Posted: 02/1999

Client-Server Technology Comes to Telephony Switching
Alternative to Mainframe CO Gear Promises Lower Costs, Open Applications Development
By Peter Lambert

Just as distributed servers and client workstations replaced many bulky, centralized
mainframe computers in data networks over the past decade, so distributed networks of
servers, workstations and software have begun to augment and, in some cases, to replace
bulky, centralized mainframe telephony switches.

Pushing the migration away from mainframe switches are new carriers, data and
telecommunications equipment component makers, a raft of startup switchmakers and the
dominant makers of mainframe telephone switches themselves.

The migration promises to begin in earnest this year, particularly for a new generation
of carriers that seek to compete at lower costs with Internet protocol (IP) networks built
from scratch. Those carriers–including competitive local exchange carriers (CLECs) such
as Convergent Technologies Inc., Englewood, Colo., and interexchange carriers such as
Level 3 Communications Inc., Denver– all wish to minimize or eliminate the use of
traditional Class 5 local access and Class 4 trunking switches and to maximize use of
packet switches for all manner of traffic, including voice.

The merging server-based switches are designed to turn that trick by effectively
unbundling functions and features that traditionally have been bundled into monolithic
mainframe switches. The service functions then reside on "feature servers," or
"policy servers," and there handle all higher intelligence related to where any
voice or data call can go, and with which features.

Most vendors and carriers expect the full-blown, mainframe central office (CO) switch
will maintain a substantial role for a long time to come. Indeed, in the early going, most
new server-based switch designs are predicated on connection with mainframe Class 5 and
Class 4 switches for access to time division multiplexed (TDM) circuits on the global
public switched telephone network (PSTN).

However, server-based switches promise carriers very low-cost market entry,
particularly into small cities where return on investment (ROI) may appear most risky. As
an alternative to building $2 million to $5 million Class 5 COs in towns that may never
exceed 1,000 voice lines, the switch server makers claim their customers can start with
700 to 1,000 lines at a $100,000 to $150,000 price range. Some claim they’ll soon undercut
the $200-per-voice-port milestone with platforms that handle not only circuit traffic, but
packet traffic as well.

Eventually, the benefits of steadily declining computer components and of open
applications programming languages may drive the adoption of these "open-systems
computing" technologies in larger networks in larger cities.

But one thing is for sure. Telecom stands in Silicon Valley’s cross-hairs.

"The telecom market is now the target of the entire computing industry," says
Joel Hughes, director of IP telephony products for Natural MicroSystems Corp., a
Framingham, Mass.-based maker of programmable computer chips for networking gear,
including server-based gateways designed to bridge the IP world to the wireline and
wireless PSTN.

Open-systems computing, as the weapon for that target, promises to distribute switching
processors far and wide around a network, and to control that distributed network with
call setup and enhancement software housed in adjunct servers.

At bottom, this removes such functions as number translation and call detail record
(CDR) data collection from Class 4 and Class 5 PSTN switch hardware, placing the features
closer to the customer and making them available to both circuit- and IP-based calls.

Critical to multiservice IP network capacity growth, this decoupling of service
intelligence from transport equipment unburdens the switch and router hardware of
sophisticated call-processing tasks, freeing them to concentrate on their main task:
pushing information across a network with minimal transmission delay, even at very large
scale.

At about one-twentieth of the base price of PSTN switches, such platforms radically
will reduce "the price of being wrong" about potential services revenues in a
given city, says Tom Burkhardt, president of Castle Networks Inc., a Westford, Mass.-based
startup that introduced its first 840-port, $125,000-to- $150,000 TDM-IP-frame relay node
switch last month.

"Circuit switching is not dead, but the unbundling of what’s on the mainframe
circuit switch has begun," says Burkhardt, whose company is staffed like its
competitors–with a mixture of data communications, telecommunications and semiconductor
experts. "CLECs can only compete with incumbents by replacing the old incumbent tools
with a new architecture of distributed switches and servers."

Applications Innovations

Beyond cost reduction, the infusion of telecommunications systems with open,
distributed computing systems promises two other general benefits:

Graph: 21st Century Mediation Services

Convergent Communications Inc., which operates as a CLEC in a dozen U.S. markets, is
"proving out the distributed switch model," says Greg Tennant, senior director
of data services. Using IEX Corp.’s DaVinci server-based node switch, Convergent
"mimics a Class 4 switch on an open platform, which enables us to take the service
logic out of the PSTN hardware for things like local number portability (LNP), VPN
(virtual private network) dialing plans, calling card processing and CDR data
collection," Tennant says. "This allows me to choose what services I want to
centralize and to choose where I do each feature, vs. 10 boxes for 10 cities, and I can
choose the most economic transport for least-cost routing (LCR)."

