Post on 05-Jan-2016
“Improving Fairness for Multi-Hop Bursts in
Optical Burst Switching Networks”
Tananun Orawiwattanakul Yusheng Ji Yongbing Zhang
Asia Pacific Advanced Network 2007, 27-31 August 2007, Xi’An, People’s Republic of China.
Topics
• Optical Burst Switching (OBS) Networks
• Hop Based Preemption (HBP)
• Numerical results
• Future works
LAN
Optical Burst Switching (OBS)
Core Switch Core SwitchEdge Switch Edge Switch
Users
LAN
UsersOBS Network
Optical Burst Switching (OBS)
Control Wavelengths
Data Wavelengths
Optical Burst Switching (OBS)
DB
C
Control Wavelengths
Data WavelengthsElectronic
IP Packets
Optical
Optical Burst Switching (OBS)
DB
C
Offset Time
Control Wavelengths
Data Wavelengths
Optical Burst Switching (OBS)
DB
Offset Time
C
Optic Electronic Process
Optic
C
Reserve this wavelength for the incoming data burst
Control Wavelengths
Data Wavelengths
Optical Burst Switching (OBS)
C
Optic Electronic Process
Optic
C C
DB
Full Wavelength Conversion
DB
Offset Time
Control Wavelengths
Data Wavelengths
Optical Burst Switching (OBS)
DB
Offset Time
C
Control Wavelengths
Data Wavelengths
Optical Burst Switching (OBS)
DB
Optical Electronic
IP Packets
General Networks
S5
S1
S2 S3
S4
Buffer
General Networks -> Electrical Buffer Loss occurs only during high traffic load
DB4
Data ChannelsDB1
DB3
DB2
Time
Control Channels
C1 C2
C3
C4
T1 T2 T3 T4 T5
Offset Time1
Offset Time1
Offset Time1
Offset Time1
Offset Time1
C5DB5
Offset Time5
No Buffer -> Contention occurs -> High Losses even in the light traffic load
The longer path -> The higher probability of Loss -> Unfairness
Challenge of OBS
Hop Based Preemption (HBP)
• The objective of HBP is to improve fairness for multi-hop bursts in OBS networks.
• The burst that has traveled through many nodes and has a high possibility to arrive at a destination can preempt a channel from other scheduled bursts.
• HBP support non/full/limited wavelength conversion networks.
• HBP is implemented only in core switches.
HBP
DB4
Data Channels
DB1
DB3
DB2
Time
Control Channels
C1 C2
C3
C4
T1 T2 T3 T4 T5
Offset Time1
Offset Time1
Offset Time1
Offset Time1
C5DB5
Offset Time5 Contending Burst
HBP Scheme Parameters
Parameters Description
Hp No. of hops between ingress switch of burst and corresponding switch.
HoNo. of hops between corresponding switch and egress switch of burst.
Wp Weight of Hp ( >=0)
Wo Weight of Ho ( >=0)
MThe minimum ratio of No. of hops that burst has traveled through and No. of hops between burst’s ingress and egress switches
I Hop Index
T Threshold ( >=0)
HBP
DB4
DB1
DB3
DB2
DB5
Contending BurstI5
I4
I1
I3
I2
Scheduled Bursts
DB5
Contending BurstI5
Source5 Destination5
Hp = 3 Ho = 3
Contending will be dropped if
HpContending / (HpContending + HoContending) < M M = 1/23/ (3+3) = 1/2
HBP
DB4
DB1
DB3
DB2
DB5
Contending Burst
I5 = 0Scheduled Bursts
DB5
Contending Burst
Source5 Destination5
Hp = 3 Ho = 3
I = (Hp * Wp ) - (Ho * Wo )
Wp = Wo = 1
I of the contending burst = 3-3 (Hp-Ho) = 0
HBP
DB4
DB1
DB3
DB2
DB5
Contending Burst
I5 = 0Scheduled Bursts
DB5
Contending Burst
Source5 Destination5
Hp = 3 Ho = 3
I4 = 1
I1 = -1I3 = 3
I2 = 0
Least I
The contending burst can preempt the channel of the original burst if
I5 – I1 >= TWhen T = 10-(-1) = 1
The contending burst wins the contention.
HBP
The corresponding switch cancels the original wavelength reservation for the original burst and instead allocates the
wavelength for the contending burst.
DB4
Data ChannelsDB3
DB2
Time
Control Channels
C2
C3
C4
T2 T3 T4 T5
Offset Time1
DB1
C1
T1
Offset Time1
Offset Time1
Offset Time1
C5DB5
Offset Time5 Contending Burst
DB5
Numerical Results
• Simulated on Optical Internet Research Center’s (OIRC) optical burst switching simulator based on ns-2.
• Based on NSF and ARPA networks
• No losses in control channels
• Bandwidth per wavelength = 1 Gbps and the processing time of a BCP at each switch = 0.01 msec
• Hybrid burst assembly
NSF Network
1100
600
1600
1000
2000 1200
2000
900500
300
300
800
24002800
800500
1
0
2
34
6
7
5
8 13
12
9
1110
600800
1100
700
7001100
600
1600
1000
2000 1200
2000
900500
300
300
800
24002800
800500
11
00
22
3344
66
77
55
88 1313
1212
99
11111010
600800
1100
700
700
NSF Network
0
0.01
0.02
0.03
0.04
0.05
0.06
3 4 5 (General OBS)
3 4 5 1:1:1
3 4 5 1:1:2
3 4 5 1:0:1
3 4 5 1:0:2
No. of hops(Wp:Wo:T )
Burst lossprobability (%)
NSF Network
0.025
0.0255
0.026
0.0265
0.027
0.0275
General_OBS 1:1:1 1:1:2 1:0:1 1:0:2Wp:Wo:T
Total burst lossprobability (%)
ARPA Network
1
2
5 12
15
18
20
4
0
3
7
6
8 9
1014
13
11 17
16
1911
22
55 1212
1515
1818
2020
44
00
33
77
66
88 99
10101414
1313
1111 1717
1616
1919
ARPA Network
0
0.05
0.1
0.15
0.2
0.25
3 4 5 6 7 8 9 (General OBS)
3 4 5 6 7 8 9 (1:1:1)
3 4 5 6 7 8 9 (1:1:2)
3 4 5 6 7 8 9 (1:0:1)
3 4 5 6 7 8 9 (1:0:2)
No. of hops(Wp:Wo:T )
Burst lossprobability (%)
ARPA Network
0.05
0.052
0.054
0.056
0.058
0.06
0.062
0.064
General_OBS 1:1:1 1:1:2 1:0:1 1:0:2
Wp:Wo:T
Total burst lossprobability (%)
Future works
• Resolve more complicated unfairness issues, such as unfairness caused by bottle-neck links.
• Provide flow level fairness
• Decrease the total loss probability.