print

The Sky is the Limit

Emerson Network Power describes how to extend ATCA into military applications.

Military programs such as Command, Control, Communications, Computers, Intelligence, Surveillance, and Reconnaissance (C4ISR) call for commercial off-the-shelf (COTS)-based equipment that can be deployed into diverse application spaces (including laser and RADAR remote sensing, simulator and trainers, LAN/WAN network analysis, client server applications, telephony/wireless base stations, SATCOM, video and image, VoIP processing, targeting and tactical intelligent ground terminals).

Some equipment required to support C4ISR must be ruggedized, because it is located in the area of the conflict. Other parts of the infrastructure have more moderate environmental specifications that make it easier to use COTS equipment that may be more cost-effective. Behind the scenes, appropriate communications infrastructure is necessary in order to handle data flows, retrieving, processing and storing data, or maintaining communication via IP networks. This infrastructure requires reliable equipment, including gateways, applications servers and storage equipment. Typically, this equipment resides in a controlled environment that provides a narrow ambient temperature range. This makes it easier to cool equipment and makes it possible to better exploit the performance range provided by the market.

Up to this point, the C4ISR infrastructure resembles that of a traditional data center. However, the application range demands support for highly reliable equipment that can help save lives during critical missions. Consequently, computer systems must host redundant servers, disk drives or power supplies. In addition, it must be possible to remove or install parts of the equipment without interrupting mission critical services.

Equipment based on the AdvancedTCA® (ATCA) open standard is well suited to meet these challenges. It provides high reliability, redundancy and hot swap capability, and is remotely maintainable. In addition, server infrastructure equipment based on the ATCA architecture can be easily scaled for performance and capabilities to meet diverse application requirements.

What Makes a Server a Server?

Performance and capacity requirements vary across the desired application range, and for a specific application they may change over time. As a result, server equipment needs to provide enough flexibility to accommodate the needs of the different application spaces in electronic warfare.

This includes:

  • Scalable compute performance from moderate to outstanding, depending on performance requirements and cost targets
  • SMP capability and use of multiple cores per CPU
  • Virtualization in both CPUs and I/O for better exploitation of compute resources and to provide additional security levels
  • Support for energy efficiency
  • Scalable main memory capacity from 10s of GB to 100s of GB
  • Scalable mass storage capacity from 100s of GB to 1,000s of GB
  • Direct attached storage (DAS)
  • Multiple RAID levels to support both storage throughput and reliability requirements

Driving it to the Extreme

Scalability is possible with ATCA technology, as proven by the wide array of product solutions. However, some of the CPU derivatives offered by the silicon suppliers today require electrical, thermal and mechanical precautions that are difficult to meet with existing standards. One could certainly wait for the next technology upgrade that promises the increase in performance that is expected of every new processor generation. But improvements to the server infrastructure are also necessary in order to exploit the increased performance of the CPU derivatives. Changing some of the boundary conditions seems to be inevitable. To meet future performance expectations, the following three major improvements are required.

Sufficient real estate on blade and system level

In general, the most compelling CPU derivatives are clocked at higher frequencies that require higher core voltages. In general, these chips also have architectural improvements that make them perform much better. This can include larger caches or more cores. As a result, power dissipation is significantly higher than the more moderately specified derivatives. For these high performers, thermal design power of 100 W and more per CPU is common, requiring larger heat sinks to be designed-in. CPUs of this type also often have wider memory and I/O connectivity, requiring the use of larger CPU packages.

To address the need for larger main memory capacity, more – and likely taller – memory modules are required. It is also desirable to provide direct attached storage close to the server blade. Ideally, it should reside directly on the blade to keep infrastructure cost down. Increasing the available real estate on future ATCA blades or systems is necessary to overcome space constraints.