In other words, says Bill Plunkett, senior vice president of CLEC sales for Lucent
Technologies Inc., Murray Hill, N.J., "a carrier will almost certainly need PBX
(private branch exchange) features and VPN support, but it may or may not need coin-phone
service, so work with each customer to create a package of features that it needs."

In the bargain, the features server is programmable by anyone with skills in Unix,
Microsoft Corp.’s Windows NT, Sun Microsystems Inc.’s Java or other standard applications
languages.

"Right now, in most cases, the vendors for the circuit switches and the
applications platform are the same company, limiting the speed with which enhanced
services and subscriber-based services can be created and launched, which is where the
real value will lie for new carriers," says Hassan Ahmad, former chief technology
officer for Alameda, Calif.-based Ascend Communications Inc.’s core switching division and
now president of Sonus Networks Inc., Westford, Mass.

Sonus’ server-based switch is designed to act both as a peer to other standard Class 4
circuit switches and, simultaneously, as a gateway to both circuit and IP backbone
networks. "A carrier can use the switch as another Class 4 trunk switch and, at the
same time, incrementally begin to migrate services over to lower-cost packet trunks,"
says Rubin Gruber, Sonus chairman and founder. He adds that, while current generation
gateways are limited to less than 100 ports, "we can scale to 64,000 calls and
higher."

The switch will leverage signaling system 7 (SS7) call controls and enhanced
applications generated by existing service control point (SCP) servers in the PSTN.

At the same time, Sonus promises to deliver an additional infrastructure of servers
less coupled to circuit switches. It plans to openly publish an applications program
interface (API) for use by any and all developers to create new services. Those services
will run on a network of call-control, call-enhancement policy servers which, in turn,
will direct the operations of switching and routing hardware.

"The ideal is to enable the service provider to build his own service on his own
server," Ahmad says. "He can put any number of open-services architecture
servers in any number of places on his network, upload an applications script to the
switch, and automate the process of deciding what services are available to a call and how
the call is to be handled and routed."

Graph: Local Services Expansion

According to Paul McNab, senior manager of systems architecture for San Jose,
Calif.-based Cisco Systems Inc., this combination of adjunct servers and open APIs
"means that the person who is managing policy doesn’t need to have network knowledge
about whether the transport is ATM (asynchronous transfer mode), frame relay or gigabit
Ethernet or TDM switching, and now, rather than configuring every switch or router, you
can configure once, through one graphical interface."

According to Natural MicroSys-tems’ Hughes, "The need is to move the $150 billion
telecommunications equipment market to mass-market computing platforms."

Open and Reliable?

In addition to leveraging low-cost, off-the-shelf computing equipment and openly
programmable software, many vendors are incorporating both data packetization and
universal gateway functionality in their distributed computing switches.

This "services mediation" function would provide a single switching fabric
for all traffic–circuit and packet, voice and data–emanating from end users.
Theoretically at least, that single fabric can, on the fly, create a virtual connection to
the appropriate wide area network (WAN) for each call, whether to the PSTN, Internet or
private network, without configuring or reconfiguring actual physical connections to those
WANs.

Startup vendors staking claims to these goals include Castle Networks Inc., Westford,
Mass.; IEX Corp., Richardson, Texas; Salix Technologies Inc., Gaithersburg, Md.; Sonus
Networks; and TransMedia Com-munications Inc., San Jose, Calif.

At the same time, dominant data and telecom vendors such as Ascend Communications;
Cisco Systems; Lucent Technologies; Nortel Networks, Brampton, Ontario; and Siemens Corp.,
Boca Raton, Fla., have begun to deliver open-systems platforms, particularly for the small
market carrier.

All these advocates admit that the key challenge to open-systems telecom is to achieve
these goals with an infrastructure as scaleable, stable and reliable as the proven PSTN.

"In mainframe telecom, proprietary systems have assured stability, reliability and
scale, but yielded little applications content, leaving enhancement development in the
hands of a few dominant vendors," Hughes says. "In contrast, in the data
communications world, open systems have yielded lower costs, high applications content and
rapid change, but datacom vendors have only just begun to address reliability and
scale," he says. "The bogey here is making open synonymous with reliable."

