The figure below shows a schematic drawing of the typical connections within the AMS-IX MPLS/VPLS infrastructure. Two customer routers at the top of the diagram are connected through a photonic cross-connect (PXC) to a set of PE routers (PE-1-red, PE-1-blue) with 10GE & 100GE connections. At the bottom a customer router connects to a PE router (PE-3) with a 1GE connection directly.
Each PE router is connected to each of the core (P) routers with one or multiple 10GE & 100GE connections.
The P routers are located in two different physical locations (core-location-1 and core-location-2). There are two P routers per core location.
Figure 2 below shows the logical connections programmed on top of the physical connections shown in Figure 1 above.
The logical connections between the PE routers are implemented as label switched paths (LSPs). Between each pair of PE routers we define four LSPs, one for each core. The diagram shows the set of LSPs for each pair of PE routers in a different colour (for example, green for the LSPs between PE-1-blue and PE-3, orange for the LSPs between PE-1-red and PE-1-blue).
Traffic between each pair of PE routers is load-balanced over the four LSPs. When any of the physical components in an LSP fails, the LSP will move over to an alternative physical path as will be shown in one of the next paragraphs.
As noted before, 10GE customer connections are not terminated directly on a PE router, but on a photonic cross connect (PXC). This PXC connects the member router at the physical level (L1) to one pair of PE routers. We use this setup because the reliability of a PXC is orders of magnitude higher than that of a PE router. In case of problems on a PE router, this setup allows us to quickly (typically 50 to 100ms) fail over 10GE customer connections from one PE router to another. Moreover, when maintenance on a PE router is necessary (hardware replacement, software upgrades, etc.), customer connections on the affected PE router can be moved to the other PE router, thereby minimising service interruptions for the customers and maximising flexibility for the AMS-IX staff.
During "normal" operations, 10GE & 100GE customer connections at a colocation are distributed over the two 10/100GE PE routers. In Figure 2 this is indicated by a half blue and a half red PE router and blue and red connections going to the customer routers. This setup further reduces the impact in case of an issue or maintenance, since only half of the customer connections on a site will need to be moved from one PE router to the other.
Figure 3 below shows the extreme case when a complete core site is down and both P routers in that site become unusable
Each LSP between a pair of PE routers has a pre-defined backup path over one of the other P routes in the other co-location. Failover from an LSP to its backup path typically takes less than 50ms. The time depends on the trigger for the failover.
In Figure 3 above both core switches at co-location-2 are down. All LSPs going over the two P routers in co-location-2 are now re-routed over the P routers at co-location-1.
Figure 4 below shows the situation when one of a pair of 10/100GE access PE routers fails or is taken down for maintenance. The PXC for the 10/100GE connections in this location now connects all member routers to the other PE router. (As a side-effect, all traffic between the 10/100GE customers at this co-location is now locally switched.)