Improved power distribution to blades

Faster CPUs, larger memory arrays and wider system buses simply require more electrical power. To run high performance compute blades, sufficient power feeding is crucial. For instance, today’s ATCA blades that use the latest processor generations typically consume electrical power in the range of 200 W to 300 W. These numbers are expected to increase as future CPU generations are introduced. Limitations in a system’s power capability could prevent them from supporting future blade generations. The trend toward increased power on server blade designs cannot be denied. To be clear, current server designs do, and future designs will, support energy efficiency by reducing power whenever possible; however, electrical support for the maximum expected workload is required.

Better system cooling capabilities

What goes in must come out. This means that feeding higher electrical power into blades results in higher thermal power dissipated on the blades. Some electrical power is channeled away through backplane interconnect and cables, but typically the effect is negligible. Therefore, future blades need enhanced cooling to transport the expected additional heat away from the blades.

Most servers, including ATCA systems, are cooled by using fans to move air. One possible approach to improving airflow in future blade designs is to widen the blade slot-to-slot pitch. This provides more space for cooling air to pass hot components and dissipate their heat. In addition, designing in more powerful fans will improve systems’ cooling capabilities.

Last but not least, limiting the maximum ambient temperature in a controlled environment can reduce cooling requirements. While equipment designed for central offices in the telecom space requires operating temperatures up to 55 degrees Celsius, data center servers typically are operated in lower ambient temperatures. The American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE) recommends a controlled environment of 18 to 27 degrees Celsius. Restricting the ambient temperature allows the use of components with higher thermal power dissipation and lower maximum operating temperature; consequently, blades with better computational performance based on faster processors can be supported. In addition, reducing cooling requirements allows the use of mainstream processors intended for the general server market, which can help to reduce costs.

Shifting Boundaries

While the constraints outlined above do not constitute an immediate problem, future server processors and architectures will present more challenges when implementing the ATCA architecture, as it will require changing some of today’s boundaries. The current specification defines a maximum of 400 W per ATCA slot. To accommodate future ATCA-based server generations, shifting that boundary to a much higher level is advisable. Doing so requires reviewing and redefining requirements for power entry, backplanes and connectors, as well as blades. Backward compliance with existing blades and systems is certainly mandatory to support the widest possible reuse of existing product solutions offered today.

For the same reason, increasing server and system real estate needs to be carefully considered. One possible approach is to make use of two ATCA slots instead of one. This approach would fit in well with the existing standard and would allow mixing blades of single and double width in a single system.

As a result of widening the real estate, airflow across boards also could be improved. In addition, improvements in system cooling capabilities would help to further shift the exploitable performance range.

Applying equipment based on ATCA technology in applications in less demanding thermal environments also raises the question of which markets – other than telecommunications – are suitable for ATCA equipment deployment.

The need to extend system capabilities has been identified by the ATCA ecosystem. The PICMG® standards organization has initiated a new subcommittee working on extensions for the existing ATCA standard that addresses some of the considerations described in this article.

Some of the published goals of the initiative are:

  • Support for higher compute density
  • Cooling enhancements
  • Significantly increased electrical power
  • Enhancements to blade and system real estate
  • Optimizations for relieved thermal environments
  • Maintaining forward and backward compatibility with existing ATCA products

Summary

Military programs such as C4ISR that make use of COTS equipment should consider the open standards-based ATCA architecture when looking for a reliable and scalable bladed solution. Varying performance requirements can be addressed by scalable compute equipment in a wide range. Exploiting future compute performance has resulted in an industry-wide effort to extend the current definitions of the ATCA standard, just as future efforts will lead to increased performance capabilities.

headshot_engels

Christian Engels is a senior technical marketing manager for the Embedded Computing business of Emerson Network Power. In his current role, Engels focuses on AdvancedTCA®, server architectures and product definitions, in addition to inbound and outbound technical marketing. He also represents Emerson in several standards development organizations, including the Communications Platforms Trade Association (CP-TA) and the PCI Industrial Computer Manufacturers Group (PICMG®).

Share and Enjoy:
  • Digg
  • Sphinn
  • del.icio.us
  • Facebook
  • Mixx
  • Google
  • TwitThis