Delivering that reliability presents a tall but–carriers agree–achievable order.

According to Convergent’s Tennant, "We don’t yet see a platform where all Class 5
switching can be done on single cell-packet-IP platform, so we can truly do voice over
X." But he concedes that Convergent directly sells long distance voice service but
only resells local phone service.

Although the converged box is "the right concept" for Conver-gent’s remote
nodes, "it’s not proven out yet that a gateway to the PSTN can work at scale and not
break, because that’s one box understanding every language and using every language very
well. But the next three or four months may reveal some breakthroughs."

Service Mediation

Both veteran and startup vendor pitches share a common refrain: The initial task is to
use the cost structures of open computing to bring PSTN functionality to IP networks at
competitive costs.

"IP-centric CLECs need at least parity with incumbents, so it’s important to carry
voice features into packet switching via feature servers," says Diane Herr, vice
president of switching access product marketing for Lucent in North America. "We are
separating some of the features and services from the transport fabric, so that services
are able to work over both circuit and packet fabrics, and we’re doing that in the Class 5
switch as well as in edge access devices like the PathStar Access Server."

Using Lucent’s AnyMedia multiservice access intelligence, the PathStar converts data
and voice traffic from phones, fax machines or personal computers (PCs) to IP packets and
routes them to IP networks. The PathStar server can be scaled from 512 to 6,000 DS-0 (64
kilobits per second [kbps]), plain old telephone service (POTS) lines while also
delivering standard PSTN features, such as PBX, Centrex and primary rate integrated
services digital networks’ (PRISDNs’) connectivity to remote carrier offices, all at an
average price of $300 per line.

Graph: 21st Century Services Architecture

The "data shelf" that is common to both PathStar and Lucent’s Class 5
AnyMedia switch also incorporates gateway functionality and CDR collection.

Nortel, which is integrating its IP Connect gateway technologies into its full range of
large and small voice switches, similarly emphasizes the notion that new, IP-centric
carriers first must match incumbent capabilities at lower costs, before they can better
incumbent offerings with enhanced services.

"CLECs realize that they need all the capabilities of the incumbents, and then
they can build features from there to differentiate themselves," says Ira Sager,
senior manager of voice over IP (VoIP) technology for Nortel Networks’ Bay Networks line
of business.

Adds Steve Breese, senior manager of IP Connect product marketing, "In addition to
providing a series of gateways for service providers across our switching products, we
provide a call-control gatekeeper function and open-services platform on top of that for
multimedia services that take advantage of the unified network fabric. Single-stage
dialing, digit manipulation, calling card, PBX dialing plans, CDRs, billing–all these
functions are available to IP."

Like other vendors adopting the client/server model, Nortel’s strategy promises to
force no overnight changes in customer behavior or access equipment. "We’ve focused
on developing a thin client for customers that translates between the huge legacy of
existing equipment, like fax machines and IP networks, which promise lower cost and
greater flexibility for electronic commerce and other voice-over-packet
applications," Breese says.

However, current incompatibilities among gateways present a key barrier to fulfilling
the open computing dream for "data bigots like us," says Tom Korte, director of
network engineering for Convergent. He places hope in current Internet Engineering Task
Force (IETF) work on the multimedia gateway control protocol (MGCP). "We’re handling
local Class 5 services through resale until the MGCP standard solidifies, and then we’ll
be able to use data protocols inside our own network, and to communicate with local
carrier switches via SS7 where we have to do so."

Few may bet against the computing industry’s ability to replicate in the
voice-networking realm what it has accomplished in data networking. The only questions may
be how soon and how widely open systems will revolutionize telecommunications.

"The idea of stacking PCs in a central office collocation site doesn’t fly,
because you really need a specialized box with integrated gateways to build out a
network," says Nortel’s Breeze.

Indeed, integration itself may have its limits, on either proprietary or open systems.
"Our lead question," says Convergent’s Tennant, "is whether there is a
point of diminishing returns in building all that function into one platform."

Yet even with limits on how far open computing can go in enabling integrated services,
Convergent is convinced that every measure will help. "Call features, which now are
coupled to hardware, will move to servers and Java applets, and there are lots of
teen-agers today writing Java," Korte says. "All that is of great benefit to
us."

Peter Lambert is features editor for PHONE+ Magazine.