Available Bandwidth

1
Available Bandwidth

• SpeakerChun Shih Lin (???)
• Date 2005/03/19

2
Outline

• Introduction
• Background
• Related work
• Proposed method

3
Outline

• Introduction
• Background
• Related work
• Proposed method

4
Introduction (1/4)

• Problem How to measure available bandwidth for
  an end-to-end network path ?

Receiver
Sender
Internet
5
Introduction (2/4)

• Available bandwidth
• Important parameter for admission control, QoS
  management, streaming applications

Receiver
Sender
Internet
According to those received probing packets, we
can measure and analyze the available bandwidth
Probing packet
6
Introduction (3/4)

• General Definition

n number of hops in a path
capacity of linki avail-bw of the path
at time t utilization of linki

• Example

Link1 2Mbps 20
Link2 1Mbps 30
Link3 3Mbps 50
Sender
Receiver
A(2) C1 x (1-u1) 1.6Mbps A(2) C2 x (1-u2)
0.7Mbps A(2) C3 x (1-u3) 1.5Mbps
A(2) C2 x (1-u2) 0.7Mbps
7
Introduction (4/4)

• Available bandwidth changes with time,
  measurement is a big challenge
• How to find bottleneck ?
• Tight link is not also bottleneck
• Other definition
• Average transmission rate

8
Outline

• Introduction
• Background
• Related work
• Proposed method

9
Background
10
Outline

• Introduction
• Background
• Related work
• Proposed method

11
IGI (Initial Gap Increasing) (1/4)

• Gap (time)

back-to-back packet pair
gi
go
P2
P1
gi initial gap (time between first
bit of P1 and P2 when they enter the
router) go output gap (time between
first bit of P1 and P2 when they leave
the router)
P2
P1
P2
P1
router
Cross traffic packets

• Benefit Reflect whether the link is bursting or
  not

• Tradeoff The length of gi
• Too short Cause probe overflow
• Too long Underestimate the available bandwidth

12
IGI (2/4)

• Single-hop gap model

• First phase

Receiver
gi
Sender

P1
P2
P3
bottleneck
Cs
Cr
Cs Cross traffic Sender Cr Cross traffic
Receiver

• packet size 700 bytes

• Probing packets length 60 packets (per phase)

gi initial gap go output gap
13
IGI (2/4)

• Single-hop gap model

• First phase

Receiver
go
Sender

P2
P1
P3
bottleneck
Stop or continue next phase
Cs
Cr
Cs Cross traffic Sender Cr Cross traffic
Receiver

• packet size 700 bytes

• Probing packets length 60 packets (per phase)

gi initial gap go output gap
14
IGI (2/4)

• Single-hop gap model

?

• Second phase

New gi gi ?
Receiver
Sender

P1
P2
P3
bottleneck
Cs
Cr
Cs Cross traffic Sender Cr Cross traffic
Receiver

• packet size 700 bytes

• Probing packets length 60 packets (per phase)

gi initial gap go output gap
15
IGI (2/4)

• Single-hop gap model

• Second phase

Receiver
go
Sender

P2
P1
P3
bottleneck
Collecting the each back-to-back packets
intervals (gi and go)
Cs
Cr
Cs Cross traffic Sender Cr Cross traffic
Receiver

• packet size 700 bytes

• Probing packets length 60 packets (per phase)

gi initial gap go output gap
16
IGI (3/4)
average gap difference (s)

• Turning point Avg_go - Avg_gi0 then algorithm
  stops

Turning point
0
initial gap (s)
gi

• Cross traffic throughput

Increased_gaps
P2
P1
P2
P1
router
Cross traffic packets
gi initial gap go output gap
Bottleneck link capacity
transmission delay on bottleneck link
17
IGI (4/4)

• Example

Bottleneck link capacity
gi initial gap go output gap
receive
P3
P2
P1
time
2
3
1
1
send
P3
P2
P1
11
6
8
time
1
2
0
18
PTR (Packet Transmission Rate) (1/6)

• Single-hop gap model

• First phase

Receiver
gi
Sender

P1
P2
P3
bottleneck
Cs
Cr
Cs Cross traffic Sender Cr Cross traffic
Receiver

• packet size 700 bytes

• Probing packets length 60 packets (per phase)

gi initial gap go output gap
19
PTR (Packet Transmission Rate) (2/6)

• Single-hop gap model

• First phase

Receiver
go
Sender

P2
P1
P3
bottleneck
Stop or continue next phase
Cs
Cr
Cs Cross traffic Sender Cr Cross traffic
Receiver

• packet size 700 bytes

• Probing packets length 60 packets (per phase)

gi initial gap go output gap
20
PTR (Packet Transmission Rate) (3/6)

• Single-hop gap model

?

