SRv6 uSID - Scale (Pt. 3) (UPA Walkthrough)
Last updated
Last updated
Although using a PCE for all end-to-end paths allows you to remove the leaking of loopbacks and locator prefixes into the L1 area, it is actually still not all that scalable. You will have a SRv6-TE policy per-PE per-color. If you are mostly using ODN policies that use the IGP metric, you will have many many policies that simply take the IGP shortest path.
The most scalable solution, in my opinion, is to use summarization of both the locator prefixes and the loopbacks. In fact, the loopback can be within the locator prefix, so the locator prefix summary alone can cover the loopback and resolve the BGP nexthops.
The problem then becomes: how do you implement BGP PIC? If a PE withdraws its loopback route in a remote L1 domain, the L1/L2 router hides this via the summarization. So BGP PIC cannot be triggered.
To handle this, the local L1/L2 router can leak the /128 or /48 prefix into the L2 with a max metric. This is called UPA (unreachable prefix announcement). This signals that the specific route is not available anymore. (Remember that the maximum possible metric value in ISIS indicates an unreachable route). To enable this, you use the following keyword on the L1/L2 router:
The router will leak the specific /48 or /128 alongside the summary prefix with an unreachable metric value. This route will only exist for a pre-determined amount of time. This is controlled using the following config:
You can also use the unreachable-component-tag keyword with the summary-prefix. This limits UPAs to components of the prefix that have that specific tag.
The ingress PE must have the following configuration in order to react to the ureachable prefix announcement. Any nodes along the path do not need this command, such as another L1/L2 router for a remote domain. This command notifies the RIB of the unreachable route, which triggers BGP PIC on the ingress PE.
Note that these commands were released in version 7.8.1.
