Herramientas para la conectividad IPv6 con múltiples proveedores

Abstract

La presente Tesis propone una arquitectura para la provisión de una solución de multihoming escalable y de la exploración de distintos enfoques. La solución actualmente disponible en IPv4 para el soporte de multihoming basada en la inyección de rutas de sitio en el sistema global de rutas impone una carga que crece linealmente con el número de sitios multihomed, lo que limita su escalabilidad y las posibilidades de crecimiento. La presente Tesis plantea una solución alternativa que garantice la escalabilidad del sistema global de rutas basada en el uso de direcciones agregables por proveedor (PA). En una configuración basada en direcciones PA, un sitio multihomed obtiene tantos bloques de dirección como proveedores tiene lo que plantea las dificultades con los filtros de ingreso, con el iniciar una nueva comunicaciones después de un fallo y la preservación de comunicaciones ante la ocurrencia de un fallo. La presente Tesis Doctoral plantea una arquitectura para solventar estos problemas basada en el encaminamiento basado en dirección origen para proveer compatibilidad con los filtros de ingreso, y una nueva capa de identificación dentro de la capa IP para brindar las capacidades requeridas de tolerancia de fallos. ________________________________________________

In this Thesis we propose an architecture for the provision of scalable IPv6 multihoming support. In the multihoming solution currently deployed in the IPv4 Internet, the multihomed site announces a route to its address blocks through all the providers using BGP. The result is that multiple routes towards the multihomed site are available in the inter-domain routing system. While this solution provides the fault tolerance and path selection features required to a multihoming solution, it presents limited scalability, since each multihomed site contributes with at least one routing table entry in the already oversized inter-domain routing tables. Because the support of the multihoming solution currently deployed in the IPv4 Internet is becoming challenging even for the current number of multihomed sites, this approach is deemed unsuitable for the expected number of multihomed sites in the future IPv6 Internet, especially when considering that the wide adoption of low-budget broadband access technologies such as ADSL or CATV will enable multihoming in SOHO environments. As a consequence, an alternative multihoming solution for IPv6 is needed. The requirements imposed to the new solution essentially include all the benefits provided by the incumbent solution, i.e. fault tolerance and traffic engineering capabilities, and also an enhanced scalability with respect to the number of multi-homed sites and other relevant Internet parameters. In order to preserve routing system scalability, aggressive route aggregation can be achieved through provider-based aggregation, precluding the injection of routes associated with individual multi-homed end-sites. When Provider Aggregatable (hereafter PA) addressing is used, multi-homed sites obtain one prefix per each one of their providers. Consequently, as each provider will only announce its own prefix to the rest of the Internet, a given provider will be used to reach the multihomed site only when the destination addresses used belong to the prefix associated with the provider. So, in order to be reachable through all of the providers of the site, each host within the multihomed site will have to configure multiple addresses, one per provider. Even if this setup guarantees the scalability of the multihoming solution, such multi-addressed configuration is not without difficulties of its own when attempting to provide the additional features mentioned above. In particular, this configuration presents the following problems: - Incompatibility with ingress filtering techniques: The incompatibility is caused by the lack of coordination between the IPv6 source address selection mechanism, performed by the host, and the path selection mechanism, performed by the intra-site routing system. As long as outgoing packets are routed through the provider that has delegated the prefix contained in the source address, packets will flow freely; but when those packets are routed through a different ISP, they will be discarded by the ingress filtering mechanism x due to source address incompatibility. It must be noted that because of this issue, packets may be discarded even in a scenario without failures. - Difficulties when establishing new communications after an outage. The difficulties arise because not all of the addresses available for a multihomed host are reachable, so in order to be able to communicate, hosts need to properly discard unreachable addresses and select those addresses that are reacahable. Current address selection mechanisms are unable to cope with such situation. - Difficulties when preserving established communications. In order to preserve established communications through outages, the endpoints of the communication have to adapt the addresses used during the lifetime of the communication according to the available providers. Moreover, this address replacement has to be performed in a transparent fashion with respect to transport and application layers, in order to actually preserve the established communication. Current applications and transport layers, such as TCP and UDP, identify the endpoints of a communication through the IP addresses of the nodes involved, implying that the IP addresses selected at the communication establishment time must remain invariant through the lifetime of the communication. But as it has been presented earlier, once that an outage has occurred in one of the available ISPs, the associated address becomes unreachable, so an alternative address has to be used in order to convey packets to the multi-homed host. These two constraints impose that after an outage, packets must carry a different address, corresponding to an available ISP, but they have to be presented to transport and application layers as if they contained the original address, in order to be recognized as belonging to the established communication. Such approach requires additional mechanisms in both ends of the communication in order to preserve a coherent mapping between the IP addresses presented to the transport and application layers and those addresses actually contained in the packets. - Difficulties when providing traffic engineering capabilities. The usage of multiple prefixes pre multihomed site imply that those traffic engineering techniques will no longer apply, and alternative mechanisms that provide equivalent capabilities are required. In this Thesis we describe an architecture for the provision of multihoming in IPv6 that deals with all the aforementioned concerns. The proposed IPv6 multihoming architecture introduces the following components: - An intra-site routing paradigm that takes into account the source address, so that source hosts can determine through the selection of the source address, the exit path of the packets. Such feature provides ingress filtering compatibility. - An address selection mechanism that takes into account address reachability information acquired through a trial and error procedure. - A new Multihoming Sub-Layer within the IP layer that will perform the required mapping between the addresses that are presented to the upper layer protocols and the addresses that are actually used for exchanging packets in the network. Such layer allows the usage of different addresses for exchanging packets during the lifetime of a communication, while keeping unchanged the address presented to the upper layers, preserving the established communication. - A mechanism for the configuration of the policy table defined in the default address selection procedure, for the provision of traffic engineering capabilities. A detailed presentation of the aforementioned mechanisms is preceded by an exhaustive analysis of the solution space that justifies the selected approach.