• Second phase

New gi gi ?
Receiver
Sender

P1
P2
P3
bottleneck
Cs
Cr
Cs Cross traffic Sender Cr Cross traffic
Receiver

• packet size 700 bytes

• Probing packets length 60 packets (per phase)

gi initial gap go output gap
21
PTR (Packet Transmission Rate) (4/6)

• Single-hop gap model

• Second phase

Receiver
go
Sender

P2
P1
P3
bottleneck
Collecting the each back-to-back packets
intervals (gi and go)
Cs
Cr
Cs Cross traffic Sender Cr Cross traffic
Receiver

• packet size 700 bytes

• Probing packets length 60 packets (per phase)

gi initial gap go output gap
22
PTR (Packet Transmission Rate) (5/6)

• Algorithm terminates at turning point

• Average transmission rate of packet train

Available bandwidth
23
PTR (Packet Transmission Rate) (6/6)

• Example

• packet size 700 bytes

gi initial gap go output gap
receive
P3
P2
P1
time
2
3
1
1
send
P3
P2
P1
11
6
8
time
1
2
0
24
Outline

• Introduction
• Background
• Related work
• Proposed method

25
Proposed method (1/10)

• Create a gap function
• Gap is increased by a better choice
• Improve IGI/PTR probing phases

26
Proposed method(2/10)

• Single-hop gap model

• First phase

Receiver
gi
Sender

P1
P2
P3
bottleneck
Cs
Cr
Cs Cross traffic Sender Cr Cross traffic
Receiver

• packet size 700 bytes

• Probing packets length 60 packets (per phase)

gi initial gap go output gap
27
Proposed method (3/10)

• Single-hop gap model

• First phase

Receiver
go
Sender
P2
P1
bottleneck
Stop or continue next phase
Cs Cross traffic Sender Cr Cross traffic
Receiver
Cs
Cr
gi initial gap go output gap

• packet size 700 bytes

• Probing packets length 60 packets (per phase)

28
Proposed method (4/10)

• Single-hop gap model

• Second phase

Sender
Receiver
New gi gi ?


P2
P1
P3
bottleneck
Cs
Cr
? generate by gap function
Cs Cross traffic Sender Cr Cross traffic
Receiver
gi initial gap go output gap
29
Proposed method (5/10)

• Gap function
• Analysis from average_gi and average_go

• Example

• At first phase function should offer an better
  increased gap
• to next probing initial gap.

• Initial gap set too small
• So set ? 1

• At second phase

New gi gi 1
30
Proposed method (6/10)

• Traffic models
• Policed by leaky bucket policing mechanism

Burst parameter
Average rate parameter
Traffic
Traffic of a connection in interval t
Time
31
Proposed method (7/10)
arrival rate of probing packet
difference of departure time between P1 and P2
arrive
departure
P1
P1
departure
arrive
P2
P2
32
Proposed method (8/10)

• Departure time of P1

burst parameter average rate parameter
output link capacity packet size
Maximum waiting time

• Departure time of P2

33
Proposed method (9/10)

• Difference of departure time between P1 and P2

34
Proposed method (10/10)

• Experiment scenario (using NS2)
• Single hop model
• Packet size 700 bytes
• Each probing uses 60 packets
• Cross traffic is generated by CBR (Constant Bit
  Rate)

35
END
36
Probing

• Active Probing
• Effectively detect endpoints situation by
  diagnosing timely information (egg. RTT)
• Introducing additional network traffic

37
Tools and Algorithms
38
Cprobe (1/2)
Sender
Receiver
Internet
first
last
Echo transmission time
Probing until last echo packet report
time between 1th and last packet
39
Cprobe (2/2)

• Problem
• No concern about cross traffic
• Only adapted for local area network

40
Delphi (1/3)
Packet train
Sender
Receiver

Internet
Analyzing packets interarrival time and
calculating available bandwidth
20
20
21
22

probing
1
2
3
4
5
6
7
8
9
10
12
13
time
41
Delphi (2/3)
Dynamic available bandwidth in T
Index of link of a queue
A link capacity of a queue
A time interval
Number of cross traffic bytes inserted in T
input
output
a router queue
Probe packet
A snapshot of a queue
Cross traffic packet
A single queue router
Empty
42
Delphi (3/3)

