Satisfies the requirements that IT folks don't understand
How low is the latency for media packets? Typically 5 to 10 µs per hop in the first implementation, which uses 1Gb/s links.
How do you achieve that? By allocating each media flow specific points in the byte stream on each link. In a switch, the incoming byte stream on each port is written to a central buffer, where it is available for about 15µs before being overwritten. The routing table for the outgoing flow shows where the packet should go in the byte stream and from where in the buffer to copy it.
What about traffic such as file transfers? Best-effort traffic is carried in all the bytes that aren't used for media packets; it pauses when it is time for a media packet, so the media packet does not have to wait until the end of the best-effort packet and no fragmentation headers are needed. The service experienced by Ethernet packets and IP datagrams is similar to current networks, and there is less overhead per packet.
So, is it the same as IEEE 802.3br Interspersed Express Traffic? It's similar in some ways, but the overheads are much less, especially for small packets.
Is special hardware needed? The situation here is similar to AVB. Switches that implement Flexilink are needed to achieve the near-analogue latency, which would not be possible with a standard Ethernet MAC layer. They use standard physical layers such as Gigabit Ethernet, and also implement the legacy MAC layer and switching, which is used if the link partner does not support Flexilink.
So is it restricted to the local area, like AVB? No, it is fully routable and suitable for use throughout the Internet. It fixes most of the problems that are being experienced with the current Internet, including issues such as address translation, mobility, and security. Allocations are per-flow, not per-class, so can't be affected by other senders "babbling". Large routers will be much less complex (and therefore cheaper and less power-hungry) than large IP routers.
How are the media flows set up? There is an out-of-band signalling protocol which allows flows to be set up that pass across different networking technologies; only the edge devices (where it moves from one technology to another) need to implement it. Addressing is very flexible and is independent of the underlying switching technology; domain names can be used directly instead of requiring end-systems to implement DNS look-up. When a flow is set up, the Quality of Service provided by the networks over which it passes is reported, allowing appropriate jitter buffers etc to be allocated if required.
Is it a standard? Forwarding plane formats are specified in ETSI GS NGP 013. A signalling protocol is specified in IEC 62379-5-2, and management is specified in other Parts of IEC 62379. The architecture is described in ISO/IEC TR 29181-3:2013 and is now being standardised in ISO 21558. More information on standardisation is available here.
------------- ooo OOO ooo -------------
Copyright ©2013-2018 Nine Tiles