Chapter 5 Introducing SDN Control in MPLS Networks - High ...

More documents

Recommendations

Info

209tunnel carries both video and VoIP traffic. In the SFO router, we simply use OpenFlow toenter rules that match on both destination IP address and the L4 transport-ports toindentify the tunnel and the traffic-type respectively; and then insert the appropriate label(as an ‘action’ on matching packets) for the tunnel the packets needs to enter. We canalso use the IPv4 DSCP bits to match on a traffic-class (if for example the San Jose routermarks the bits).Fig. 5.17 GUI Snapshot: Routing of Traffic-Classes and Load SharingIn this experiment, we do not perform per-class admission control as we did notreserve bandwidth from class specific sub-pools of link bandwidth. But this is a simplechange to our CSPF algorithm (and nothing else). More importantly we can createmultiple sub-pools each for a different traffic-class; to match with tunnels created formultiple traffic-classes (as shown); which is something that MPLS-TE cannot providetoday without changing the routing protocol.The second example of deviation from shortest-path routing shows load-balancing. InFig. 5.17, traffic between SFO and KAN takes two routes: Video flows go through a
210 CHAPTER 5. INTRODUCING SDN CONTROL IN MPLS NETWORKStunnel from SFOKAN, which is actually routed via the Seattle and Chicago routers;and all other flows take the IP links between SFODENKAN. This is an example ofload-sharing to a tunnel-tail, between a tunnel-path and an IP-link-path; a feature thatMPLS-TE cannot provide today due to the nature of Auto-Route in MPLS-TE (asdiscussed in the previous section).Conclusions: To summarize, we draw the following conclusions from our SDNbased MPLS network prototype:• We achieved the initial goal of verifying our architectural ideas of introducing themap-abstraction in MPLS networks. We showed that we can not only perform thebasic operations of creating and maintaining LSPs in an OpenFlow enabled MPLSdata-plane; but we can also provide higher-level services such as MPLS-TE bywriting applications above a map-abstraction purely in the control-plane.• We validated the simplicity of our approach when compared to more traditionalapproaches seen today. Our implementation of the traffic-engineering applicationtogether with all the features described in Sec. 5.2.1, took only 4500 lines-of-code. Toaccomplish the same in today’s network would require implementations of OSPF-TEand RSVP-TE (together ~ 80k LOC) and implementations of all the features. At thetime of this writing we could not find an open-source implementation of MPLS-TEfeatures; and so we can only estimate, based on experience that each feature couldtake at a minimum 5-10k lines-of-code, pushing the total LOC well above 100k. It iseasy to see that our SDN based implementation is much simpler given the two-ordersof magnitude difference in the lines-of-code required. While our implementations isnot production-ready, two orders of magnitude difference gives plenty of room togrow; giving us confidence that even production-ready code will be simpler toimplement with the map-abstraction.• Finally, we achieved our second goal of demonstrating that using SDN and the mapabstraction,we can implement the TE features in ways that a) either greatly simplifythe feature; or b) provide a feature that MPLS-TE cannot provide. Examples of the
Page 1: Chapter 5Introducing SDN Control in
Page 5 and 6: 176 CHAPTER 5. INTRODUCING SDN CONT
Page 7: 178 CHAPTER 5. INTRODUCING SDN CONT
Page 10 and 11: 181routers make decisions on LSP pa
Page 12 and 13: 183benefits of the map-abstraction
Page 14 and 15: 185all upgraded at the same time. M
Page 16 and 17: 187However the tunnel’s reserved
Page 18 and 19: 189Admission-Control: The MPLS cont
Page 20 and 21: 191This latter point can best be ex
Page 22 and 23: 193different customer’s traffic d
Page 24 and 25: 195the start of the tunnel, to the
Page 26 and 27: 197forwarding traffic-type (or clas
Page 28 and 29: 199controller, as well as link-stat
Page 30 and 31: 201netdev-linux module that abstrac
Page 32 and 33: 203packet-flow routing. The other t
Page 34 and 35: 205In Fig. 5.14 we show two tunnels
Page 36 and 37: 207forced to route along a less-uti
Page 40 and 41: 211former include (i) our implement
Page 42 and 43: 213these backup tunnels usually do
Page 44 and 45: 215annotated fiber topology, on whi
Page 46: 217of magnitude fewer than implemen

Chapter 5 Introducing SDN Control in MPLS Networks - High ...

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?