August 2001

Rebel By (Open) Design

BY ALAN DEIKMAN

[ Go right to: Building On An Open Architecture ]

There are numerous ways for design engineers to meet this challenge, but the focus of this article is cutting time-to-market by shortening the design cycle, particularly in software. Savvy usage and the best resources available to cut design time can enable OEMs to surpass competition by introducing products into the market more quickly and efficiently. This choice can lead the OEM to greater profits and market share, lowering research and development expenditures. By solving this issue, OEMs save time, money, and gain the benefit of first-mover competitive advantage that is so essential to retaining market share and customers' goodwill.

There are two elements comprising communications design for high-availability IP-based telecommunications systems: hardware and software. But with the widespread use of open standards (i.e., Linux and Ethernet) and the coming adoption of specifications such as PICMG 2.16, the key design issues today rest with software development.

ENTERING THE EMPOWERED AGE
A golden age of software development has just begun, where all the familiar commands and flexibility of the UNIX operating system reside in a more flexible operating system: Linux. Linux software is currently the most up-to-date and complete software code base in the world, providing tremendous empowerment to the developer community.

It is not unusual for simple software problems to be solved by hacking the Linux shell script, allowing programmers to do elegant things by writing in the shell's command language. Designers can implement fairly complex tasks with very little work. For example, one can write a shell script watching for a specific event, such as an intermittent error. The script can generate a warning e-mail when the event occurs, providing a diagnostic solution without any new equipment. This simple, remote diagnostics hack illustrates the power of Linux.

OPEN ARCHITECTURE
Despite the advent of operating systems like Linux, the major problem facing the embedded design community is writing complex software routines into their designs, which dramatically affects how quickly and efficiently their products are brought to market. So, what can be done to master these complexities?

One choice is to completely change all existing elements to fit the design specifications. This torturous route doubles or even triples design time, making the time-to-market issue a showstopper. Programming proprietary software can mean additional man-months of programming time, stretching already constrained product delivery cycle times and resources.

A more effective approach is to choose a switching product with an architecture that easily incorporates existing industry software without recoding. Such a switch would allow engineers to run existing software without modification. (See sidebar.)

The potential applications that can be designed with this type of product include: server load balancing, Web load balancing, VoIP, firewalls, virtual private networks, call management and concentration, multicasting, network monitoring, wire-speed routing, and SS7 signaling. Even more importantly, management protocols or daemons can be added to accommodate existing network management software.

For example, let's say you are writing BGP routing software, which must be compliant with both Juniper Networks JUNOS and Cisco IOS. The likelihood of these companies having current BGP stacks for utilization by programmers on their Web sites isn't particularly high. What is most likely is that one of the thousands of Linux application programming teams will be the first to post the latest BGP implementation. That implementation can then be added to your stack immediately without recompiling -- a stark contrast to waiting for delivery by a vendor. Scouring Linux software sites is relatively easy with Freshmeat, Linux Today, and Linuxapps.com listing the latest worldwide code updates on an hourly basis.

NEW TO COMPACTPCI
New specifications are also helping design engineers who have been using the CompactPCI specification to win the time-to-market race. Before CompactPCI, telco equipment vendors would use proprietary card cage designs that would implement the hot-swap and rear panel I/O features required. The CompactPCI standard has made possible multi-vendor and outsourced solutions, which offer designers these same features.

Telecommunications companies using CompactPCI switches have also discovered that many of the Ethernet cables being used were for board-to-board connections within the chassis, in addition to going outside the chassis. The existing bus architecture of PCI was not sufficiently scalable for the applications being implemented by engineers.

The new PICMG 2.16 sub-committee formalizes these architectures by simply bringing the Ethernet point-to-point connections onto the backplane. The generic PICMG 2.16 architecture is illustrated in Figure 1. Note that in this architecture there is no common bus and no single point of failure. Each node has two paths to communicate to the Internet or other nodes. This architecture can scale to the limit of the switch fabrics' ability to handle ports. By overlaying an embedded Ethernet switch onto the backplane, the subsystems function as standalone systems. Instead of moving via a shared bus, data is moved through an embedded switched 10/100/1000 Mbps Ethernet network, increasing overall system reliability.

The PICMG 2.16 design advantages are most clearly observed in the areas of hot-swapability and fault tolerance. In the past when box level components failed, total replacement was the only option. This meant removing Ethernet cables and the failed unit and then installing a new unit, recabling, and reprogramming to avoid degrading overall system performance. The proliferation and inaccessibility of co-located telco sites creates a painful nightmare for maintenance technicians, making hot-swapability an essential design feature.

With PICMG 2.16, the individual boards are isolated from the chassis so that users can perform system diagnostics and repairs without adversely affecting other subsystems. By isolating every board on the backplane, products can ensure successful mission-critical applications without costly downtime. As of this writing, although the PICMG 2.16 specification is still in draft status, pre-standard products are being introduced to the market for OEM evaluation and testing.

FOR THE FUTURE
Ethernet's reliability and ubiquity has been the touchstone for much of the successful design implementations in the embedded systems world. Since the ratification of IEEE 802.1, a full two generations of programmers have been successfully using Ethernet LAN topologies in their designs. In addition, all operating systems support some form of Ethernet and IPv4 protocol. Proven LAN technology has created a backbone for the proliferation of rich IP-based connectivity devices.

Alan Deikman is CTO of ZNYX Networks. ZNYX is a supplier of scalable, high availability LAN solutions for servers, workstations, and embedded systems. With a comprehensive line of PCI, PMC, and Compact PCI network adapters, ZNYX has what you need for your commercial, telephony, or industrial applications.

[ Return To The August 2001 Table Of Contents ]

A large telco provider generally has an expensive set of requirements for qualifying telco equipment. Generally there are dozens of projects that they need to bring to market. So what typically happens is that in order to cut budget requirements, the telco provider will attempt to develop a standard platform from which to base new products. In other words, if they can develop a chassis, a power supply, a switch fabric, and a set of boards such as single board computers that have already gone through the qualifying process, new projects can use those products without paying for that qualification again.

CompactPCI makes this possible. But one of the problems in this type of project that cPCI does not address is that each project has different software requirements with regard to management. In other words, Project A will need SNMP management, Project B will need HTTP management, and Project C needs a completely proprietary protocol, which they already have in place. So how can the telco provider conform to these different requirements with one piece of equipment?

Through use of a platform such as ZNYX Networks' OpenArchitect, the telco provider could have the switch conform to all of those different management methodologies with the same hardware. The result being that they do not have to qualify three different sets of hardware to complete three different projects. By using a platform like OpenArchitect, telco providers are able to keep costs down and bring products quickly to market.

[ Return To The August 2001 Table Of Contents ]