Algorithm

• Idea:– Slower growth of cwnd

• At every ACK reception– cwnd = cwnd + 1/ cwnd – cwnd = cwnd + MSS*MSS/ cwnd (in bytes)

• The resulting growth is linear – cwnd grows by 1 MSS per RTT

Page 29: Colored GSPN Models  for the QoS Design  of Internet Subnets

30

Congestion AvoidanceSample Trace

Page 30: Colored GSPN Models  for the QoS Design  of Internet Subnets

31

When a Segment is Lost …

• …the transmitter rate has exceeded the available bandwidth

• Idea:– Reset the window size (cwnd=1)– Quickly recover the transmission rate

• The TCP transmitter detects the loss when the timeout expires, or 3 dupacks are received

Page 31: Colored GSPN Models  for the QoS Design  of Internet Subnets

32

Graphically …

5

10

15

20

cwnd

Time [RTT]

ssthresh

slow start

congestionavoidance

RTO

Page 32: Colored GSPN Models  for the QoS Design  of Internet Subnets

33

TCP Fairness

• The congestion control algorithm in TCP is AIMD (additive increase, multiplicative decrease)

• Fairness: N TCP connections sharing one bottleneck link of capacity C, obtain each C/N

Page 33: Colored GSPN Models  for the QoS Design  of Internet Subnets

34

R

R

Fair bandwidth sharing

Throughput connection 1Th

roughput

connec t

i on 2

loss: window reduced by factor 2

congestion avoidance: AI

Fairness with 2 TCP connections

• AI: linear increase

• MD: proportional decrease

Page 34: Colored GSPN Models  for the QoS Design  of Internet Subnets

35

AQM: RED

P(d)

Avgminth

maxth

1

Pmax

RED

Page 35: Colored GSPN Models  for the QoS Design  of Internet Subnets

36

Consideration

Accurate TCP models must consider:

closed loop behavior

short-lived flows

multi-bottleneck topologies

AQM schemes (or droptail)

QoS approaches, two-way traffic, ...

Page 36: Colored GSPN Models  for the QoS Design  of Internet Subnets

37

Developing accurate analytical models of the behavior of TCP is difficult.

A number of approaches have been proposed, some based on sophisticated modeling tools.

Consideration

Page 37: Colored GSPN Models  for the QoS Design  of Internet Subnets

38

Outline

The Internet today

Dimensioning IP networks

GSPN and Queuing network models

Fluid approaches

Conclusions

Page 38: Colored GSPN Models  for the QoS Design  of Internet Subnets

39

T. Lakshman and U. Madhow, "The performance of TCP/IP for networks with high bandwidth-delay products and random loss," IEEE/ACM Transactions on Networking, vol. 5, no. 3, 1997.

M.Ajmone Marsan, E.de Souza e Silva, R.Lo Cigno, M.Meo, “An Approximate Markovian Model for TCP over ATM”, UKPEW '97

J. Padhye, V. Firoiu, D. Towsley, J. Kurose, "A Stochastic Model of TCP Reno Congestion Avoidance and Control“, UMASS CMPSCI Technical Report, Feb 1999.

Literature

Page 39: Colored GSPN Models  for the QoS Design  of Internet Subnets

40

C.Casetti, M.Meo, “A New Approach to Model the Stationary Behavior of TCP Connections”, Infocom 2000

M.Ajmone Marsan, C.Casetti, R.Gaeta, M.Meo, “An Approximate GSPN Model for the Accurate Performance Analysis ofCorrelated TCP Connections”, SPECTS 2000

M.Garetto, R.Lo Cigno, M.Meo, E.Alessio, M.Ajmone Marsan, “Modeling Short-Lived TCP Connections with Open MulticlassQueueing Networks”, PfHSN 2002

A.Goel, M.Mitzenmacher, "Exact Sampling of TCP Window States", Infocom 2002

Literature

Page 40: Colored GSPN Models  for the QoS Design  of Internet Subnets

41

R.Gaeta, M.Sereno, D.Manini, "Stochastic Petri Nets models for the performance analysis of TCP connections supporting finite data transfer", QOS-IP 2003

R.Gaeta, M.Gribaudo, D.Manini, M.Sereno, "On the Use of Petri Nets for the Computation of Completion Time Distributon for Short TCP Transfers", ICATPN 2003

Literature

Page 41: Colored GSPN Models  for the QoS Design  of Internet Subnets

42

1 2

3

4...

N

URLs/sec

URLs/sec

greedy flows

4N F

23 F

finite flows (mice)

finite flows

greedy flows (elephants)

IP core

Problem statement

Page 42: Colored GSPN Models  for the QoS Design  of Internet Subnets

43

Input variables: only primitive network parameters:

IP network: channel data rates, node distances, buffer sizes, AQM algorithms (or droptail), ...

