End-to-end resource management in DiffServ Networks –DiffServ focuses on singal domain –Users want end-to-end services –No consensus at this time –Two proposals: Integrated Services over Differentiated Services Bandwidth Broker

Integrated Services over Differentiated Services –Apply integrated services model end to end across a network with one or more DiffServ domains. –DiffServ networks are treated as a link layer medium –Integrated Service requests are mapped to diffserv capability Services to PHB –Two tier resource allocation DiffServ distributes aggregated resources at core IntServ allocates the resources to indivudual users –Example: Figure 3.8 –Do we have end-to-end QoS guarantees in this model?

Bandwidth Broker (BB) –Resource allocation between different backbone service providers, as well as support QoS on local LAN. –Each network is associated with one or model BB, which keep track on local network usage Admission control, policy control, reservation aggregation –BB needs the knowledge of routing –Example, Figure 3.9 –Per flow end-to-end QoS guarantee?

Multiprotocol Label Switching Virtual circuit at IP level –A short, fixed length label is used for forwarding. –Similar too other circuit switched technology: ATM, Frame Relay, WDM Convergence of datagram and virtual circuit. –How virtual circuit/datagram works? –Fundamental trade-off: complexity at routing and signaling.

Motivation: IP over ATM integration. –Without MPLS: Overlay model (classical IP over ATM) –ATM is treated like another subnet technology. –When ATM is treated as a subnet, it is different from both point-to-point or shared media network: »multiple neighbors, but no broadcast capability (ARP) »ATM may contain multiple logical IP subnets (LIS) –ATM ARP: need to emulate ARP. –NextHop Resolution Protocol (NHRP) Problems: –Have scaling problems. Need VC to connect hosts in a logical subnet. O(N^2) connections and O(N) neighbors. –Going through IP layer in an ATM cloud is not efficient. –Lose QoS capability

Motivation: IP over ATM integration. –With MPLS: IP routing and signaling protocols take over the control path. ATM switches are used only to data transmission. ATM switches are now in the IP routing domain. NHRP and on demand SVCs are no longer needed MPLS matches ATM’s capability –Labels are SVC circuit IDs –Switching is done by ATM switches –QoS can be preserved in a LSP.

Some advantages for MPLS: –Simpler packet classification and table lookup Trade off between simple forwarding and complexity in signaling. –Protocol independent forwarding Even IP has many protocols. (CIDR, v4, v6, etc) –Finer forwarding granularity Can treat different customers differently. –Traffic Engineering MPLS can setup explicit route between two nodes. key application

IP packet forwarding (Figure 4.2): –Control path: making routing decisions. unicast routing protocols and multicast routing protocols –Data path: move packet from input ports to output ports. unicast forwarding table, unicast forwarding engine, multicast forwarding table, multicast forwarding engine Forwarding is done by longest match between dst address and entries in the forwarding table.

MPLS packet forwarding (Figure 4.4): –Control path: making routing decisions. unicast routing protocols and multicast routing protocols Other control protocols (ATM, FR) –Data path: move packet from input ports to output ports. Label-switching table Label-based forwarding engine

–IP centric control: IP is here to stay. IP control protocols are more mature and scalable than other alternative protocols. IP control protocols for everything (ATM, FR, optical WDM) –A unified control plane integrated IP with different link layer technologies and simplify the management.