33
QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

Embed Size (px)

Citation preview

Page 1: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

QOS Routing: The Precomputation Perspective

Ariel Orda and Alexander Sprintson

Presented by: Jing, Niloufer, Tri

Page 2: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

2

Outline of the Presentation• Introduction• Algorithm for hierarchical network, bottleneck wt.

– Clustering Algorithm

– Find Algorithm

• Pseudo-Polynomial Solution for Additive weights• RTP and RTCP• Conclusion

Page 3: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

3

Motivation

• QOS path selection that – meets connection demand – uses fewer resources in network.

• Use Precomputation instead of On-Demand for QOS path selection.

• Use hierarchical structure to improve the computational complexity.

Page 4: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

4

QoS Routing Is Complex

• Many requirements specified which translate to constraints on path selection.

• Need to consider global network resources also.

• Identity of optimal path is connection dependent but doing path selection for every request is expensive.

Page 5: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

5

Making it Simpler

• Individual Connection requirements:– Translate different types of QOS requirements

to Bottleneck requirements. Easier than additive.

• Global Network Optimization:– Use hop minimization. Fewer hops Fewer resources.

• Computations for each Request:– Do Precomputation. Rate of requests < rate of changes in

network state

Page 6: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

6

Restricted Shortest Path Problem

• Given are source node s, destination node d and a QOS requirement w. Find a path p from s to d such that:– W(p) <= w

– C(p)<= C(p1) for every other path p1 that satisfies the restriction W(p) <= w

– there does not exist another path p2 for which C(p2) = C(p) and W(p2) <= W(p)

Page 7: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

7

Bottleneck weight Hop cost

• Given are a source node s, a destination node d, and a bottleneck QoS requirement w. Find a path p from s to d such that:– W(p) <= w

– |p|<= |p1| for every other path p1 that satisfies the restriction W(p) <= w

– there does not exist another path p2 for which |p2| = |p| and W(p2) <= W(p)

Page 8: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

8

Simple Precomputation Scheme• Use Bellman-Ford algorithm.• Compute minimum weight for each

possible hop count between s and d.• Upon connection request, choose minimum

hop value for which the corresponding path meets QoS requirement of connection.

• Precomputation Complexity: O(MH)• Find Complexity: O(logH + |p|)

Page 9: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

9

Clustering Algorithm

• Input: – Layer i peer group C

– Node s’

– Arrays TC’[v1,v2,h] for peer groups at level i-1 that make up this peer group C

• Output:- Compute TC[s’,v,h] for all v and h, where v is border

node of C and h is hop count.

Page 10: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

10

Bottleneck Weight, Hop Cost Algorithm for Hierarchical NW

• For each peer group, at each layer, considering each border node as s’, call Algorithm for Clustering

Page 11: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

11

S:

(a,b,1,10)

(b,c,1,20)

(a,e,1,25)

(b,d,1,30)

(d,e,1,30)

(c,d,1,40)

(b,e,1,50)

a

b

d

e

c

50

30

20

10

25

30

40

Border node

Border node

S’

Output:(To be used at layer 2)TC[s’,e,1] = 50TC[s’,e,2] = 25TC[s’,e,3] = 25TC[s’,e,4] = 25

Page 12: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

12

a

b

d

e

c

Border node

Border node

S’

Page 13: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

13

Page 14: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

14

S’

Page 15: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

15

Page 16: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

16

Page 17: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

17

Page 18: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

18

FOR THE FIRST LAYER (ACTUAL NETWORK).Identify peer group that contains the destination node

.Remove all links that do not meet the constraints.Use Breath First Search, find “best” path from the border

nodes to the destination node

FOR ALL OTHER LAYERS (layer 2 to k).Identify peer group that contains the destination node

.Use pre-computed data (Tc array) and Dijkstra to find “best” path from the border nodes to the destination node

.Repeat the process until reach the last layer (layer k)

Page 19: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

19

PRECOMPUTATION SCHEME FOR ADDITIVE METRICS & PSEUDO-POLYNOMIAL SOLUTION

Bottleneck metric:Weight of the path = Weight of the worse link

Additive metric:Weight of the path = Sum of the weight of all links

NP-hard problemBecause need to exhaust the solution space O(M C)

M: number of links C: all of possible value of weightPseudo-polynomial because the value of C does not correlate (polynomial) to number of links or vertices in the network.

Page 20: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

20

Additive QoS Constraints (Delay)

• Mathematic Approach:– Example: Pseudo-Polynomial Solution for RSP problem

• Heuristic Approach:– Example: RTP

Page 21: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

21

RTP Overview

• RTP (Real-time Transport Protocol)– Designed as an end-to-end transport protocol

that is suitable for unicast or multicast transmission of real-time data

– Used in combination with other transport protocols---typically on top of UDP

Page 22: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

22

RTP Overview (Cont.)

• Why Not TCP?– RTP is primarily designed for multicast, the

connection-oriented TCP does not scale well and therefore is not suitable

– For real-time data, reliability is not as important as timely delivery

Page 23: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

23

RTP Overview (Cont.)• RTP can be viewed as a sub layer of the transport layer

Page 24: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

24

RTP Overview (Cont.)

Page 25: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

25

What RTP Provides:

• Timestamping– Set according to the instant the first octet in the packet

was sampled

– Used to reconstruct the original timing in order to play out the data in correct rate

• Sequence numbering– Used to place the incoming data packets in the correct

order

– Also used used for packet loss detection

Page 26: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

26

What RTP Provides: (Cont.)

• Payload type identification– Specifies the payload format as well as the

encoding/compression schemes

Page 27: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

27

What RTP does NOT provide:

• Does not make any QoS commitments

• Does not guarantee reliable, timely, or in-order delivery

• Does not enforce any error treatment measures

Page 28: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

28

RTCP Overview

• RTCP (Real-Time Control Protocol)– Designed as a control protocol that works in

conjunction with RTP

– Used to convey feedback on quality of data delivery and information of membership in an RTP session

Page 29: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

29

What RTCP Provides:

• QoS monitoring and congestion control – By sending feedbacks to applications about the quality

of data distribution

• Source identification – Information may include user's name, telephone

number, email address and other information

Page 30: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

30

What RTCP Provides: (Cont.)

• Control information scaling – Control traffic is limited to at most 5% of the overall

session traffic

Page 31: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

31

Useful Links

• Http://www.cis.ohio-state.edu

• http://www.cs.columbia.edu

• Http://www.Seas.upenn.edu

• Http:// www4.informatik.uni-erlangen.De

Page 32: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

32

Conclusion

• There are different approaches to tackle QoS constraints:– Algorithmic based

• Bottleneck

• Additive

– Heuristic based

Don’t let NP-Hard stop us!!

Page 33: QOS Routing: The Precomputation Perspective Ariel Orda and Alexander Sprintson Presented by: Jing, Niloufer, Tri

33