Packet-based blade solutions are redefining the landscape for embedded modular computing. The new paradigm emphasizes the network over the control bus.

The embedded OEM board "blade" and system community is in the midst of a number of transitions. There are major changes in not only technology innovation and adaptation, but also corresponding standardization of the way that modular building blocks are put together. The shift that is underway is to a packet-based midplane/backplane infrastructure that readily supports converged networking and computing applications for the new generation of communication equipment being built for edge and gateway services.

The shift also includes much more thought and planning for higher-powered solutions and more easily managed building blocks. The internal block diagram for a contemporary embedded system today looks more like an IT LAN schematic than it does the more traditional microprocessor "control bus" schematic of days past. Figure 1a and 1b compare these old and new approaches to embedded computing.

Standards: A Pragmatic Blend

Successful standards are pragmatic blends of innovation and adaptation of existing technologies in such a way as to provide high value, cost effectively. VME was successful in the 80s because it married the Motorola 68K bus interface to a backplane standard. CompactPCI® was successful in the 90s because of the adoption of the PCI bus, built into most commercial silicon for the desktop market as a passive backplane implantation.

The landscape is littered, however, with attempts to standardize "over engineered" solutions that while technically superb are not supported by basic widespread availability of the fundamental technology that defines the solution. Futurebus was a perfect example of a "great idea" that could not turn the corner to widespread use because of a lack on building block availability. Similarly, attempts to "extend" old architectures by layering in new protocols on top of old protocols provide the illusion of forward compatibility. The reality is that the new technology often is only supported by one or two companies that enjoy a brief marketing advantage in providing this "renaissance" but also fall prey to lack of pervasive building block availability at the silicon and tool level.

With all this in mind, It's important to realize that the market defines the parameters for a successful standard, and rarely the other way around. The best a standards-based body can do typically is to collect together existing technologies and sprinkle in a bit of innovative "mortar" to offer a compelling architecture for widespread adoption. The focus should be on ragmatism, not revolution.

Pressure from Moore's Law

Moore 's Law for semiconductor improvement constantly applies pressure to system builders to provide a balanced solution set. There are three fundamental elements that are fundamentally always "at siege" by Moore 's law:

• Power In - The power it takes to feed ever denser solutions
• Balanced I/O and memory interconnection with the next generation of silicon
• Heat Out - Removing the heat efficiently

This slow steady pressure that Moore 's Law applies to the industry eventually causes "revolution," which in fact are nothing more than reactions to the overall balance of system architecture getting off center. The transitions that this causes are opportunities for new approaches to supplant old, so the industry evolves.

Interconnects: Down to Two

Embedded suppliers of modular building blocks have traditionally provided bus-based solutions that were adaptation, either directly or indirectly, of microprocessor interconnects -- sort of an inside-to-out architectural approach. Standardization in the form of VME and CompactPCI formalized this approach in a way that allowed for an open merchant market of board and system solutions from multiple vendors. This made sense in the 1980's and 1990's -- but not for the 2000 and beyond. Most new embedded applications, whether standardized or not, are employing more of an "outside-to-in" architecture that embed external features such as LAN technology and system management inside the box. This has led to the standardization of Ethernet on the backplane in the form of architectures, for example, PICMG 2.16.

For any standard architecture to be successful, it must draw upon technology that will be readily available to designers and be cheap or even "free" to the designer. In terms of serial interconnects, there are only two players that will be ubiquitous for the foreseeable future and they are Ethernet and PCI Express. There have been hundreds of attempts to create "the perfect serial interconnect" with some of these actually seeing some level of adoption.

Ethernet is King

It is Ethernet that is the king of the current serial interconnects and it will be PCI Express and PCI Express/Advanced Switching that will be the next "free" technology that will be widely adopted in the embedded industry. While Ethernet and PCI Express are very different in their designed intent, they share on trait in that they are (Ethernet) or will be (PCI Express): everywhere. All others will fill niche roles and may even be technically better than either Ethernet or PCI Express, but the simple availability and cost of these two interconnects will more than offset the potential technical advantage that any other solution will purport to have. Ethernet has been used to solve many problems that it was never designed for simple because it was available and "good enough" for the task. And so it will continue to be.

Originally conceived as a LAN technology, Ethernet won that battle for dominance in the 80's, and has become embedded at every level of integration from simple microcontrollers to Wide Area Networking (WAN) and Ethernet-to-the-home initiatives. It only makes sense that Ethernet is a popular interconnect for blade standards and will continue to be so "forever" -- it simple works and is free to blade designers because it is so widely supported in silicon solutions. PICMG 2.16 is an excellent example of this adoption and the AdvancedTCA specification defines Ethernet as the Base Interconnect for inter-blade communication for this very reason.