Backplane-based architectures have advanced and continue to advance to meet market needs. We’ll take a look at the evolution of many of the VITA and PICMG-based architectures, see where they fit today, and analyze some upcoming technologies.
Origins of Eurocard and VME
The advent of the microprocessor created the foundation for the microcomputer bus industry. The VERSAbus by Motorola in 1979 based on the 68000 microprocessor formed the electrical basis for the new VMEbus (Versa Module Eurocard) along with the preexisting Eurocard mechanics. Eurocard is a term that collectively represents the products based on the DIN 41612 and IEC 60603-2 connector standards, the IEEE 1101 printed circuit boards standards and the DIN 41494 and IEC 60297-2 rack standards . The VMEbus was created in 1981 as a result of the joint efforts of Motorola, Philips/Signetics and Mostek. Subsequently it was adopted as a standard by IEEE (IEEE 1014-1987) and IEC (IEC 60821). At the time, the VMEbus was a 16 bit, easily upgradeable, microprocessor independent, non-proprietary bus standard with a proven mechanical form factor. Any vendor could adopt interoperable and compatible products.
In the years since, VME has proven to be a de-facto standard and the pioneering joint development efforts has been repeatedly and successfully adopted to drive new emerging standards and specifications to this day. The unprecedented success of VMEbus architecture was in no small part due to the robust Eurocard form factor and its ability to retain backwards compatibility while advancing the technology.
As technology progressed with faster boards and chips, the necessity to upgrade from 16 bit, 40 Mbytes/sec to 32- bit (3U) and 64-bit (6U) with twice the bandwidth (80 Mbytes/sec) caused the VMEbus specification to be revised through IEEE 1014-1987 and ANSI/VITA 1-1994. The latter is also referred to as VME64. This improvement along with an automatic ‘Plug n Play” feature, Auto Slot ID, and a host of others served to enhance the popularity of VMEbus in embedded applications.
VME is very popular particularly in mil/aero and industrial designs due to its consistency/long life span, backward compatibility, multitude of vendors, and rugged form factor in an open specification. However, bandwidth and I/O constraints started to become a problem for some applications where more performance was required. So, ANSI/VITA1.1-1997, also called VMEbus Extensions or VME64x, was the answer. This revision added new capabilities including a new 160 pin connector, a 95 pin P0/J0 connector for additional I/O, 3.3V power plane, 160Mbytes/sec bandwidth, more 5V power, rear plug-in units, and EMC front panels with injector-ejector handles and geographical addressing. VME64/64X has also proven to be an unqualified success in ATR (Air transport Rack) type enclosures with conduction cooling per IEEE 1101.2.
By 1997/1998, Elma Bustronic Corp in cooperation with Arizona Digital released another improvement to the speed of the VMEbus – the VME320. VME320 has a data transfer rate of 320Mbytes/sec using a ‘Star’ interconnection scheme. Actually, a highly similar concept for switched serial fabric backplanes is used today.
Legacy VME and VME64x will continue to have its place in the market. For some applications, the bandwidth, I/O, and reliability of VME/VME64x are more than enough. Even 32-bit J1 VME backplanes are still designed-in on new applications to this day. Later, we’ll look at how switched fabrics are changing the game. First, we’ll discuss VME’s nemesis, CompactPCI.
CompactPCI
At about the same time as the VME320, a new emerging bus based architecture called CompactPCI was born. This was based on the 2mm HM connector standard and the IEEE 1101.10/11 mechanicals with PCI being the core electrical portion. PCIbus was a proven bus with an enormous installed base in different market segments like telecom, industrial automation, and several others. The CompactPCI (cPCI) architecture could also leverage off the lower cost, widely available PCI silicon , WinTel (Windows/ Intel) architecture and rapidly improving performance. A 64-bit implementation could boast a data transfer rate of 533Mbytes/sec. cPCI was created by a group of PCI manufacturers under PICMG. PICMG released a series of specifications in the late nineties that addressed critical requirements in the telecommunications industry. This included features like hot swap and five nines availability (99.999) and the high pin count of the 2mm HM connector.
PICMG 2.1 (Hotswap), PICMG 2.5 (H110 Telephony bus), PICMG 2.7 (Dual system slot) and PICMG 2.9 (System management bus) were some of the specifications that made using cPCI bus a compelling reason for new products catering to computer telephony, VoIP and a wealth other applications. With a soaring demand for more speed and bandwidth fueled by the Internet boom, cPCI became extremely popular with its open standard that helped speed the “time to market” for new entrants in the telecom arena. The Eurocard form factors of 6U by 160mm for front line cards and 6U x 80mm for rear transition cards was tailor made for platform providers to develop rackmount equipment that met the NEBS criteria for 300mm depth as well as cable management. However, as faster processors emerged over time, it was clear that CompactPCI would need changes to compete in higher-end applications.