TCP: number of elephants, mice establishment rates and file length distribution, segment size, max window size, ...

Output variables: IP network: link utilizations, queuing delays, packet loss probabilities, ...

TCP: average elephant window size and throughput, average mice completion times, ...

Problem statement

Page 43: Colored GSPN Models  for the QoS Design  of Internet Subnets

44

IP networksub-model

TCPsub-model

1load

1

load N

packet loss probabilities, queuing delays

TCPsub-model

N

decomposition of the whole system into subsystems: sub-models are built for groups of homogeneous TCP connections (same TCP version, similar RTT and routing, ...) and for the IP network.

iterative solution with FPA (Fixed Point Algorithm).

Our modeling approach

Page 44: Colored GSPN Models  for the QoS Design  of Internet Subnets

45

Our modeling approach

Page 45: Colored GSPN Models  for the QoS Design  of Internet Subnets

46

GSPNs or . / G / queues describe states of the TCP protocol

tokens or customers stand for TCP connections

transition probabilities and service or firing times depend on TCP rules and network feedback (packet losses, round trip times, ...)

in the case of mice, colors or classes are introduced to represent the number of segments still to be transferred

TCP sub-model

Page 46: Colored GSPN Models  for the QoS Design  of Internet Subnets

47

TCP sub-model (Elephants)

Page 47: Colored GSPN Models  for the QoS Design  of Internet Subnets

48

TCP sub-model (Mice)

Page 48: Colored GSPN Models  for the QoS Design  of Internet Subnets

49

The IP network sub-model is an open queuing network, where each queue represents an output interface of an IP router, with its buffer of finite capacity.

Different queuing models were tested:

M / M / 1 / B: very simple, but only suitable when dealing with elephants and heavy load links

M [D] / M / 1 / B: to better model the traffic burstiness of mice under variable link utilization

M [D] / M [D] / 1 / B: a very accurate model, capable of coping with complex multi-bottleneck topologies

IP network sub-model

Page 49: Colored GSPN Models  for the QoS Design  of Internet Subnets

50

Bottleneck 1

Bottleneck 2

Numerical results: topology

Page 50: Colored GSPN Models  for the QoS Design  of Internet Subnets

51

Numerical results: settings

0.1

0.5

10 20 100

0.4

length (segmen

ts)

probability

Packet size: 1000 bytesBuffer size: 64, 128, 512 packetsMaximum TCP window size: 64 segmentsTCP tic: 0.5 s

Flow length distribution (when mixing different flow lengths)

Page 51: Colored GSPN Models  for the QoS Design  of Internet Subnets

52

modelsim

2040

6080

100120

N1

100200

300400

N2

2

3

4

5

6

Average window size

Elephants crossing both bottlenecks

Page 52: Colored GSPN Models  for the QoS Design  of Internet Subnets

53

modelsim

20 40 60 80 100 120N1

100200

300400

N2

0.01

0.1

Packet loss probability

Elephants crossing both bottlenecks

Page 53: Colored GSPN Models  for the QoS Design  of Internet Subnets

54

0200

400600

8001000N1

03000

60009000

12000

N2

0.01

0.1

Packet loss probability

Elephants with increased channel data rates (100 Mb/s -- 1 Gb/s)

Page 54: Colored GSPN Models  for the QoS Design  of Internet Subnets

55

0.001

0.01

0.1

0.75 0.8 0.85 0.9 0.95 1Offered load

Bottleneck 1analysis - B = 128analysis - B = 64

Mice (NewReno)P

acke

t lo

ss

pro

bab

ilit

y

Page 55: Colored GSPN Models  for the QoS Design  of Internet Subnets

56

0.2

0.5

1.0

2.0

5.0

10

0.75 0.8 0.85 0.9 0.95 1

Ave

rag

e co

mp

leti

on

tim

e (s

)

Offered load

10 packets20 packets100 packets

Mice (NewReno)

Page 56: Colored GSPN Models  for the QoS Design  of Internet Subnets

57

Outline

The Internet today

Dimensioning IP networks

GSPN and Queuing network models

Fluid approaches

Conclusions

Page 57: Colored GSPN Models  for the QoS Design  of Internet Subnets

58

V. Misra, W. Gong, D. Towsley, "Stochastic Differential Equation Modeling and Analysis of TCP Windowsize Behavior“, Performance'99

T. Bonald, "Comparison of TCP Reno and TCP Vegas via FluidApproximation," INRIA report no. 3563, November 1998

V. Misra, W. Gong, D. Towsley, "A Fluid-based Analysis of a Network of AQM Routers Supporting TCP Flows with an Application to RED“, SIGCOMM 2000

