There are four primary traffic patterns in a backplane environment: equal access systems, primary access systems, multiple access systems, and centralized access systems. Each of these differs in complexity, aggregate throughput, and latency.

Point-to-point is often favored for high-speed design, but the switch from multi-point architecture to point-to-point architectures is a radical change with many system implications.

The terms parallel and serial are often used to describe the two major types of backplane implementations. Better terms for these would be multi-point and point-to-point (or single-point), respectively. Multi-point describes a backplane in which a signal has more than two connections, with each connection often able to transmit or receive signals. This feature is also called shared media or bus. Point-to-point describes a backplane in which a signal has only two connections — a transmit and a receive connection. In either case, these signals can be uni-directional, bi-directional, or full duplex.

Another feature that distinguishes multi-point backplanes from point-to-point backplanes is the number of separate channels typically available. In a multi-point backplane, the normal configuration is a single channel. This configuration is due to the width of the channel (number of signals) required to provide high-bandwidth data through the backplane in a multi-point environment.

High bandwidth is an overriding concern because one set of connections provides the single communication path to all cards. Due to the nature of the multi-point backplane with many connectors and loads on each signal, the frequency of the backplane is limited. In some cases, this channel may be duplicated to provide redundancy, but this is not the normal approach. By its very nature, a point-to-point backplane requires that multiple channels be implemented. The advantage of multiple channels is that they extend the backplane's potential for very high aggregate bandwidth. The single channel bandwidth is, though, by its nature, a fraction of the aggregate bandwidth.

Consider the complexity of connecting each card to each other card directly. This would be a mesh network connection. Also, there are many systems implementations that may need only limited communication links between cards.

So, how do you get from one card to every other card in the system when you have only a point-to-point connection? In such a system, a crosspoint switch device is often used. A crosspoint switch is a device that has multiple ports in which any input port can be connected to any output port or, in some cases, can be connected to any number of output ports (for multi-casting). Each connection within the crosspoint switch is completely independent of all other connections within the switch.

This feature is known as a non-blocking switch. Even though the switch is non-blocking, multiple streams of traffic to a given output will cause blocking at the output and require arbitration between streams. Just as a multi-point backplane overcomes blocking by providing bandwidth that suits the requirements of all streams of information, so does a point-to-point switched backplane.

To implement a point-to-point switched backplane, each point-to-point connection from a card is connected to a switch device, often on a switch card. When the primary operation of the card is other than a backplane function, this is refered to as a function card. These are cards that perform some system function, such as line cards, memory cards, I/O cards, or CPU cards.

Such cards are distinguished from crosspoint switch cards. Usually the crosspoint switch device is on a separate card, not directly on the backplane. Each function card will have its channel(s) connected to a switch card, as will every other function card. More than one independent channel on each card is certainly an option.

From the other perspective, a switch card will have connections for each channel of each function card. There can also be more than one switch card per backplane. For example, two are needed to provide redundancy, load sharing, or both.

In the case of redundancy, it is possible to have the same signal from the function card sent to both switch cards. More than likely, there would be separate drivers such as 1-to-2 fanout drivers that connect to each switch. If load sharing is required, then two completely separate channels are needed on each function card. Using a 2 x 2 crosspoint device for these connections on the function card would provide optimum flexibility. The load sharing implementation could also provide for half-performance redundancy as well.

The point-to-point switched backplane is the same as the star topology seen in many network designs — the network switch is replaced with the backplane switch card and the nodes of the network are replaced with the function cards of the backplane.

A star topology, traditionally used in networks, is where a group of nodes (computers) are connected to a central point for all communications links. The central point that connects all of the nodes is a switch system.There are two types of switch control that can be used in a switched backplane — out-of-band and in-band. The difference is whether the switch is controlled by information that is passed through the same port as the data (in-band) or through a separate, centralized port dedicated to the control of the switch (out-of-band). An example of in-band control would be Ethernet, in which header information, such as the destination address in sent along with the data in a packet format.

Another set of backplane architectures that is finding a level of interest is the mesh backplane. Since there is no switch, the complexity of handling these multiple point-to-point signals now resides on the function card. Something needs to be available on either the transmitting or receiving function card that identifies the data. On the transmitting function card, the data would need to be directed to the receiving function card that requires the data. Or, if the transmitting function card does not discriminate, then the receiving function card must identify the data that it requires.

Point-to-point differences
A point-to-point card interconnect solution is good when specific cards, but not all cards, need to communicate and when multiple streams of simultaneous communication are helpful. A point-to-point switched card interconnect solution is suitable when all cards need to communicate, when multiple simultaneous channels are valuable, and when the required aggregate bandwidth exceeds that which can be attained with a multi-point backplane.

Keep in mind that each point-to-point link in either of the point-to-point solutions described will have bandwidth that is often much lower than can be attained with a multi-point configuration. However, the total bandwidth potential can be many times the total bandwidth capability of a multi-point solution. Once again, this is appropriate when many communications links to multiple points are required. Of course, the bandwidth limitations of the point-to-point solutions can be easily overcome by adding more signals to each path. This is an example of where a point-to-point solution could be parallel.

Not long ago, the multi-point backplane was the only backplane architecture available. Access to any card from any card can be accomplished with a multi-point backplane. With the introduction of GTLP, the performance limits on multi-point backplanes have been pushed well beyond what was available a few short years ago. Now, multi-point backplanes can perform well above 100 MHz with 8 or more cards, and others have been developed with 40 cards and 20 MHz performance. In certain systems, it is even possible to push the performance up to 200 MHz, even for high card counts.

An interesting characteristic of the point-to-point switched backplane is that it can easily be used for both primary-access and the equal-access systems. By adding a single high-speed port to the switch cards of an equal-access system, primary-access traffic can be accommodated. So, in the case of network equipment, a single system can be used for both the switch and the router function.

However, switching from multi-point architecture to the point-to-point architecture is a radical change, although it is a straightforward change from most network designs to take the point-to-point architecture of the network and embed it in the system. In almost every case of transition from the multi-point architecture to the point-to-point architecture, there are significant changes in all aspects of the implementation. Address, data, and control are no longer separate, isolated entities but, most likely, packets. Transceivers are replaced with serializers and deserializers. Transactions are replaced with packets, frames or cells containing address, error and link information. Master/slave configurations and bus control signals are replaced with in-line arbitration. High-speed signals are replaced with RF signals and all of the expected implications of extremely high-speed signals. Clock signals are replaced with embedded clocks and PLLs or DLLs. Noise and skew concerns are replaced with loss, jitter and latency concerns. Connectors begin looking like walls to the electrical signal. Memory buffers are required. Cut-through and adaptive cut-through techniques need to be understood. Switch-to-switch communication needs to be understood.

The concept of point-to-point switched backplanes is an outgrowth of the networking market. It was a very important breakthrough in the networking architectures when it became possible to replace the bussed network with switching systems, since the collisions on a bussed network can limit performance. Due to this history and the need to reduce wiring, clocking became something that was embedded in the data using rather sophisticated techniques to ensure that transitions of the signal happen often enough to allow a Phase Lock Loop circuit to continue to generate an accurate clock. And, this still makes sense when it is traces that are being considered rather than wires.