Internet Minute Drives Packet Acceleration
Intel’s latest two-chip Crystal Forest platform adds data packet acceleration to existing telecommunication application and control plane functionality.By Steve Price, Intel
Staying true to its telecommunication workload roadmap, Intel has just announced the addition of its next generation communications platform, codename Crystal Forest, based on Intel® microarchitecture, codename Sandy Bridge. This addition will consolidate application, control and accelerated data control processing into a two chipset multimedia system.
|Figure 1: The Internet Minute demands greater bandwidth and faster deployment from telecommunication manufacturers, operators and service providers. (Courtesy of Intel)|
Shipments of Internet-based consumer electronic devices is expected to exceed those of traditional PC platforms for the first time in 2013, according to IHS iSuppli. The phenomenal growth of mobile phone and social media applications (see Figure 1) are placing a serious strain on network resources like bandwidth for both the telecommunications equipment manufacturers (TEMs) and original equipment manufacturers (OEMs).
Perhaps hardest hit are the Internet service providers and operators, customers of the TEMs and OEMs. The rise of the mobile environment has increased the need for security and deployment of faster networks at lower cost.
Several years ago, in anticipation of these needs, Intel laid out a Four-to-One (4:1) communication workload consolidation roadmap (see Figure 2). Similar to the familiar OSI layered network model, Intel’s communication path started with the application and control processing layers. Based on the Nehalm architecture, the Intel® Xeon® processor C5500/C3500 was the company’s first offering to cover the communication network application and control domains. Since that time, the company’s hardware and software offerings have been used in applications such as billing and accounting services. In the network control domain, Intel has focused largely on the radio network controller (RNC), a critical element in wireless area networks (WANs) that handles the set-up and tear-down of phone calls. The RNC heralds back to the days of telephony. Today, however, this same technology must deal with social media networking usage like text messaging, instant messaging and Facebook sessions. These additional tasks mean that TEMs have had to develop a different way to design and manage the control plane.
|Figure 2: Sandybridge-based Crystal Forest integrates communication workload functionality, including packet acceleration, into two chipsets.|
After the control plane, the next network challenge is moving data plane packets while maintaining the best performance per watt ratio. The key is to keep the thermal heat generation of the packet process as low as possible due to the thermally constrained form factor environments of today’s mobile devices. Shrinking form factors vary from standard boards like ATAC and AMC to proprietary form factors from Eriksson, Huawei (pronounced “wa-way”) and other TEMs.
The former Intel Xeon processor C5500/C3500 platform was designed specifically for all three telecommunication challenges in the application, control and data planes. It reduced the overall number of chips from 3 to 2, namely, the Nehalem processor and its chipset that included Input/Output (IO), memory control and DMA engine.
Crystal Forest Acceleration
Maintaining the same reduced chipset, today’s Crystal Forest technology includes a 22nm processor and based on Sandy Bridge microarchitecture a 32nm chipset that adds acceleration to the existing IO and memory subsystems. The accelerators include 0-100Gbps cryptography, compression and deep packet inspection functionality via Intel® QuickAssist technology. This means that no separate third party acceleration is needed at the board level.
Data processing interfaces include SATA and USB, while Ethernet handless the communication I/Os.
In addition to greater acceleration and security offerings, scalability remains the chief benefit to both communication hardware developers and programmers. “Our customers currently have nine different silicon architectures - multiple ones for application, control, packet plane from multicore providers like MIPS, Intel, Freescale and others,” explains Steve Price, director of marketing for Intel’s communications infrastructure division. “Plus they had all the different tools that went along with that–a very complicated environment.”
The value proposition offered by Crystal Forest is one of efficient design based on the same Intel ® architecture (IA) -based tool suite. In other words, the same instruction set architecture for the processor can be used in a variety of implementation domains. “The bottom line is customers can build a common platform that runs different software loads for application, control or packet and use it in different points of their network element or different elements of the network like RNC, Gatways or edge routers,” said Price.
Another advantage with IA tools is that common C/C++ code can be used to program the acceleration technology, instead of more costly and involved DSP specific or network processing code. Additionally, a new set of C/C++ libraries that run on native IA, will move packets much faster than before - up to 160 million packets per second. These optimized libraries can be acquired from Intel or the company’s ecosystem of providers, such as WindRiver, 6Winds and others.
Today’s Crystal Forest technology announcement introduces the next milestone in the company’s communication platform, i.e. acceleration of network services. The new platform is scheduled to be available later in 2012.
Steve Price is the marketing director for Intel’s Communications Infrastructure Division.