|FROM THE DESK OF JOHN THOMSON
Ethernet & The Military
which in turn are connected via PCI Express, with Serial RapidIO for node-to-node transfers within a back-end multiprocessing system. The results could then be relayed via Ethernet to a host computer for visualization (Figure 1). That kind of highly typical application architecture is probably what Freescale had in mind when integrating all three fabrics into the 8640 and 8640D processors.
Figure 1: The strengths of individual serial switched fabrics can be leveraged within a single high performance system
Although Ethernet technology emerged in the early 1980s, the initial cable segments were difficult and expensive to interconnect. The first repeaters and bridges made creating large networks a challenge. With the advent of “Cheapernet” (10Base5), and hubs to wire up twisted pair cabling, things simplified. It took until nearly 1990 for the first Ethernet switch to appear. This made wiring Ethernet networks simple and efficient. Now, given the proliferation of Ethernet-based networks, it is unsurprising that the range of available Ethernet switches has also expanded hugely. Ethernet switches can roughly be grouped two ways: which ‘layer’ do they work at, and how configurable (or ‘managed’) are they?
In practice, there are three classifications: Unmanaged Layer 2, Managed Layer 2, or Managed Layer 3. “Layer” here refers to the once-famous OSI Reference Model - but it’s simplest to say that Layer 2 means that it makes decisions based on the Ethernet (or Layer 2) address. So: a Layer 2 switch knows about Ethernet and nothing above it. An unmanaged Layer 2 switch is “plug-n-play” – it will handle anything that is Ethernet based – but there may be some things that it can’t do as efficiently, since it’s very general. Moving to a managed Layer 2 switch allows configurability – it is possible to do things like set up VLANs (Virtual LANs), or make the switch aware of multicast. This could be the key to running the network at the right level of efficiency for a particular application.
The next class is managed Layer 3 switches. These still know about Ethernet – but also about the Internet Protocol that commonly sits above Ethernet (Layer 3 = IP). A managed Layer 3 switch will normally do everything that a managed Layer 2 switch will, and a lot more: since it is sensitive to IP addresses, it can even do some router-like functions.
Managed switches (at Layer 2 or Layer 3) are highly desirable in the kind of mission-critical situations that characterize military applications: for example, not only can they provide the opportunity to design redundant or fail-safe solutions, but they can also provide automatic alerts to network managers in the event of failure.
There is one particular example that is common in military use, and illustrates the need for a managed switch. The situation is where there is a lot of sensor-type data, being “multicast” (sent to many recipients.) On an unmanaged switch, the switch will send multicast traffic out on all ports, meaning that the overall traffic on the network is higher than it needs to be. A managed switch, supporting a feature known as ‘IGMP snooping’, will be aware of which recipients are interested in listening to this sensor data, and will limit that traffic to only those ports. By controlling the overall traffic rates in this way, a managed switch can make the difference between a workable and an unworkable application.
With increasing requirements of access control and security in networks, and the push towards the new version of the Internet Protocol (IPv6), Ethernet switches in military applications - including backplane switches - are becoming steadily more sophisticated. As the need for speed drives the signalling technologies higher, the steady migration from 100Mbps to Gigabit is changing gear to 10Gig.
GE Fanuc has perhaps the broadest range of Ethernet switches available today in 3U and 6U form factors, for the VME, CompactPCI, VXS and VPX architectures, layer 2 and layer 3, IPv4 and IPv6, managed and unmanaged, 100Mbps to 10Gbps, with front or rear port routing, for benign and hostile environments. To find out more, visit us here.
At a time when systems designers are beginning to leverage the enormous performance potential of Serial RapidIO – especially in the military market, where the VME-centric, serial switched fabric orientation of the VPX architecture is enabling the development of new generations of high performance solutions – it’s easy to forget about Ethernet and, especially, Gigabit Ethernet. Although primarily used ‘outside the box’, Gigabit Ethernet should be considered a serial switched fabric like Serial RapidIO or PCI Express, suitable for use on backplane interconnects.
Ethernet's detractors have been quick to point out its shortcomings. One common criticism was its non-deterministic nature because of access to the original shared cable. Much criticism was also directed at the higher-layer protocols often associated with Ethernet – the Internet Protocol (IP) and protocols which make use of it (TCP, UDP and so on). These are criticized as being inefficient, using CPU resources, or being unreliable –the “send and pray” nature of UDP, for example.
But most of these shortcomings have been addressed or are easily worked around. Full-duplex, switched networks remove the low-level non-determinism, and packet prioritization allows predictable delivery rates. Most Gigabit Ethernet devices offer hardware checksum and segmentation offload support, similar to that carried out in interconnects such as Serial RapidIO. These significantly reduce the processing overhead on the host CPU.
These are just two examples of Ethernet’s potential disadvantages, and how they have been addressed. The question is: why would anyone bother?
The fact is that Ethernet has some very significant advantages – advantages that are very attractive, especially to military users. The first of these is that Ethernet is all but ubiquitous: it is the most common computer networking architecture in the world. Hundreds of millions of commercial Ethernet ports are sold every year – and virtually all the traffic on the Internet terminates with an Ethernet connection. Implicit in this widespread usage is the huge ecosystem of COTS Ethernet hardware and software products, Ethernet services and Ethernet expertise – not to mention the promise of virtually unlimited interoperability.
It is probably that widespread use that has led to Ethernet becoming relatively inexpensive – another facet that is very attractive to military customers. Beyond that, Ethernet has been around a long time: the standard was first published in 1980 – making it slightly older as an industry standard than the venerable VMEbus, another favorite of military systems developers. And, just like VMEbus today, Ethernet provides backward compatibility, allowing for easier upgrade and integration of legacy systems. A Gigabit Ethernet network interface card can communicate seamlessly with previous slower versions of Ethernet.
Beyond this, Ethernet is widely agreed to be low-cost to implement, simple to install and maintain, and robust. It is also ‘future proof’. In the same way as it is backward compatible, so yesterday’s 100Mbps Ethernet is being replaced by today’s Gigabit and then upgraded with tomorrow’s 10 Gigabit, or soon-to-be 100 Gigabit, as networks carry bigger data loads, higher definition graphics, or whatever more demanding application is around the corner.
So: despite its perceived disadvantages, Ethernet is in widespread use throughout the military – in applications as diverse as unmanned vehicles, navy vessels and a wide range of aircraft and land vehicles. That’s no surprise, given the focus on network-centric warfare – and the network that underpins that strategy is an IP-based network.
It is also important to bear in mind that the various serial switched fabrics currently achieving a high profile are not mutually exclusive. Each has its strengths and weaknesses, and the best systems designs leverage those strengths. It’s possible, for instance, to envisage a high-performance system in which sensors are connected to high-speed ADCs,