Redirecting thousands of users from a primary datacenter to a backup represents a huge shift in terms of the WAN traffic flow. Traditional QoS mechanisms, such as MPLS Class of Service, simply cannot deal with these sudden changes automatically. Manual intervention is required to reconfigure the QoS parameters to match the new situation.
Datacentre replication is also pushing the network to the limit. The secondary datacentre must keep an up to date version of all application content and states. To do so, high speed links are used between the geographically distant datacentres known as fat pipes. These fat pipes can suffer from extreme Bandwidth Delay Product problems where the full bandwidth cannot be used due to TCP limitations. Over these fat pipes, backup applications (EMC, Network Appliance, Legato, Veritas, Tivoli, Microsoft, CommVault, DoubleTake, CA …) or virtual machine transfers (VMware, Microsoft, etc.) are used to perform the data replication. They all rely on the TCP protocol, which is not optimised for fat pipes and most of them do not fully optimise their bandwidth requirements with application level compression. As a result, datacentre replications are slow and business continuity is jeopardized.
Active-Active DR creates “meshed flows”. Branch offices can receive flows from both datacenters at the same time. Following Dual use DR principles, critical applications are handled by one datacenter and less critical application by the other. In such deployments, non critical applications can easily use all the available resources at the branch and thus jeopardize the performance of critical applications. With tactical solutions, competition between flows coming from both datacenters can only be handled by a device at the branch, necessitating a device at every single branch in the network. In most cases, the costs involved with such a massive hardware deployment are clearly incompatible with the benefits of using Active-Active DR.