With the growth in end-user needs for high bandwidth due to Netflix, WhatsApp, YouTube, Instagram, etc, there is a demand for much-sophisticated network management and monitoring of the traffic. Service providers want to give customers better user experience by proactively identifying problems in the network, making the network secure and by extracting metadata for third party applications.
In a typical network architecture, there are multiple 10 thousand eNodeB connected to multiple data centres with up to a hundred compute nodes which handle the traffic.
The challenges for monitoring include:
” There is no load balanced, load balancing!”
An efficient load balancing (LB) can only be done with one powerful entity based on a deterministic identifier. Any solution based on additional meta-information will fail because a meta-information had to be computed and there is no error-free algorithm. The problem with Software-based LB is limited performance. The limit of a CPU solution is much lower than an ASIC approach.
Example: If traffic of 800 Gbit has to be load balanced to various Probes, but the CPU supports only 200 Gbit, a load balancer that correlates signalling and user traffic is needed upfront. This could not work and it does not scale at all.
Let’s look at the user plane traffic flow.
The only deterministic identifier for the user plane traffic is the user IP address. Signalling plane does not have user IP.
The issue with S1U interface is, that the transport network parameters are changing when the user moves from a base station to base station. This is the reason why the common approach to load balance on the TEID (tunnel endpoint ID) will not work. Because this ID is changing when the customer moves. And to follow this changes would need a lot effort and will never be 100 % because of ciphered traffic of the signalling in an LTE network. The only technical and commercial useful solution is to load balance on the inner IP address of the customer because this address is not changing during a session.
Cubro has a solution for this issue: We can load balance on inner IP with our Next Generation Network Packet Brokers EXA32100 and EXA48600 which are easy to configure via Graphical User Interface.
Cubro’s next-generation network packet broker can perform 3G/4G GTP load balancing of user plane up to 3.2 Tbit/sec per unit (19 inch 1U). It is possible to perform full line speed filtering to separate user and signalling. The advanced network packet broker is capable of performing user session aware load balancing, i.e, specific user session always arrives at the same Probe interface.
See this video to learn more about LTE GTP Load Balancing