HERE'S the BARS REPORT on the role of the ATM in broadband...VERY GOOD STUFF:
THE ROLE OF ATM IN BROADBAND SERVICE
PROVIDER NETWORKS:
Why Carrier-Class ATM Is Not Another Technological Mirage
Paul Johnson, CFA (212) 407-0415 paul_johnson@rsco.com
Tim Weingarten (212) 407-0444 tim_weingarten@rsco.com
Ara Mizrakjian (212) 407-0406 ara_mizrakjian@rsco.com
Continuing with our theme of research on exciting opportunities in the service provider
equipment market, we recently attended a conference in Miami, Florida, on the state of
ATM technology and usage in carrier networks. After several conversations with a host
of CLECs, IXCs, and RBOCs and sitting through the aforementioned conference, we
thought it would be interesting to present our view of the 1998/1999 market demand for
both traditional as well as soon-to-ship core switching and transport hardware. While we
have aggressively pounded the table since the end of 1997 on companies with heavy
exposure to this space - particularly Ascend Communications (ASND $43 7/16, Buy) -
and have recently been joined by most of our fellow analysts, we believe it would be
helpful to explain the underlying dynamics of ATM demand by examining why and
where ATM has been deployed and at the same time uncovering its potential going
forward.
Along the lines of some of the more prevalent investment ideas in the networking and
telecom equipment sectors that we continue to see as presenting strong upside
opportunity in 1998, carrier-class ATM deployment represents what we believe to be the
most important catalyst for both above-industry-average growth and upside surprise to
expectations this year and next. Based on our belief that a number of both new and well-established
service providers will spend an increasing share of capital on core WAN
switching equipment in 1998, we expect to see a continuous stream of large ($30 million
+) tenders contributing to a greater than anticipated (53%) growth rate in ATM
equipment. Additionally, we believe the opportunity for ATM switch vendors is
expanding as new, unaccounted-for markets open up to ATM switches incorporating
functionality that currently requires stand-alone hardware. In particular, we view the
SONET ADM market as the most vulnerable to ATM vendors in the short term as both
linear and ring-based add/drop functionality begins to be engineered into interfaces on
ATM switches. With these themes in mind, let's examine how ATM is in use today, why
carriers are choosing ATM, and which vendor's ATM gear is being chosen.
BancAmerica Robertson Stephens 1
We believe the most critical issue facing service provider network planners today is how to build the
most future-proof, scaleable architecture that will expand to meet the phenomenal rate of growth in
demand for their bandwidth and services. While in the past it was fine to overlay each data type or
service offering on a separate switching and routing infrastructure, which sat on top of a common
underlying SONET network, this has proved to not cost-effectively scale to the extent that is required
today. There are substantial non-linear increases in network complexity and cost as providers add new
customers and new services. For these networks, which are comprised of routed IP networks with
`clouds' of frame relay switches, it is not cost-effective to scale due to the simple fact that they have
been built with point-to-point circuits in mind. Adding more customers or core capacity means adding
more leased circuits, more router ports, and more frame relay switches to an already expensive mesh of
DS1s and DS3s.
Looking at the use of frame relay more specifically, today's network providers use frame relay switches
as multiplexors. Because Cisco (CSCO $73 15/16, Buy) and Wellfleet (BAY $23, Buy) couldn't build
routers that were cost effective at aggregating low speed (below DS3) access lines, almost all carriers
use frame relay switches (typically Cascade 9000s or Newbridge (NN $27 7/8) 36170s) to multiplex
these access links into a router (typically Cisco 7500) that is `one-armed' off of the frame switch. In
many cases, it is possible that with the use of a router, costs rise as a function of growth, so that the cost
per customer is actually inverted. For example, an ISP's router could have 50 T1 ports and one FDDI
or Fast Ethernet blade. In the same PoP, there is a router with an FDDI or Fast Ethernet interface and a
bunch of DS3s for backhaul into a core backbone. So, essentially the first router is used to put 50
customers onto the LAN in the PoP and the other router is used to take traffic off of the LAN and onto
the backbone. The problem arises when this example is scaled up to a typical PoP where there are 500
T1 customers on 10 access routers along with maybe 3 backbone routers. The cost inversion is due to
the expensive FDDI interfaces on each of the 13 routers that don't generate any revenue, but are
required to connect all of these customer T1s onto backbone circuits. It is easy to conclude that we
need bigger, faster routers, but today's selection, namely the Cisco GSR and the Ascend GRF, are
designed and priced more for core backbone applications and are not economical for access line
aggregation.
