Kein Folientitel

1
Traffic Engineering for Hard Real Time Multicast
Applications Using Genetic Algorithm
Shu Zhang, Lothar Kreft and Ulrich
Killat 15.03.2005
Communication Networks
Prof. Dr. U. Killat
2
Problem Setting
End Systems 1
Gateway 1

• N (Ngt100) multicast groups
• Guarantee of end-to-end
• delay bounds

Gateway 2


Gateway M
End Systems N
Communication Networks
Prof. Dr. U. Killat
3

• Index
• Problem setting
• EDF scheduler and delay bounds
• Genetic encoding
• The performance of our algorithm
• Conclusion

Communication Networks
Prof. Dr. U. Killat
4
The Earliest Deadline First Scheduler
Packet of flow 'j' arrive at time t2
EDF Scheduler
Packet of flow 'i' arrive at time t1
Assigned a deadline of t1di
Assigned a deadline of t2dj
di , dj predefined delay bounds for flow i, j
on this link.
Deadlines are guaranteed when the flows are
"schedulable".
Communication Networks
Prof. Dr. U. Killat
5
The Schedulability Function
Schedulability function F(t)
F(t)
Ct
A1(t-d1,j)
A2(t-d2,j)
A3(t-d3,j)
All flows are schedulable iff.
t
0
d1,j
d2,j
d3,j
Communication Networks
Prof. Dr. U. Killat
6
The Scheduling and Bandwidth Margins
With token bucket as traffic envelope, we define
the scheduling margin
F(t)
Bj
define bandwidth margin
Sj
?1
?2
?3
s1
s2
s3
t
0
d1,j
d2,j
d3,j
gt Optimization objectives
Communication Networks
Prof. Dr. U. Killat
7
Assigning Delay Bounds to Links
delay bounds assignment procedure
Communication Networks
Prof. Dr. U. Killat
8
Creating Alternative Routing Candidates
4
0
4
0
3
3
1
5
1
5
6
6
2
2
Shortest path tree with original topology
Shortest path tree with topology without link 1-2
Communication Networks
Prof. Dr. U. Killat
9
Genetic Encoding
Gene Routing tree Chromosome Complete routing
planning Fitness Scheduling and bandwidth
margins
A gene
Parents

Chromosome A
1
2
3
N

Chromosome B
N
1
2
3
Crossover
Crossover
Offspring

Chromosome A'
1
2
N
3

Chromosome B'
N
1
3
2
Mutation
Communication Networks
Prof. Dr. U. Killat
10
Performance of the Planning Algorithm
Scheduling margin (byte)
Bandwidth margin (byte/ms)
Generation
Generation
210 multicast groups
Communication Networks
Prof. Dr. U. Killat
11
Performance of the Planning Algorithm
Scheduling margin (byte)
Bandwidth margin (byte/ms)
Generation
Generation
350 multicast groups
Communication Networks
Prof. Dr. U. Killat
12

• Conclusion
• Hard real time QoS can be guaranteed by proper
  scheduling mechanism and routing selection.
• A traffic engineering approach using genetic
  algorithm has been developed.
• The performance of the algorithm shows its
  effectiveness.

Communication Networks
Prof. Dr. U. Killat
13
The Traffic Envelopes and Solutions for the
Multi-node Case
Flow j with envelope Aj(t)
Flow j with envelope Aj(tdj)
(1)
EDF Scheduler
Flow i with envelope Ai(t)
Flow i with envelope Ai(tdi)
Shaper with envelope Aj(t)
Flow j with envelope Aj(tdj)
(2)
Flow j
EDF Scheduler
Flow i
Shaper with envelope Ai(t)
Flow i with envelope Ai(tdi)
Communication Networks
Prof. Dr. U. Killat
14
Comparison to a naive approach
Scheduling margin (byte)
Bandwidth margin (byte/ms)
Computation time (minutes)
Computation time (minutes)
210 multicast groups
Communication Networks
Prof. Dr. U. Killat