There was so much as the Bus and Board show that I had to split it into two parts. In addition to covering the other companies I met, with I needed to finish up some of the things I left dangling in the first part of this series.
One of the main items that I only mentioned in part one is Aurora, something I have not written about in depth yet. It is also a technology that many outside of the FPGA industry—especially non Xilinx users—do not know much about. The big change is that it is now a backplane standard.
What is Aurora? It is a point-to-point data-streaming system that uses the same SERDES technology as other high-speed serial fabrics like PCI Express, Serial RapidIO, Gigabit Ethernet, and InfiniBand. This collection can utilize the same fabric backplane because the main difference between these are protocols. All the serial fabrics impose a higher level protocol that includes everything from virtual channels to in-band management.
Aurora is similar but much more simplistic. Where an endpoint for the other fabrics can typically communicate with any other endpoint in the fabric, Aurora is really designed to connect a specific pair of end-points. This is not surprising as Aurora was originally designed for linking a pair of FPGAs.
Aurora operates at speeds in excess of 622 Mbits/s per channel. It can operate in full duplex or simplex mode. It has flow-control support and handles any size frame. Yes, the protocol is packet based, but it does not include the routing and management overhead that other fabrics have. This, in theory, makes it more efficient and it has the potential for higher throughput. But in practice this difference is not a good reason to choose Aurora over something like RapidIO.
Instead, Aurora’s simple system configuration and lower complexity tends to be its big selling points. Aurora can be routed by a hard backplane providing a high-speed link between boards. This can be done using a full mesh backplane or using rear transition modules (RTM), but more likely an Aurora switch would be used. The term switch is probably a misnomer from my standpoint since it is more logically a cross-bar switch for connecting board A to board B. The switch does operate like other fabric switches, but the routing is fixed. This is similar to what Advanced Switching Interconnect (ASI) does when tunneling PCI Express to the latter, it connects a limited subset of ASI nodes. However, PCI Express is a more complicated protocol than Aurora.
Aurora tends to be implemented in FPGAs. This is not surprising for Bus and Board vendor attendees that most likely have FPGAs on their boards anyway. This typically allows the FPGA to be configured for any switch fabric, including Aurora. It is just that the FPGA usage is lower for Aurora. Still, migration to fixed-purpose chips or embedded support for fabrics like Serial RapidIO is increasing. Eventually, dedicated support such as a Serial RapidIO interface in a DSP will be the rule instead of the anomaly.
It was interesting to hear the discussion regarding Aurora. Not surprisingly, those that had support for Aurora were psyched about it and related announcements. Those that did not were pushing their alternatives.
The bottom line is that Aurora provides a very simply way to extend the communication between devices across multiple boards. It has its uses and overlap between Aurora and any other fabric is likely to be more limited than one might think. Aurora will find its own spot just as the other fabrics are settling down into particular niches, some rather large.
Small Is In One other item I didn’t bring up in the first part of the Bus and Board article is the 3U form factor. It has always been a standard and it has found a home in an ever increasing number of spots. Still, interest in 3U boards seems to be higher than I expected, especially in rugged environments.Shrinking designs to 3U seems to be driven by a number of factors. Processors and FPGAs with lower power consumption than their predecessors seem to be a big factor. It is now possible to pack in significant compute power without resorting to a 6U form factor. Likewise, more rotating and flash storage is showing up in a 3U form factor. The side effects are important as well, given the reduced size and cooling requirements. It will be interesting to see what the impact of MicroTCA will have on this area next year. Final approval of MicroTCA should occur this year, so products will be at the Bus and Board show next year. However, there’s no need to wait to evaluate the form factor because prototypes and AMC cards are already available.
It was an interesting show and it is a harbinger of a great year. Next year’s show should be even more interesting.
For more information about the Bus and Board show, see the following:
Mellanox Pushes The Limit With InfiniBand
Micro Memory’s CoSine Was A Great Sign
Artesyn Delivers Pre-Integrated Blade Systems
Pentek Records Wideband Signals
TekMicro and QinetiQ Take a Note
www.busandboard.com