The Network Applications Platform (NAP) is fast becoming the ubiquitous workhorse of both infrastructure and enterprise networks.
While NAPs often take on many guises, IP traffic is at the core of all applications. Maximising reuse and development efficiency, while creating innovate and feature rich solutions is crucial for success. This paper, part one in Advantech's Network Application Platform series, examines the challenges of processor selection.
We all have a tendency to look at the past through 'rose tinted spectacles'. The traditional networks all seemed so simple and straight forward. There were clear demarcation lines between the voice and data networks of the core infrastructure. Even as enterprise networks evolved from basic Ethernet backbones to major data centers and farms of webservers, the PBXs running the phone network were in a whole different world.
Each element and piece of equipment, whether a voice switch, DSLAM, router or PBX, was designed based on the type of traffic and physical interfaces required. Specialized boards, chipsets, software and connectors were all required. Architecting the platforms required all these elements to be considered and the hardware concerned nearly always required dedicated, discrete circuitry and support chips.
Next generation networks and convergence have changed the game significantly. As we move closer and closer towards the all IP network the distinctions between network elements are significantly more in software and application than the underlying hardware or interconnection media. This has lead to new challenges and design considerations.
General purpose processors (GPPs) played a significant role in the underlying architecture of all communications equipment. GPPs however were always designed to be a flexible computing 'workhorse'. However, as the applications became more specific and demanding, new breeds of specialized processing engines began to emerge. With the data and voice elements of the network still separate two key devices were used to boost performance and capacities, the Digital Signal Processors (DSP) and Network Processors (NP).
The DSP catered to the specific needs of the voice network. The key application function was to handle the analogue signals. These needed to be converted backward and forward to digital, and relatively simple (linear) processing performed very rapidly. The mathematical engine was optimised towards greater performance for floating point operations as distinct from a GPPs primarily integer based core.
Yes, a GPP could do it but a DSP was more efficient, performant and cost effective. The other key difference between DSPs and GPPs is their deterministic nature. In time critical applications where delays, even in the milliseconds will be noticed, DSPs were the clear winner. When talking on the phone the conversation must be processed in real time where even minute delays will be noticed as stuttering or gaps in the conversation.
Similar scenarios were applicable to the development of the Network processor. The early needs of switching and routing were handled through software running on a GPP. As the multilayered protocols became more complex the first processors dedicated to specific layers were created. The first key development was the Ethernet switch focusing on only the requirements of layer 2.
This led to a revolution in specialised chips and ASICs (Application Specific Integrated Circuits) each with specialised architectures, memory configurations and low level microcode, streamlined and optimised for the application and protocols for which it was designed. Again GPPs could manage the same tasks but the specialists did a better job. A GPP would need to apply 'brute force' which usually meant more processors and more power.
Email us for the full story, firstname.lastname@example.org, please mention the article heading.