June 2001

Marching Towards The Core: Board Level Platforms Enter The Network

BY RON KENNEDY

[Go right to Call Density In Carrier-Grade Systems]

Bigger, better, faster: few casual observers back in the early 1990s might have thought that those three simple words would be used to describe the evolution of board level platforms into the new millennium. After all, CTI back then still struggled for mind and market share. Some CPE environments ran small four-port voice mail systems as an adjunct to the on-premises PBX. Many network managers, however, couldn't be convinced to even try out the new computer telephony paradigm. Application developer needs in most cases could therefore be met with single DSP analog boards capable of supporting 10 to 100 simultaneous users, depending on the service.

Yet commercial board level platforms today are quickly evolving on several fronts. Driven by demands from emerging competitive service provider markets, densities and scalability are orders of magnitude higher than just a few short years ago. IP connectivity is also becoming much more central. As well, eight years ago standard DSPs could perhaps run a single IVR application over 12 channels. This is also growing by an order of magnitude, to the point where DSPs can now run multiple applications over 256 channels and more, with onboard co-processors or ASICs (Application Specific Integrated Circuits) providing functionality such as TDM-to-IP conversion.

Board quality and reliability are also beginning to approach the expectations of many "five-nines" customers with the emergence of CompactPCI technologies and standards. The more recent emergence of Compact Packet Switched Backplane (PSB) technologies will move this even further ahead, and help drive system-wide scalability into the hundreds of thousands of customers.

Time To Advance
Board level platforms that form the heart of CT systems are taking on many of the characteristics required within the very core of service provider networks. Vendors in the new ASP, CLEC, and voice portal markets are increasingly turning to CT systems as a way to cost effectively and quickly differentiate their services while gaining a leg up in fiercely competitive environments. The original promise of the Advanced Intelligent Network (AIN -- the SS7 network) of seamless and easy service creation is coming to fruition in a commercial standards-based form factor.

This trend will likely reach its next logical plateau over the next two to three years with the emergence of an increasingly ubiquitous softswitch-based network. Commercial board level platforms will have transitioned from running small voice mail systems in CPE environments ten years ago, to running CO adjunct systems hanging off the SS7 network today, to running the functionalities of the traditional proprietary telco switches at the heart of the softswitch network in the very near term.

Central to this paradigm shift will be continued advancements in signal processing technology. Developments in DSP horsepower, smaller footprint packaging, and reductions in power dissipation will all help improve overall system scalability while reducing costs. Increased processing in particular can translate into cleaner signals, better compression, or more voice or modem channels per chip.

Arm-In-Arm
But DSP chips no longer have the signal-processing world to themselves. Today, designers can readily add some of the key DSP building-block functions and instructions to devices ranging from low-cost general-purpose micro-controllers to the latest generation of x86 and RISC (Reduced Input Set Computer) CPUs. Specialty architectures, such as vector processors that conduct single-instruction/multiple-data (SIMD) operations and very-long-instruction-word (VLIW) processors, can perform many operations in parallel. These powerful architectures also are being used in signal-processing applications.

The initial attempt to perform DSP-like functions on general-purpose processors like the Intel Pentium resulted in mediocre performance thanks to low clock rates. But today's high internal clock rates (up to 900 MHz in a few cases) and the addition of a few key features -- such as single-cycle multiplier-accumulators, improved floating-point units, and some signal-/image-processing instructions -- make newer processors worthy competitors to traditional DSPs. And with the addition of dedicated vector coprocessors that can perform SIMD operations, RISC CPUs can now surpass the performance of all but the most advanced DSP chips.

Whether through DSPs or newer variants of CPUs, more processing power will soon be available to help effectively deliver many types of next-generation applications such as voice over IP (VoIP). Of equal importance to many next-generation service providers will be the ability to squeeze as much capacity as possible out of each chassis. Standard port densities now reach 24 analog lines or 8 T1/E1 spans, with 16 spans a near possibility for some cPCI cards. The next logical step in the form of T3 boards is now emerging, with each port providing carriers the equivalent throughput of 28 T1s. Dual T3 cards that are fully VoIP capable should be more than able to handle up to 670 active simultaneous users. Dual OC3 boards in the future hold out the possibility of supporting 4,600 users.

Into The Core
At the heart of service provider needs for highly scalable, cost-effective, robust board architectures for the next-generation public network is the requirement to do away with the static allocation of CPU resources, voice processing capabilities, and individual ports to each and every CTI application. Such traffic engineering calculations were essential just a few short years ago. Service providers that wanted to deploy CTI applications running on board level platforms would need to first gaze into their mystical crystal ball. They could then arrive at a complex series of "best guess" estimates as to how many customers would need what kind of access to how many applications at any given point in the day. These numbers would then allow engineers to bang resources into each and every card. Changes in demand for one application over another, of course, threw this entire cumbersome architecture off balance. This inflexibility makes it totally unsuitable for the hosted application service model.

Telecom board architectures today essentially deliver full flexibility at a multi-chassis level to shift and share resources on the fly. For hosted application service providers, this means that CTI systems within large centers can be designed to scale in all dimensions of an application over multiple hardware platforms. The same system within smaller centers can also be collapsed into a single chassis, while still preserving the same feature-rich suite of services and potential scalability. From an application standpoint, system-wide resources within this architecture appear almost as a virtual pool running across multiple CTI chassis.

Onto The Backplane
Where this reaches its true potential is with the emergence of a specification called Compact Packet Switched Backplane (cPSB). This new telecom bus architecture basically maps existing switched Ethernet/IP networks directly onto the backplane of each CTI chassis. An end-to-end IP architecture offers significant cost and system integration advantages over other technologies like ATM an inter-chassis connect schemes. Standards such as RSVP and MPLS both offer a way to achieve this. Both have attracted substantial industry resources, time, and effort over the past several years.

