Next-generation networking needs more high-end real estate

3Next-generation networks are evolving rapidly, with a number of emerging technological trends creating new challenges for operators, service providers, and in particular, network equipment manufacturers. The move towards more software-driven networks using Network Intelligence (NI) technologies running on inexpensive commodity computer platforms promises to create a more flexible, efficient network infrastructure to deliver new services, greater security, and reduced costs. At the same time, workload consolidation strategies and technologies such as Software-Defined Networking (SDN), Network Functions Virtualization (NFV), and Deep Packet Inspection (DPI) are all changing the shape of the next generation of network system architectures. The goal of generating deeper levels of intelligence across all types of network infrastructure is aimed at helping service providers improve traffic management and optimization, policy enforcement, billing, Quality of Service (QoS), and security; it is almost a foregone conclusion that this newfound visibility into data will create further opportunity for new and powerful revenue-generating applications.

The challenge for architects and platform designers, particularly those working on high-end systems, is two-fold. The transition to an intelligent, software-driven network requires an increase in processor capacity and the capabilities needed to meet the latency requirements of the communications network. When coupled with the forecasted 300 percent broadband increase in subscriber demands over the next five years, performance and I/O capacity demands will increase exponentially. Even with the communications acceleration and offload functions built into newer processors, it is clear that the expected performance and cost enhancements of commercial multicore processors alone will not be sufficient to deliver the required performance of high-end equipment handling millions of simultaneous subscribers.

To support these next-generation networks, what architects need is a platform that is tailored for high-end communications applications where significant flexibility and application-specific optimization is frequently required. Such a platform must support a wide variety of processing architectures and accommodate both off-the-shelf and custom-developed acceleration options that are able to provide Original Equipment Manufacturer (OEM) differentiation. To meet growing demand, systems that are capable of throughput measured by several Terabit per second (Tbps) switching capacity along with extensive I/O connectivity will be mandatory. Of course, in today’s world, the platform must be based on open standards that allow designers to use off-the-shelf building blocks supplied by a dynamic ecosystem. And, when building such a system, it must be easily managed as a single entity and not as a chassis with a random collection of blades.

The Advanced Telecom Computing Architecture (ATCA) used in communications platforms today meets many of these requirements.

ATCA status and direction

ATCA has delivered on its promise developed over a decade ago. The now-mature ATCA ecosystem provides the telecommunications and enterprise networking industries with a flexible, scalable, manageable, and highly reliable platform to meet the needs of the increasingly software-driven communications environment. ATCA has a proven track record, and the extensive ecosystem has shown that they are able to easily incorporate the latest processor, storage, and I/O technology. Collectively, the PCI Industrial Computer Manufacturers Group (PICMG), the embedded vendors, and the user community have shown how the adoption of this technology has vastly improved time-to-market for new applications, exemplified by its wide deployment, particularly in worldwide 4G networks.

While current ATCA-based systems can lay claim to being a high-density, scalable I/O solution, it is still not enough to keep up with the forecasted increase in performance required by next-generation networks using NI technologies. ATCA specifications are evolving to support increased power requirements, higher transport speeds, increasingly sophisticated management, and extended versions (of the base specification) for higher performance networking and datacenter applications.

This last point – extended versions – is particularly crucial, as ATCA is faced with a processor and I/O density challenge. More simply put, ATCA has a real estate challenge. There simply isn’t enough room for the advanced I/O, acceleration, and offload capabilities found in the latest component technologies. To enable support for software-driven networks requires a dramatic improvement in the processing and I/O density of ATCA platforms.

PICMG has groups working to evolve the ATCA specification to meet the needs of the next 10 years. For example, the proposed PICMG 3.7 specification addresses the real estate issue mentioned previously with several base extensions being defined to provide more capabilities. The 3.7 specification provides for double-wide boards, two-sided shelves, and new board types, including Extended Transition Modules (ETMs). Currently, the PICMG 3.7 specification is in progress.

For those that want to “push the envelope” to deliver solutions now, there are a number of alternate options: proprietary technology, general-purpose Intel-based “appliance” platforms, and ATCA-based solutions with managed extensions and customization capabilities. The pros and cons of proprietary and Intel commodity approaches are well known. The remainder of this article focuses on how to extend existing ATCA solutions.

extended ATCA (eATCA)

extended ATCA (eATCA) is a system architecture based on ATCA that delivers increased I/O and packet processing performance for networking platforms facilitating borderless enterprise, secure datacenter, and cloud computing applications (Figure 1). eATCA systems integrate standard ATCA blades with extended Rear Transition Modules (eRTMs), providing almost four times more real estate for PCIe-based I/O connectivity and enabling new capacity for acceleration and offload .