Literature

Page 58: Colored GSPN Models  for the QoS Design  of Internet Subnets

59

Y.Liu, F.Lo Presti, V.Misra, D.Towsley, Y.Gu, "Fluid Models and Solutions for Large-Scale IP Networks", SIGMETRICS 2003

F. Baccelli, D.Hong, "Interaction of TCP flows as Billiards“, Infocom 2003

F.Baccelli, D.Hong, "Flow Level Simulation of Large IP Networks“, Infocom 2003

Literature

Page 59: Colored GSPN Models  for the QoS Design  of Internet Subnets

60

Abandon a microscopic view of the IP network behavior, and model packet flows and other system parameters as fluids

The system is described with differential equations

Solutions are obtained numerically

Modeling approach

Page 60: Colored GSPN Models  for the QoS Design  of Internet Subnets

61

A simple example:

One bottleneck link

RED buffer

Elephants only (no slow start)

Modeling approach

Page 61: Colored GSPN Models  for the QoS Design  of Internet Subnets

62

TCP model

dWs(t)/dt = 1/Rs(t) – Ws(t) s(t) / 2

Where:

• Ws(t) is the average window • Rs(t) is the average round trip time

• s(t) is the congestion indication rate

of TCP sources of class s at time t

Page 62: Colored GSPN Models  for the QoS Design  of Internet Subnets

63

IP network model

dQk(t)/dt = Σs Ws(t) (1-P(t)) / Rs(t)

– - C 1{Qk(t)>0}

Where:

• Qk(t) is the length of queue k at time t

Page 63: Colored GSPN Models  for the QoS Design  of Internet Subnets

64

IP network model

Rs(t) = PDs + Qk(t)/C

Where:

• PDs is the propagation delay for source s

Page 64: Colored GSPN Models  for the QoS Design  of Internet Subnets

65

IP network model

s(t+Rs(t)) = Ws(t)/Rs(t) P(t)

Where:

• P(t) is the loss probability at time t

P(t) = RED(Q(t))

Page 65: Colored GSPN Models  for the QoS Design  of Internet Subnets

66

Fluid models – work in progress

• Slow start (mice)• Droptail buffers• Finite window• Threshold • Distributions • Fast recovery• Core network topologies

Page 66: Colored GSPN Models  for the QoS Design  of Internet Subnets

67

Fluid models – results

Page 67: Colored GSPN Models  for the QoS Design  of Internet Subnets

68

Fluid models – results

Page 68: Colored GSPN Models  for the QoS Design  of Internet Subnets

69

Fluid models – results

Page 69: Colored GSPN Models  for the QoS Design  of Internet Subnets

70

Fluid models – results

Page 70: Colored GSPN Models  for the QoS Design  of Internet Subnets

71

Fluid models – results

Page 71: Colored GSPN Models  for the QoS Design  of Internet Subnets

72

Fluid models – results

Page 72: Colored GSPN Models  for the QoS Design  of Internet Subnets

73

Fluid models – results

Page 73: Colored GSPN Models  for the QoS Design  of Internet Subnets

74

Fluid models – results

Page 74: Colored GSPN Models  for the QoS Design  of Internet Subnets

75

Fluid models – results

Page 75: Colored GSPN Models  for the QoS Design  of Internet Subnets

76

Fluid models – results

Page 76: Colored GSPN Models  for the QoS Design  of Internet Subnets

77

Fluid models – results

Page 77: Colored GSPN Models  for the QoS Design  of Internet Subnets

78

Fluid models – results

Baccelli and Hong obtained results for an access network with over one million TCP flows, about ten thousand routers, and link capacities from 6 Mb/s to 50 Gb/s.

Page 78: Colored GSPN Models  for the QoS Design  of Internet Subnets

79

Outline

The Internet today

Dimensioning IP networks

GSPN and Queuing network models

Fluid approaches

Conclusions

Page 79: Colored GSPN Models  for the QoS Design  of Internet Subnets

80

Conclusions

Fluid models today seem the most promising approach to study large IP networks

Tools for the model development and solution are sought

Fluid Petri Nets may be helpful for the model construction

Efficient numerical techniques are a challenge

Page 80: Colored GSPN Models  for the QoS Design  of Internet Subnets

81

Conclusions

The modeling paradigms to study the Internet behaviour are changing

This is surely due to scaling needs, but probably also corresponds to a new phase of maturity in Internet modeling

Reliable predictions of the behavior of significant portions of the Internet are within our reach

Page 81: Colored GSPN Models  for the QoS Design  of Internet Subnets

82

A comment on model complexity

time

models used

understanding of mechanisms being modeled

models proposed

early middle late

modelcomplexity

Adapted from [Hluchyj 2001], [Kurose 2001]

We need to go down

Page 82: Colored GSPN Models  for the QoS Design  of Internet Subnets

83

Thank You !