A multipoint VPLS network can make an entire WAN look like an Ethernet switch. Some providers build this type of network in a way that their is no discrimination between different types of packets. Basically, the provider guarantees the specified carrying capacity of the network and police it to this rate at their edge. This can cause challenges for those with strict QoS implementation and packet loss budget. In order to best understand this problem, take a look at the diagram below.
At first glance, it might look as though a hierarchical priority queue configuration could solve this problem. The problem is the customer routers don’t really know what the other sites are sending at a given point in time. In the above example, the site at the bottom can send up to the contract rate of 10Mb/s. The site on the right, could also be sending 20Mb/s. If both of these are sending at full capacity toward the site on the left, it would cause an approximate drop rate of 33%.
This challenge forces a network administrator to either accept that a network is best effort or to determine how to lessen the likelihood of dropping high priority packets. In that case, an administrator would create hierarchical queuing policies. However, the QoS policies would need to be written in a way that the sum of all potential traffic going to any site doesn’t exceed the contract rate for that site.
The advantage of this is that the administrator has control over all of the bottlenecks and should be able to reorder packets effectively. The drawback is that it is no longer possible to use the full capacity of the network. If the network traffic is fully meshed, this would create reserve a lot of otherwise usable capacity. When traffic follows a hub and spoke pattern, this is at least a little more predictable and is a little more efficient.
There are QoS challenges with multipoint networks that aren’t configured to discriminate between different types of packets. When oversubscription occurs, packets will be dropped. Since the provider is indiscriminate, important high priority packets could be dropped. Without very aggressive application of hierarchal priority queuing, unintended drops will likely occur. Thinking through the process of what a queueing configuration may look like for a given network may help administrators understand the challenges and come up with a reasonable solution.