The concepts of pooling and backhauling DSP resources are not new. Board level platforms on the market today are built from the ground up to fully accommodate these architectures. Such systems can currently allow up to 256 voice channels to be pushed across a telecom bus and another 256 channels pulled across an internal switching fabric to a maximum of 8 T1 spans -- all on the same telecom card. This allows developers to build extremely dense CTI systems with all the built-in redundancy inherent to resource sharing. Hardware costs and system complexities are substantially reduced, while overall internal scalability is increased. A ubiquitous internal and external IP bus architecture based on cPSB technology will enhance the scalability of this optimized CTI infrastructure into the tens of thousands of ports range.

The cPSB specification is still not well defined, although appears to have the blessing of some of the industry's heaviest hitters. Spearheaded by Performance Technologies, an industry consortium is targeting mid 2001 for fast track ratification of a cPSB standard as an extension to the PICMG 2.x family of specifications. The proposed changes are expected to retain many of cPCI's attributes, as well as leaving the H.110/H.100 buses intact for systems that support those standards. This type of model will enable the creation of integrated systems in which any card can reside in any link slot and run any operating system, provided that each card supports standard Ethernet protocol interfaces. Product time-to-market can be improved since integration occurs at the system level rather than at the driver level.

One way to ensure that systems can evolve over time might be to include both the H.110/H.100 and cPSB pin-outs on the same resource card. This would offer maximum interoperability and resource sharing between systems. H.110-based resources could also access the faster bus architecture by simply crossing a card-level switching fabric to a larger GigX on-card switch.

Aside from the intriguing media convergence possibilities, this emerging architecture will help further fuel the rapid adoption of CT platforms into large scale hosted applications and central office environments. This should prove especially true again for the emerging softswitch market, where telephony systems that are increasingly more dependable, smaller, and cheaper to own and use will quickly become an industry imperative.

Ron Kennedy is VP of marketing at Pika Technologies. For more information, please visit www.pikatech.com.

[ Return To The June 2001 Table Of Contents ]

BY JOSH ADELSON

Traditionally, port density has been a major focus for developers of carrier-oriented communication boards and systems. Vendor claims are constantly leapfrogging one another. While port density is important, focusing only on this single dimension can eclipse awareness of other critical attributes necessary for carrier-grade systems to meet the capacity requirements of service providers. These other attributes include high-availability, processor efficiency, manageability, and calls per hour, or call density.

Call density refers to the number of calls a system can process in a given time period. It seems self-evident that high call density is desirable; however, it's useful to review the underlying reasons for this, as follows:

Service availability: That is, the availability of the service to subscribers at the moment they request it. Availability expectations in telecom environments are so high that making a subscriber wait even a few seconds to access the service may be unacceptable.

Real estate cost reduction: Real estate in telecom and networking environments (i.e., co-location rack space) is so valuable that operators need to maximize their service capacity for a given amount of floor space.

Network cost reduction: Minimizing the cost of network resources required by the system. Depending on the environment, network costs may be actual monthly costs for access lines connected to the system, or they may be in the form of scarce ports off the main telecommunications switch.

Efficient host CPU utilization: The system needs to be able to handle a high number of calls while still leaving plenty of host CPU cycles for the higher-layer aspects of the application.

Requirements Vary By App
"Call density" is defined as the number of calls per unit of time, as opposed to "port utilization," which is defined as the number of calls active within the system at any given time. Systems that handle short-duration connections require high call density attributes because they will, by definition, need to handle the highest number of calls per unit of time. Systems that handle long-duration connections also benefit from high call density attributes, but it is less of a critical factor for them.

Gateways, such as voice over IP gateways, softswitch media gateways, or wireless voice gateways, typically carry real-time voice conversations that last for half a minute to an hour. The average call duration may be several minutes, so these systems -- even when fully utilized -- may not be handling a large number of calls per unit of time.

Enhanced services platforms, including intelligent peripherals and softswitch media servers, are used for applications such as messaging, announcement services, connection services, or voice portals. These systems typically handle shorter sessions than gateways do. This varies considerably by application. Voice messaging sessions are usually in the 10 to 20 second range although they can be much shorter -- for example, when no message is left.

Many network announcements are in the sub-5-second range, for example, talking call waiting prompts or playing "all circuits are busy." Speech-recognition sessions for voice portals and directory assistance systems vary in duration, but these too are generally brief when the main purpose of the session is to direct the call. Connection services, such as call screening, are nearly instantaneous from the perspective of the platform performing the connection function. The graph (see Table 1) illustrates relative call durations for different types of applications.

When testing your system's call density, focus on three major parameters:

Calls per hour: The number of successful calls, under a variety of call types ranging from simple connect/disconnect to playout of voice prompts.

CPU utilization: The extent to which the host CPU power is consumed by the execution of call-related functions, i.e., call setup and tear-down, media play and record. (The CPU cycles not required for these functions are available for use by higher-level applications.)

Stability: In addition to supporting high call volume, it is also important for systems to keep running without errors. A system able to handle a heavy call load for an extended period -- at least 24 hours -- is further evidence of its robustness.

Summary
Call density is a critical attribute for carrier-grade systems, especially for enhanced services platforms. Thoughtful design at the board level is needed to ensure high call density at the system level.

Josh Adelson is Market Development Manager, Brooktrout Technology. For more information, please visit their site at www.brooktrout.com.

[ Return To The June 2001 Table Of Contents ]