Quality-of-Service Routing in IP Networks

1. Quality-of-Service Routing in IP Networks IEEE TRANSACTIONS ON MULTIMEDIA JUNE 2001 Donna Ghosh, Venkatesh Sarangan, and Raj Acharya

2. Outline • Introduction • Two-Level Approach • Routing Table Maintenance • Packet Forwarding Mechanism • Experimental Results and Discussion • Conclusion

3. Introduction (1/3) • QoS routing Goal => To find a loop-less path satisfying a given set of constraints on parameter like bandwidth, delay, etc. (optimize the global network resource utilization) • (RSVP + IP) v.s. (RSVP + Qos routing)  Need a QoS routing algorithm in IP networks.

4. Introduction (2/3) • Need of QoS Routing Design • Minimal changes to the existing routing protocols. • Tradeoff between the average number of messages exchanged and impreciseness of maintained global state. • Source QoS routing • Distributed QoS routing • Maintain global state (source routing) • Not Maintain global state (flooding)

5. Introduction (3/3) • Maintain global state • Specify the best next hop. • Overhead in maintaining the global state and state impreciseness. • No Maintain global state • High overhead on establishing a connection. The paper’s method complements this bounded flooding approach.

6. Two-Level approach (1/2)

7. Two-Level approach (2/2) • The advantage of Two-Level: • Message overhead and the impreciseness will not be as large as maintaining the global state. • Using the information about the second-degree neighbors, forward intelligently instead blindly flooding. • The approach could extend Nth degree neighbor.

8. Example network

9. QoS Routing algorithm • Table Maintenance • Building the routing table • Update Policies • Packet Forwarding • Bounded Two-Level forwarding

10. Routing Table Maintenance – building the routing table • Building the Routing Table • LTN table (Link-to-Node) • Construct by exchanging Hello packets with neighbors. • Forwarding table • contains information about the metrics of all the links in E1(v) and E2(v). • Routing table (two-level) • First level entry => copy the LTN table of v • Second level entry => updated by Hello2 packet

11. Routing Table Maintenance – Update Policies (1/2) • The update policy used decides when these Hello2 packets are sent. • Based on timers • Based on Thresholding (adopt) • Each node remembers the last advertised metric on each link. If the ration is above (or below) a threshold , an update is triggered.

12. Routing Table Maintenance – Update Policies (2/2) • Advantage of threshold-based update policy the impreciseness could be easily modeled using probabilities. • If bandwidth b is advertised on a link ,and say is 2, cab be modeled as a uniform distribution in [b/2, 2b] • There are some approaches[10],[12] to do efficient routing with such imprecise information. Ref: 1.INFOCOM 97, “QoS routing in networks with inaccurate information: Theory and algorithms” 2.INFOCOM 98, “QoS routing in networks with uncertain parameters”

13. Packet Forwarding Mechanism • The connection setup process has three phases: probing, ack, failure • Probe packet format [k, QoS(Bandwidth=B), s, t, cid, {l}.] K: the router that has forwarded to probe to v QoS(): QoS requirement function s : Source t :destination cid :unique identifier {l} :list of neighbors to which v should forward this probe

14. Fig. Packet forwarding at a node

15. Drawback of forwarding mechanism • The destination sends an ack only to the first probe it receives and discard all the duplicates. • Improve way: 1. Receiver can retain the best one of probe packet, besides the first arrive. 2. Failure happened, send failure message to receiver and sender in the same time.

16. Bounded Two-Level Forwarding(1/2) • If the network load is light, it is not a wise idea to blindly flood the probes on all eligible links. • The shortest eligible path is preferred. • Each probe is assigned an age. Initially, age(p) =0

17. Bounded Two-Level Forwarding(2/2) • Forward condition on link(i, j) at node i:  bandwidth(i,j) B and (age(p) + dj,t +1  L ) • Much less overhead when the network load is light. • First, L = ds,t Second, L =  , when network load heavily. • Simulation: Unbounded – without the hop constraint (L = ) Bounded – with the hop constraint (L = ds,t)

18. Experimental Results and Discussion • To have such a reduced overhead, additional information about the second-degree neighbors must be stored at each router. • Prove: overhead of table maintenance is much less than the savings in probe forwarding. • Summary  The savings in the probe forwarding is dependent on resource availability and the network topology.

19. Simulation Network Topology

20. Compare manner • First set: compare between the unbounded versions of flooding and two-level forwarding. (L = ) • Second set: compare between the bounded versions of the two approaches. (L = ds,t)

21. Fig5. Unbounded-flooding Overhead on Mesh-1 Fig7. Unbounded-flooding Overhead on ISP Compare Overhead per Call Admitted (L = )

22. Fig6. Unbounded-flooding Bandwidth admitted on Mesh-1 Fig8. Unbounded-flooding Bandwidth admitted on ISP Bandwidth admission ratio : the ratio of bandwidth admitted into the network to the total bandwidth requested. (L = )

23. Compare Overhead per Call Admitted (L = ds,t)

24. Bandwidth admission ratio : the ratio of bandwidth admitted into the network to the total bandwidth requested. (L = ds,t)

25. Conclusion • Propose a new distributed QoS routing algorithm which has a very low call establishment overhead. • Characters: • Two-Level routing table maintain • Threshold-based update policy • The age of a probe is defined