Backwards Compatible Switched Fabrics – PICMG 2.16/ 2.17/2.20
Eventually, the slot limitations of CompactPCI and the bottlenecks to higher data transfer rates prompted the foray into switched fabric architectures like PICMG 2.16 (cPSB), 2.17 (StarFabric) and 2.20 (Serial Mesh). CompactPCI is limited to 8 slots at 33 Mhz and 5 slots at 66 Mhz. CPSB (Compact Packet Switching Backplane) was the first and the most successful. Since these were all based on the PICMG core 2.0 cPCI specification, the form factor remained 6U x 160mm (Eurocard). The specification took undefined portions of a standard cPCI backplane and assigned pinouts for the fabric. Switch slots, which run the fabrics, were new. Otherwise, the backplane slots had the same form factor and accepted legacy cPCI cards.
This allowed backwards compatibility and preservation of investment while leveraging their higher performance levels for high-end applications. Not only is the bandwidth vastly increased, but switch-fabric interconnection is also preferable for high availability applications due to their self-healing features and options for redundancy. The cPCI bus on the backplanes are still limited to 8 loads (at 33 Mhz); not all slots need to implement the cPCI bus on a PICMG 2.16 or 2.17 backplanes. Also, low-profile bridges allow more cPCI slots without interfering with rear I/O. However, the bridges still occupy a load in each segment. The performance of PICMG 2.16 is approximately up to 830 Mbps slot-to-slot, where standard cPCI accommodated approximately 266 Mbytes/sec (64-bit @ 33 Mhz). Shortly after this timeframe, VITA specifications started to go through similar changes.
Backwards-compatible Switched Fabrics -- VITA
Borrowing from the backwards-compatible advancement of PICMG 2.16/CompactPCI, VME has now integrated switched fabrics into the technology too. GigE over VME adds switched fabric performance to the VME64x backplane. The specification takes the previously undefined portion of the backplane in the P0 section and defines the pin-outs for a Gigabit Ethernet fabric routing. The new feature of the backplanes is switch card slots that drive the fabric. Otherwise, the backplanes are fully backwards compatible. This provides the ability to have data traffic at high speeds in one “plane of the backplane and have the control functions of VME in another “plane”.
However, as semiconductor technology advanced and new applications and capabilities arose, it became apparent that the 2mm HM connector in cPCI and the P0 section of VME64x had its limitations. Thus, the
VXS (VME Switched Serial) specification uses a new high-speed connector. VXS is basically the same concept as GigE over VME, except it uses the MultiGig RT-2 connector. Where the 2mm HM started to see performance problems above approximately 1.3 Gbps, the MultiGig claims to perform above 6 Gbps. VXS also is fabric-agnostic – fabrics like InfiniBand, Serial RapidIO, Gigabit Ethernet, and so on, can be implemented. The performance of VXS is approximately up to 3000 Mbps slot-to-slot.
Although, the new P0 connector in VXS is not backwards-compatible, many VME64x cards do not use the P0 connector and the backplane can always be designed to have legacy VME64x slots (see Diagram 1). In 2005/2006, Elma Bustronic created a new technology called VXS Processor Mesh (VITA 41.7), which will be discussed further below. It adds a mesh fabric to VXS. The bandwidth is approximately up to 5000 Mbps slot-to-slot.
Non Backwards-compatible fabrics – PICMG (AdvancedTCA, MicroTCA)
In late 2001, PICMG formed a committee to develop a new series of specifications aimed at the next generation of Telecom requirements – PICMG 3.0 Advanced Telecom Computing Architecture or AdvancedTCA (ATCA) calls out for an 8U x 280mm front boards and 6U x 60mm rear boards. Thus the Eurocard mechanics continues to exert its influence as the form factor of choice in emerging technologies.
ATCA was not designed to be backwards compatible with cPCI, it is a whole new form factor – geared towards central office applications. AdvancedTCA has the following features aimed specifically at the central office:
- 8U x 280mm boards, with wider 1.2” spacing
- 48V DC as the sole power delivered to boards (most CO systems use that voltage exclusively)
- Serial interconnects for bandwidth
- Flexible user I/O
- Mandatory shelf management based on IPMI
- High level of service reliability (five nines or more) from integrating features for redundancy, serviceability, and manageability
- Low cost sheet metal solution (in high volumes)
Using the ZD connector rated at 5 Gbps and with Dual Star , Mesh, or Replicated Mesh topologies, ATCA has a huge bandwidth improvement over CompactPCI. The performance is approximately 5,000 Mbps slot-to-slot. However, ATCA is overkill in several applications. Therefore, MicroTCA was created featuring many of the design concepts as ATCA in a smaller simpler platform.
