Media-oriented networks require CMPs
Keywords:communications and media processor? multimedia content delivery? multimedia content? memory-to-memory architecture? multithreading?
Networks are rapidly evolving from simple connectivity providers to media providers. This is a change from a data-oriented network to a multimedia-driven network. The end result will be on-demand access to high-quality streaming media applications such as HDTV, VoIP, gaming and high-definition audio. An example of such an environment is the digital home, where wireless home gateways, gaming consoles, VoIP phones, digital media players, TVs and network-attached storage deliver, distribute and play media apps simultaneously.
This new and exciting media-rich environment, however, presents new challenges. In particular, processor designers are facing problems they've never had before. Traditionally, processor designers have used the "bigger hammer" approach to solve any problem—create faster processors, and everything will be solved. In this new environment, however, speed is not the only factor. Higher performance is necessary because media applications are performance-intensive.
There are several other media-applications requirements that cannot be solved effectively or economically with legacy architectures and solutions. These requirements include high sustained throughput, deterministic processing, multiple concurrent application management, enhanced security, optimized cost-performance ratio and RTOS. This shift in network apps has given birth to a new class of processors specifically designed from the ground up to meet the challenges of the new media age. Known as communications and media processors (CMPs), these processors meet the stringent requirements needed for delivering multimedia content.
Sustained performance
Legacy architectures such as RISC-based processors were previously used for standard networking needs. These are bursty in nature—high burst of throughput followed by quiet time. In contrast, media apps, such as streaming video and audio, require sustained high-bandwidth packet processing with ultralow jitter. Legacy RISC-based architecture processors use cumbersome context-switching, interrupt-driven processes to handle traffic. But this architectural model breaks down when required to handle sustained high-bandwidth traffic with ultralow jitter and with required QoS for media content delivery. Some of the additional fundamental shortcomings of legacy processors are low architectural efficiency, long hazard time, long memory wait time and use of general-purpose OS.
The number of instructions it takes to complete a typical task or benchmark is referred to as architectural efficiency. If one instruction set uses fewer instructions, it will complete the task sooner, even if the clock frequency is the same. RISC-based processors have a load/store instruction set that means data must be moved to registers before being used. Moving data from one location to another takes two instructions—a load and a store. This is a good choice for servers, for which most of these processors were originally designed, but is not optimal for the packet processing required for communications and media applications. To process a packet, the CPU scans the packet, makes minor changes and then perhaps copies it. For packet processing, a memory-to-memory instruction set is the most efficient architecture, but it is not implemented by legacy processors.
 |
| Traditional processors use cumbersome context-switching interrupt-driven processes. |
Architectural edge
CMPs have many architectural advantages over RISC-based processors such as on-chip memory, multithreading and dedicated OS. Besides raw speed, these architectural advantages enable CMPs to offer high sustained throughput, deterministic processing, multiple concurrent application management, security acceleration, optimized cost-performance ratio and RTOS.
The most efficient design to meet the challenge presented by media applications is a memory-to-memory architecture combined with multithreading. Benefiting from on-chip memory, packets can stream directly to on-chip memory, be processed completely by any of the independent threads and then be sent out. This maintains high sustained throughput and reduces jitter by 95 percent over traditional architectures. In contrast, legacy processor applications run from off-chip DRAM through a cache and spend 30 percent to 60 percent of their time waiting for cache misses.
 |
| CMPs use memory-to-memory architecture and multithreading for efficient processing. |
Media applications are very sensitive to jitter. Jitter influences the responsiveness of the system and can affect user experience. Deterministic processing provides a fixed time for processing each packet, thus reducing jitter from milliseconds to microseconds. Combined with the ultralow latency of multithreading, it provides the best QoS and user experience for media applications.
Today's digital environment is applications-rich. This means multiple and diverse applications need to be processed in parallel and with low jitter. The advantages of a memory-to-memory and multithreading processor are evident because it can independently process multiple applications with low jitter.
A dedicated hardware security engine offloads security processing from the core for higher performance and maximum network protection. This engine, combined with encryption and authentication enhancements, and stateful and stateless professional-grade firewalls, protects networks from outside hacking and identity theft. It also protects digital assets from unauthorized access.
CMPs and RISC-based processors all execute one instruction per clock peak. Hence, the perception is that their performance is completely determined by the clock frequency. This is not the case though. Additional factors such as architectural efficiency, hazard time, memory wait time and OS type directly determine the performance of a processor. Thus, CMPs inherently have much better performance at the same frequency than legacy processors. This gives CMPs a higher cost-performance ratio.
A more efficient software implementation uses fewer instructions to do the same amount of work. Most RISC-based processors rely on a general-purpose OS such as Linux. These general-purpose operating systems are not optimized for communications and media applications. They are mostly designed for computing applications, so it's not surprising that they are slow, have low efficiency, produce high latency and jitter, and lack the flexibility needed for emerging media applications. In contrast, CMPs take advantage of dedicated operating systems for low jitter, real-time processing and up to three times more efficiency.
The change in network applications and the move from simple connectivity to media playability present new challenges and opportunities to processor designers. While the transition is still happening, one thing is evident: legacy RISC-based architecture cannot meet the challenges of new media effectively or economically. There is a need for a fresh and innovative approach specifically designed for media applications, not a re-packaging of old solutions. The features of the new and innovative CMPs certainly fit this bill and provide an optimal method for delivering multimedia content.
- Ali Simnad
Sr. Product Manager, Ubicom Inc.
Visit Asia Webinars to learn about the latest in technology and get practical design tips.
All rights reserved.