Using LSPs to squeeze the most out of bypass wavelengths
Jun 29, 2015
In Metro rings, typically most of the traffic goes from the nodes on the ring to a head end.
When the traffic on the ring is low (compared to the size of the links) it is obvious that the “most-efficient” solution is to build a ring of point-to-point connections between the nodes. When the traffic is high, it is equally obvious that the “most-efficient” solution is to connect every node directly to the head-end aggregation router(s) in a star or double-star "hub-and-spoke" formation.
Where it gets interesting is when you are some place in between and when there is at least some local traffic between the nodes on the ring, as well as the main traffic direct to the head-end(s). How do you mix point-to-point connections along the ring with direct connections to the head-end in the most efficient way?
Here you find that the routing protocol is a very important aspect of this problem. It turns out to be very difficult to use the capacity efficiently if you use an “open” protocol such as ISIS or OPSF to route the traffic. However, if we RSVP-TE and combine LSPs, routing with constrained shortest-paths (CSPF), with Juniper’s auto-bandwidth to configure their bandwidth, then we can squeeze out the capacity on the links in the most efficient way.
The following pictures show a simplified worked example where we open up a 4-node ring into a chain to make it easier to figure out the numbers. The last node, D, connects into another part of the network, E (not shown). Here we have some local traffic but most of the traffic goes to the external node connected into the gateway aggregation router at D. The "network" is physically in a chain with the WDM cards allowing wavelength bypass through the routers.
It is easy to calculate the bandwidth requirements when we have point-to-point links only and also easy we use LSP (CSPF) routing to use the capacity efficiently. Where we use an open protocol, such as ISIS or OPSF, the traffic must follow the shortest-path come what may so this is much harder to calculate. When there are equal-cost-paths (ECMP) involved, you may need to take off your shoes and socks to help with the counting, as the traffic is split recursively each time the paths branch!
So we can see a massive reduction in the capacity requirements if we mix direct bypass-links to the head-end with point-to-point connections along the ring (or chain), using MPLS-TE and LSP routing to steer the traffic efficiently. If we use an Open protocol such as ISIS or OSPF, then the improvements are less clear and it is very sensitive to the exact details of the routing metrics on the links. Of course the amount of gain you can get depend on the details and scale of the traffic patterns. However, LSP/CSPF routing always gives you the best improvement.