Figure 1: Pictured here are front and rear views of a two-slot eATCA system. eATCA allows enterprise and datacom customers to manage ATCA like a single entity, not a collection of blades, while specialized SMM-5060 Shelf Management Modules facilitate customization.

The system architecture responds to an increasing demand from network equipment providers for greater I/O connectivity and a simpler way to add special-purpose acceleration hardware beyond that currently offered. The eATCA architecture is specifically targeted at OEMs deploying solutions in large enterprise network security systems, or in datacenter and service provider environments where the highest available performance, enhanced I/O, and application-specific optimizations plus time-to-market are uncompromising factors of choice.

The ability to scale the new systems to over 1 Tbps of throughput also hits the performance sweet spot targeted by customers deploying solutions at the high end of today’s network computing and packet processing spectrum. By using Commercial Off-The-Shelf (COTS) ATCA blades with enhanced I/O, and with the big switching capabilities of ATCA, equipment providers can utilize technology from a wider supplier ecosystem for a future-proof architecture that can truly scale. Moreover, customers can gain distinct product differentiation through the ability to add their own hardware innovation more easily on an ETM.

Utilizing the new real estate

To support the many possible types of application-specific customization required for high-end platforms, the eATCA architecture is based on Advantech’s Customized COTS (C2OTS) technology. C2OTS is a semi-custom, system-level approach to implement high-end communications platforms that allow for OEM differentiation, but also offers the distinct advantages of COTS building block re-use. C2OTS is based on standard products that have been designed with later customization in mind to enable distinct features and custom IP integration without the risk of expensive, time-consuming, full-custom Original Design Manufacturer (ODM) designs.

C2OTS platforms, including Advantech’s standard ATCA blades and eRTMs, support several approaches for adding acceleration, offload, and I/O options:

  • Standard Advanced Mezzanine Cards (AMCs)
  • Network Mezzanine Cards (NMCs), which are similar in form factor to AMCs but cost optimized for use in Advantech network servers and appliances
  • Fabric Mezzanine Modules (FMMs), which use an on-board fabric mezzanine interface to support acceleration, offload, and I/O options

With a wide range of AMC, NMC, and FMM options currently available, OEMs are able to quickly tailor a platform for particular applications (Figure 2). Fully custom solutions can also be implemented by designing custom AMCs, NMCs, FMMs, and eRTMs with vendor IP built right in at the hardware level.

Figure 2: The Advantech MIC-5333 is a quad Network Mezzanine Card (NMC) carrier that allows customers to run I/O ports directly into the blades rather than through the switches. The NMC can be reused on eATCA, and allows co-processing and offload to take place next to the compute blades. Note that the Zone 1 power connector can supply up to 200 W to the eRTM.

An example would be the increase in demand for viewing and controlling networks based on “flows,” a unidirectional sequence of packets all sharing a set of common packet-header values. SDN and OpenFlow represent trends that require flow capabilities, and being able to easily build a specialized flow processor into a custom eRTM enables a differentiated platform for a wide variety of cybersecurity applications.

A further example of the flexibility built into this solution is how Advantech NMC and FMM options utilize Intel’s QuickAssist Technology. As the complexity of networking and security applications continues to grow, network platforms need more capable and complex processing resources. This is especially true for workloads such as cryptography, data compression, and pattern matching. Intel QuickAssist Technology is designed to optimize the use and deployment of algorithm accelerators in these kinds of applications, and also helps manage them as resources to simplify accelerator implementation across platforms and technologies. Using Intel QuickAssist Technology makes it easier for developers to integrate customization options into their solutions and to migrate to future technologies.

Real estate-enabled innovation

Software is defining and transforming the network, but more software requires more processing power. The preferred design strategy for implementing systems is shifting from a “do-it-yourself ASICS”-based approach to one that utilizes standard commercial processors with both on-chip and external co-processors to provide essential performance and I/O improvements.

To meet the needs of high-end network solutions providers, Advantech has designed the eATCA architecture with the necessary real estate to accommodate a wide range of acceleration, offload, and I/O options, allowing OEMs to rapidly build tailored solutions for their customers.

Paul Stevens is Telecom Sector Marketing Directorfor Advantech.

Advantech Co., Ltd.