MPLS Compared to Other IP-over-ATM Schemes

MPLS Compared to Other IP-over-ATM Schemes

In ATM networks, MPLS allows ATM switches to directly support IP services, giving maximum efficiency compared to other approaches. Traditional IP-over-ATM connects routers over Permanent Virtual Circuits (PVC).

Cisco also supports an alternative IP-over-ATM scheme called Multiprotocol over ATM (MPOA), which uses the Next Hop Resolution Protocol (NHRP). Unlike MPLS, MPOA overlays IP-over-ATM rather than fully integrating them. Although they do not share many of the advantages of MPLS in the WAN, MPOA and NHRP are cost-effective technologies for interconnecting nearby emulated LANs (ELANs) at high speeds. MPOA and similar proprietary approaches carry IP traffic over Switched Virtual Circuits (SVC). Traditional IP over ATM, MPOA, and proprietary approaches all have similar disadvantages:

• It is difficult to offer some types of IP services on the networks. For example, IP Class of Service cannot be offered natively by traditional ATM switches, and must be offered by translation to quite different ATM Forum Quality of Service concepts.
• Where IP services are offered, they are difficult to administer. Two levels of routing must be administered: IP routing (via OSPF or EIGRP or similar) and PNNI or similar routing for ATM. MPOA requires additional administration. Service translations, for example IP Class of Service to ATM Quality of Service, also require administration.
• IP services can be quite inefficient over ATM networks. For example, IP Multicast over ATM networks is difficult to achieve on a large scale due to the interaction of multicast routing, multicast group membership processing and ATM VC maintenance.
• There can be scaling limitations and/or dangerous interactions between IP routing (OSPF, and so on) and the ATM network, leading to unstable networks. Traditional IP over ATM can lead to storms of IP routing updates and subsequent network meltdown, if more than 30 OSPF routers are connected in a full mesh over PVCs. MPOA is unsafe when connecting routers to each other, and is intended only to connect hosts to routers or hosts to hosts. (See below.)
• IP services require a substantial implementation and management effort. For example, an MPOA implementation requires PNNI, SVC signaling, ATM ARP, an ATM ARP server, NHRP, and a NHRP server, in addition to AAL5, IP routing (OSPF, and so on) and an IPv4 stack.

MPLS in ATM networks avoid all of these disadvantages.

Problems of Running IP Routing over An ATM Network without MPLS

If N number of routers are running OSPF and are connected in a full mesh over ATM PVCs, a single physical ATM link failure may result in ATM-layer rerouting of a large number of PVCs. If this takes too long, or if the ATM network cannot reroute PVCs at all, a large number of PVCs effectively fails.

The number of PVCs involved may be of the same order magnitude as N, and even N² in some cases. In any case, it is likely to be seen by O(N) routers, where "O(N)" means "a number proportional to N". So, a single ATM link failure will cause each of O(N) routers to send a link state advertisement (LSA) of size (at least) O(N) to (N-1) neighbors. Thus a single event in the ATM network results in O(N³) to O(N⁴) traffic.

When a router receives an LSA, it must immediately recalculate its routing table because it must not forward packets based on old routing information. The processor load caused by a storm of routing updates might cause the routers to drop or not send keep-alive packets, which appears to the neighboring routers as further link failures. These lead to further LSAs being sent, which perpetuates the problem.

The net result is that a full mesh network can go persistently unstable after a single network event.

This critical failure occurs because the routers do not see the state of the ATM links and switches directly. IS-IS has somewhat better performance than OSPF in full mesh conditions because IS-IS has more sophisticated flooding capabilities (these capabilities, specifically the ability to pace flooding and block flooding on some interfaces, are also becoming available on OSPF). However this does not address the underlying problem.

The solution is to enable IP routing to directly see the state of ATM links, which is what is done by ATM MPLS.

MPLS also addresses a different problem that arises when the ATM network runs PNNI routing: the basic conflict between routing protocols. PNNI routing at the ATM layer can make decisions that conflict with OSPF or similar routing at the IP layer. These conflicting decisions can lead to persistent loops. (See the NHRP Protocol Applicability Statement, RFC2333, for more on this. Further investigation on router-to-router NHRP at the IETF revealed that router-to-router NHRP was not practical.)

The only reliable solution to this problem is to use the same routing protocol at the IP layer and ATM layer. This is exactly what MPLS does in ATM networks.