This leads to the first role of ATM in service provider networks to date: aggregating medium speed
access circuits in the same manner that frame relay switches are utilized, but at higher bandwidths. In
most large PoPs today, backbone routers and edge ATM switches (Ascend CBX 500, Cisco BPX, and
Newbridge 36170) are used to switch between access DS3s and backbone DS3s. While these are the
switches that in the past have been used for providing transport in the core, this generation of ATM gear
is now being pushed to the edge of service provider networks and retrofitted with various revenue-generating
service interfaces in order to create multiservice ATM switches. In this sense, multiservice
means that the switch supports frame relay, low and high speed ATM, IP, and other access services
such as circuit emulation for voice. For these applications, ATM becomes one more access technology,
but at the same time remains as the switching layer that multiplexes all sub-OC3 rate traffic onto OC3
and OC12 circuits.
Along with the use of frame relay switches and edge ATM switches as service multiplexors, there is a
need for ATM-based integrated access platforms -- typically called multiservice concentrators. These
switches (ADC Kentrox (ADCT 30 5/16), Ascend Sahara, and Newbridge 36140) are just beginning to
ramp as they encounter the strong demand for a switch that will cost-effectively terminate hundreds of
DS1s while supporting any revenue-generating carrier service from sub-DS1 to DS3 rates. Because
ASIC-assisted routing and the ability to `speak' IP routing protocols will be standard features on this
class of device, the need for a stand-alone router is diminshed, and a carrier's service provisioning costs
are lowered. Going forward, ATM-based multiservice concentrators will sit at the very edge of a
2 BancAmerica Robertson Stephens
service provider network as the first level of access into the provider's frame relay switches and
multiservice ATM switches.
In the last few years, the second role of ATM has been to create a backbone between access PoPs in
order to eliminate the need for meshes of physical circuits. Instead, by trunking the hundreds of access
routers and switches in a carrier network with meshes of ATM Virtual Paths (VPs) and Virtual Circuits
(VCs), strong economies are created. Because ATM was
implemented early on in carrier networks for the basic purpose of dumb transport, the first and second
generation ATM switches that were deployed in 1995,1996 and early 1997 were
typically only utilized as simple VP and VC multiplexors and not for most of the higher value
capabilities they supported. Typically, almost all customer traffic - data or otherwise - was mapped
into the Constant Bit Rate (CBR) service layer; very little service level differentiation and ATM QoS -
certainly some of the strongest arguments for the use of ATM - were used. The use of ATM as a
transport mechanism for, in particular, frame relay-based LAN traffic and IP-based Internet traffic is
simply a matter of price/performance and economies of scale. There is no other cost-effective method
of switching traffic from low speed access DS1s up to OC48 trunks in the core.
As ATM, especially the software side, has matured and new revisions of code have been spun-out, the
use of carrier class ATM switches has evolved, shifting the dynamics of ATM demand and driving its
use in service provider networks. First and foremost, the wildly successful use of frame relay for
business access - especially at DS1 and below speeds - continues to drive ATM into the core, but this
use of ATM is no longer just for basic CBR transport. Instead, there is a visible trend in carrier
backbones towards the use of Variable Bit Rate (VBR) and Unspecified Bit Rate (UBR), thereby
allowing the network to treat different types of traffic with different priorities and service levels and
allowing carriers to derive more value from their ATM backbones. Most of the carriers we talked to
were running about 60% to 70% of their traffic in CBR PVCs with the rest split between the other two
service classes. Typically, Internet traffic and some LAN data are mapped into the UBR class, more
important LAN traffic is run in VBR, and real-time traffic requiring fixed latency such as voice and
video is run in CBR mode. Furthermore, carriers are tariffing VBR PVCs at a lower aggregate rate than
CBR in order to push customers into this service level so that additional statistical gain can be gained
from bandwidth allocation in an ATM backbone. Two years ago, CBR would have been used for
video, now carriers are using real-time VBR for the same application. We just talked to one large
CLEC with 150 Ascend ATM and frame relay switches in its network that planned in the next couple of
months to run every one of its service offerings over a common ATM switching layer - even toll voice
calls - using the QoS capabilities of ATM. This is not an uncommon plan; as competition grows and
customers demand more bandwidth for both traditional data services as well as newer data, voice, and
video services, service providers of all sizes are finding that the only way to build a cost-effective
common switching layer is with the use of ATM switches.
