Edge servers ease network congestion

-November 06, 2017

Many upcoming networked applications demand massive bandwidths and real-time communication in small form-factor edge servers with dedicated interfaces. COM Express Type 7 server-on-module boards are appropriate platforms for designing such dedicated micro servers for the edges.

Public and private network operators need to provide an appropriate infrastructure for 1 GbE (Gigabit Ethernet) enabled devices. As more and more devices get connected, they need to eliminate oversubscription ratios in 1 GbE switched networks. A 10 GbE network is consequently the next logical step, as existing CAT 6/7-class infrastructures can be re-used. But it is not only the oversubscription that calls for more network bandwidth. There are also many high-performance applications demanding increased speed. Application areas include but are not limited to:

  • Access edges to broadcasting infrastructures
  • Service provider datacenters for video and audio streaming as well as SaaS
  • Local carrier-grade infrastructures for the mobile edge
  • Metropolitan and larger private networks
  • Cloud and edge servers on enterprise level
  • Storage attached networks (SANs) for Big Data storage
  • Intelligent switching technologies and smart NAS devices
  • Fog servers in Industry 4.0 applications
  • Edge nodes for wireless smart sensor networks
  • Collaborative deep learning computers


Different real-time demands

Most of these applications have not only massive bandwidth demands, but also demand real-time communication capabilities – a video stream for example. Tolerable latency here ranges from 6 to 20 ms [1]. Similar latency demands occur everywhere in networked applications because nobody wants to wait for the system once an application’s button has been clicked.

Consider autonomous vehicles used in intralogistics applications, or cyber-virtual factories where many hard real-time PLCs need to be synchronized, and where collaborative deep learning robots need to have situational awareness by ultrasound and video streams, and act without any delay. Here, the latency limits get even tougher, and overstepping a limit is critical as it can lead to unacceptable system failures or hazardous situations. So one can see that there are different real-time demands in networked applications, which is why any server technology has to deal with real-time capabilities.


Balancing the networks and servers

Such real-time capabilities can be achieved with load balancing of the communication channel. If there is a need for hard real-time, the channel also needs to be deterministic. The same applies to the applications servers. They can be balanced by using virtual machines, sharing their resources as needed. Such virtual machines can even consist of networked servers or networked resources such as NAS or SAN storage, and for hard real-time, all this can be configured in a deterministic manner as well.

By installing such a networked system, not only are high quality real-time services achieved. The entire system becomes more economical by running many applications on virtual machines. This ensures that server inventory gets consolidated and that no processing power is wasted. That’s why more and more virtualized server technologies are deployed in harsh environments, from carrier-grade network edge servers on rooftops, down to industrial systems, robotics, and machine control.


Managing the design challenge of customization

Exactly for these heterogeneous systems, the PCI Industrial Computer Manufacturers Group (PICMG) launched the new COM Express Type 7 server-on-module specification to help engineers overcome the design challenge of building dedicated server technologies at viable price points by utilizing commercial, off-the-shelf components. Servers-on-module are application-ready components that offer engineers design efficiency, as they only need to design an application-specific carrier board.

Furthermore, designers can scale the performance not only within one processor family but across all relevant processors from all vendors. The standardization of the module interfaces to the user’s carrier board delivers high design lifetime and upgrade opportunity.

COM Express Type 7 server-on-modules that support 10 GbE bring extremely compact, scalable real-time server performance to control cabinets, redundant switched small cells, and cloudlets at telecom edges.

Time-sensitive networking support

Since real-time support is key for these server designs, server-on-modules also support a software-definable pin for each of the 10 GbE interfaces. This physical pin can be configured as an input or output and is driven by the corresponding Ethernet controller. A typical application is the implementation of a hardware-based IEEE 1588 timing protocol for high-performance real-time applications to realize 802.1 timing and synchronization of distributed real-time systems. Possible applications include converged networks with real-time audio/video streaming, and real-time control streams which are used in automotive or industrial control facilities.

By implementing 802.1 compliant time-sensitive networking, designers can ensure that all devices have a common understanding of time, and that they use the same rules in processing and forwarding communication packets, selecting communication paths, and in reserving bandwidth and time slots – possibly utilizing more than one simultaneous path to achieve fault-tolerant failover modes [2]. The latter demand is the major reason why COM Express Type 7 modules can support up to four 10 GbE ports natively, as horizontal process-level communication needs at least two pairs of channels to build a cable-saving line or ring architecture. The same demand applies to endpoint servers utilizing a switched storage network, or executing, for example, Big Data or deep learning algorithms.


An example COM Express Type 7 module and carrier board combination with 4 Gigabit Ethernet ports to build fault-tolerant topologies in accordance with the 802.1 time-sensitive networking (TSN) protocol. This board supports up to a 16-core Xeon, 48 GB of DDR4 ECC RAM, two 10 GbE ports, and 32 PCIe lanes (source: congatec).


Hypervisor implementation

For application developers who also want to virtualize their server platforms, application-ready bundles can help engineers get their solutions to market faster. COM Express Type 7 module/carrier devkits support hypervisors – from Real-Time Systems for example – which is particularly well-suited to industrial robotics, control systems, and medical applications. Such a quick-starter set that also includes remote management and maintenance simplifies the evaluation of the servers-on-module.



[1] https://en.wikipedia.org/wiki/Real-time_computing

[2] https://en.wikipedia.org/wiki/Time-Sensitive_Networking

Related articles:

Dan Demers is the Director of Sales & Marketing – Americas, at congatec, Inc., and has over 19 years of experience with embedded computing in the industrial, medical, and communications markets.





Loading comments...

Write a Comment

To comment please Log In