MicroTCA was developed as a smaller, cheaper architecture based on the ATCA concept. It reduces size and cost by eliminating the ATCA carrier, enabling AdvancedMC modules to be plugged directly into the backplane. MicroTCA provides scaleable bandwidth from 1-40 Gbit/sec, and scaleable availability ranging from three nines (99.9%) to five nines uptime (99.999%). The MicroTCA Carrier Hub (MCH) provides interconnect, power conversion, clock distribution, and system management functionality needed to support up to 12 AMC modules. The high performance of MicroTCA coupled with its compact size, low cost and low power consumption make it a good fit for wireless base stations, digital loop carriers, optical ADMs, enterprise networking, storage servers, non-rugged Mil/Aero communications, and many other applications. Chassis come in three formats – cube (usually 7U high where multiple cubes can be loaded horizontally in a rack), pico (the cards are loaded horizontally, saving vertical rack space), and shelf (various heights, often 19” size). Like ATCA, the performance is approximately 5,000 Mbps slot-to-slot. To offer advanced performance, but maintain some compatibility with legacy VITA and PICMG standards, some new hybrid technologies have also been developed.
New Hybrids
Hybrids are what I call “partially” backwards compatible. One could use the legacy technology, but requires a bit more changes than the fully backwards compatible technologies. One of the “hybrids” is CompactPCI Express.
In 2004 CompactPCI Express (cPCIe) began to be developed with the expected dominance of PCI Express in the marketplace. Similar in concept to VME’s VXS, cPCIe uses a new connector to overcome performance limitations of 2.16/2.17 A new ZD connector rated at 5 Gbps and using the PCI Express fabric, the performance of cPCIe is vastly increased while maintaining some backwards compatibility. The support of legacy 32 or 64 bit CompactPCI boards is accomplished by a PCIe-to-PCI bridge. The use of CompactPCI boards of 33MHz, 66MHz, or 133MHz is possible. CompactPCI Express utilizes a serial point to point bus with a read-only bandwidth up to 16x 2.5 Gigabits/second or 8x 2.5 Gigabits/second full duplex bandwidth. Providing support for several different card form factors with connectivity in 1x, 2x 4x, and 8x increments, each link represents one full duplex 2.5 Gigabit/second interconnect path. Because the CompactPCI Express architecture continues to support the P3, P4 and P5 in all 6U slot types, CompactPCI Express can continue to support for all existing CompactPCI secondary architectures such as PICMG 2.5, 2.20, 2.16, 2.17 and 2.18. They can be used as either functions on native CompactPCI Express cards or as legacy cards in the original CompactPCI format.
VPX (VITA 46/48) is another hybrid where some backwards compatibility is possible. The technology is highly customizable and most backplane designs will be custom as the routing is highly undefined. This allows a lot of flexibility, but a less straightforward, follow step A to step B approach. VPX comes in 3U and 6U versions and the chassis can be convection cooled (forced air), conduction cooled, or liquid cooled. V48.3 is the most aggressive version with special modules for liquid flow-through cooling. On the backplane, the key difference is it offers a mesh configuration and the VMEbus is optional. The aggregate bandwidth slot-to-slot is approximately 5,000 Mbps.
VXS Processor Mesh in simplest terms is a mesh VXS backplane. Processor Mesh allows backwards compatibility to VXS and legacy VME64x slots and has a control plane, and central I/O slot. Many ask what the difference is between Processor Mesh and VPX (VITA 46). Processor Mesh is more defined and backwards compatible and has higher performance slot-to-slot (approximately 7,500 Mbps). It also has defined areas like the central I/O and control plane. VPX comes in a 3U version, which can be very attractive for applications where space or weight restrictions are a critical element.
Other new hybrids include Industrial PCI Express and Serial Mesh. Industrial PCI Express (PICMG 1.3) is a specification for next generation PICMG 1.0 backplanes. Serial Mesh (PICMG 2.20) used the ZD connector in the P4 section of a CompactPCI backplane. It offered higher performance for TDM traffic and supported multiple simultaneous transport protocols like ATM, IP, Frame Relay, GPRS Tunneling Protocol, etc. New hybrids have used new connector technologies and other modifications to expand performance, but have maintained a degree of compatibility to cPCI or VME, preserving the investment in software and hardware.
A Question of Backwards Compatibility
One of the key elements in Eurocard is the question of backwards compatibility. The industry has developed specifications that have tremendous performance with maintaining a partial or full backwards-compatible premise. Backplane architectures like legacy VME and CompactPCI are still going strong. The Hybrids will also help preserve the investment in these technologies and keep them an important part of the industry product mix. The completely new specifications like AdvancedTCA offer their own set of advantages along with high bandwidth. But as you’ll notice, the backwards-compatible specs are catching up in performance. Will it be too late for architectures like CompactPCI Express, which got going a little late? We will have too wait and see.