• Problem
• Exponentially increasing packet-spacing may
  overestimate the available bandwidth due to
  burst of cross-traffic

43
TOPP (Trains of Packet Pairs) (1/7)
Back-to-back packet pair
Sender
Receiver

Internet
tc
ta
tb
?TP
?TP

Source
Sink
b
b
b
b
b
b
ta gt tb gt tc
Probe packet with size b
b
44
TOPP (2/7)
Rate omin
Receiver
Sender
N packet pairs (flow 1)

Internet
back-to-back packet pair initial transmission
rate
Omin
45
TOPP (3/7)
Receiver
Sender

Internet
Measuring the receiving rate and decide next
probe transmission rate
back-to-back packet pair initial transmission
rate
Omin
46
TOPP (4/7)

• Increasing probing rate each probing rate
  increased by ?

New Rate omin?
Receiver
Sender
N packet pairs (flow 2)

Internet
Measuring the receiving rate and decide next
probe transmission rate
back-to-back packet pair initial transmission
rate
Omin
47
TOPP (5/7)

• Until probing rate met omax

Final Rate omax
Receiver
Sender
N packet pairs (flow n)

Internet
Compare the total transmission rates and
receiving rates
back-to-back packet pair initial transmission
rate
Omax
48
TOPP (6/7)

• Find congestible link

o
o transmission rate f receiving rate
Li congestible link capacity A available
bandwidth CTi cross traffic
f o
L1
L2
L3
f

• Decide available bandwidth

Link1 L12Mbps CT11Mbps
Link2 L21Mbps CT20.5Mbps
Link3 L32Mbps CT31Mbps
Sender
Receiver
A(2) L1-CT1 1.0Mbps A(2) L2-CT2
0.5Mbps A(2) L3 -CT3 1.0Mbps
Agt
49
TOPP (7/7)

• Problems
• Rate bound
• Packet pair length
• The correctness of Cross traffic could affect
  the estimation of available bandwidth

50
TOPP
f receiving rate O sending rate m cross
traffic rate l link capacity
t1 , t2 ,t3 each links
av-bw
51
TOPP
Omin initial probe rate Omax probe
packets end up with this rate
52
PathChirp (1/4)
B (sink)
A (source)
A chirp N probe packets with same size

Time
T?N-1
T?N-2
T?N-3
T
T?
Probe packets are separated by exponentially
spacing time
53
PathChirp (2/4)
A chirp typical queuing delay signature
54
PathChirp (3/4)

• Studying from the signature
• Excursion 1 if qk lt qk1 then Ek Rk,
• Ek per packet avail-bw
• Rk P/?k ,
• P packet size
• ?k spacing time between packets k and k1
• qk queuing delay of packet k
• Excursion 2
• l start packet of excursion 2
• Others

Ek Rl, kgtl
Ek Rx , x N-1
55
PathChirp (4/4)

• Final estimation

per chirp avail-bw per packet avail-bw
spacing time between packets k and k1
56
Pathload (1/4)

• SLoPS (Self-Loading Periodic Streams)

Sender
Receiver
internet
Sending K packets size L with a transmission rate
R which being sent in period T
Estimating the Avail-bw
probe packet timestamp
57
Pathload (2/4)

• Iterative rate adjustment algorithm

1
58
Pathload (3/4)

• Iterative rate adjustment algorithm

2
A is bounded in
59
Pathload (4/4)

• Problems
• SLoPS does not instantly reflect the burst of the
  link
• Iteration algorithm convergence time takes too
  long

60
OWD (One Way Delay)
Sender
Receiver
internet
one way delay per packet number of hops
packet size link capacity queuing delay
61
OWD (One Way Delay)
62
Delay detection
one way delay per packet group number
63
OWD (One Way Delay)
OWD of packet with size Lk N number of
links probe packet size capacity
of linki queue delay of packet k at link
i processing delay at link i
Receiver
Sender
Internet
????
64

• Available bandwidth Definition

n number of hops in a path
capacity of linki avail-bw of the path
at time t utilization of linki

• Example

Link1 2Mbps 20
Link2 1Mbps 30
Link3 3Mbps 50
Sender
Receiver
A(2) C1 x (1-u1) 1.6Mbps A(2) C2 x (1-u2)
0.7Mbps A(2) C3 x (1-u3) 1.5Mbps
65
IGI algorithm
66
PTR formula