Though chassis and backplane vendors have announced support for 100Gb/s over the backplane (100GE), it does not appear that this capability will be required in the immediate future. In fact, it might be overkill for the time being.
Just recently, Intel announced the general availability of the Fortville, a 40GE (and not nearly a 100GE) NIC, while other chipset vendors are not even able to deliver Fortville’s level of throughput. Furthermore, even if NICs are able to offer full 40GE backplane ports, this does not assure us that applications can actually handle that speed. Most cannot.
We can make a pretty good intuitive guess that by the time 100GE throughput is actually needed for backplane support, the industry will already be changing its spots. By then, it will adopt a new paradigm to replace top-of-rack architecture: compute nodes at the network edge will host applications dispersed all over the network and there will be no requirement for direct connectivity to a central processing location or to a certain chassis that will need to process tens of terabits per second.
So, when we speak today about the necessity for 100GE on an ATCA switch, we are really referring to 100GE uplinks rather than the fabric interface. The traffic can be directly re-routed back to the network again without accessing the node blades. Traffic can be filtered or load-balanced per flow and redirected to the relevant node blade. Therefore, total bandwidth will not approach the per-port 100GE level.
Telco Systems’ T-ATCA510 100GE switch blade meets the challenges of 100GE uplinks. It also delivers smart engines for flow control and inspection to assure optimized traffic utilization. Implementing this switch provides solutions for today’s requirements while paving a growth path into the future.