After publishing a series of marketing articles, I suggest the reader to dig a little deeper and find out how to achieve the highest performance to date, and is not lockable switching on all ports devices Brocade. Code.org is a great source of information. Actually no secret here. The architecture of devices based on the good old mathematics Charles Kloss, who has long been used in telephone Switch layer where the lock to put it mildly discouraged, and then was taken into service for the construction of all the older lines Brocade, including BigIron, NetIron mlx and xmr. So. NetIron mlx.

The device was released in 2006, tested by time and a lot of installations, with a capacity of factories 7,68 Tbps, routing 7.6 billion pps, and supported to date 256 is not blocked ports of 10 GbE. Architecture of the device described in detail in the relevant document on the website vendor. We also look at the scheme and discuss the most interesting facts: Switch modules are used by factory and the ideology of the device are the same for all devices range from mlx 4-slot chassis with up to 32 slot. No lock on the level of line modules and a Switch-factory for all types of traffic and packet sizes involved constantly and entirely up to the last module, ensuring maximum performance with redundant N +1 Control Plane is separated from the Data Plane. The decision to send the package was adopted locally on the cpu line card that serves a particular port. The diagram shows the main elements of the line card and switching the factory. np – Network Processor tm – Traffic Manager fe – Fabric Element So NP-network processor is responsible for handling traffic at the port and having a capacity of 20/40 Gbps sends the incoming packet to the traffic manager (TM) which is connected together with all elements of all factory-installed Switch factories. tm received a package from the network processor generates from it cells of equal size, and sends them at once to all elements of Switch-factory (FE).

Package re going to outgoing tm and sent to the output port in its service processor. Thus, we achieve the stated bandwidth of the device without blocking and delays. But it’s good for L2, you say, but at the third level? How crazy is reached the figure of billions of packets per second? And everything is quite simple. News main functionality is implemented in hardware, to ASIC-s local maps.