Further driving the usage of ATM is the maturation of frame relay to ATM network interworking and
the recent trend to service interworking. Through the use of network interworking software on an ATM
switch, service providers can trunk together two frame relay sites while maintaining a transparent ATM
core between the end sites. This is a mature and well understood technology. More recent is the
availability of service interworking software by ATM switch vendors. With this technology, a carrier
can provision a service to multi-site business where small branch offices connect via frame relay and
larger sites connect via ATM - all to one ATM `cloud' offered by the carrier. We expect to see a
strong ramp in the use of service interworking as it allows revenue generation off an ATM core without
the need for much, if any, native ATM access. The usage of service interworking is strategically
important for ATM switch
vendors and carriers since ATM is no longer just a transport vehicle in the core, but is instead an
enabler of revenue generation for owners of an ATM backbone. This is why even IP-centric carriers
BancAmerica Robertson Stephens 3
such as Qwest (QWST $37) and Level 3 (LVLT $68) are now building ATM and frame relay networks
(both are based on Ascend ATM and Frame Relay switches) - not just Packet over SONET (PoS)
backbones using routers such as Cisco's GSR12000. If they want to generate any kind of revenue
outside of their bandwidth wholesaling operations, they have to offer both frame relay and ATM
services to their business customers. And as there is no economical way of tunneling frame relay and
ATM through a PoS backbone, carriers have to build ATM cores to offer these types of services.
Looking ahead, we believe businesses no longer want to worry about picking different technologies and
transport services from the portfolio of services that are offered. Instead of deciding between native
ATM, frame relay or maybe IP/PPP for transport (and between such issues as PVCs or SVCs, for
example), all the customer should see is a Service Level Agreement (SLA). The SLA will simply
describe the bandwidth, latency, and price of the connection into the carrier "cloud", but will make no
mention of ATM or IP or frame relay. We believe that SLAs such as the one just described will
typically be a nicely packaged Transparent LAN Service (TLS) based on ATM. We believe businesses
don't really want to buy a DS1 or a DS3-based data service - these access speeds do not match their
LAN speeds - instead they want to buy a service that very simply connects their LANs together at the
same speed as their LANs. With the use of ATM, a carrier could offer LAN connectivity at LAN
speeds and at the same time offer voice and video services over the same connection.
Even though many of our readers may now accept that ATM is currently the network of choice among
service providers and that going forward the use of ATM will bring more and more value to network
operators, there are those who will point to 1999 as the year that IP becomes the dominating network
protocol and therefore more PoS routers than ATM switches begin to be installed into carrier networks.
Here's why we do not think so and why we are strong believers in the long-term picture for ATM
switch manufactures that can adapt to the events we describe below. There are two very strong trends
that we see developing in the vendor community. First, in the IP world (particularly in the IETF) there
is a movement to make IP more connection-oriented and more QoS-aware (in other words, more like
frame and ATM). Second, in the switching world, ATM manufacturers are moving "down" the OSI
stack by building traditional SONET transmission functionality into their ATM switches.
Vendors are feverishly working on two IP-centric initiatives for the next year or so: Multiprotocol
Label Switching (MPLS) and Differentiated Services (DiffServ). What do these technologies do and
what are the implications? DiffServ is a mechanism for building tiered services by enabling three
service classes: today's best effort, assured access, and premium service. MPLS (which will be based
on Ascend's IP Navigator and Cisco's TAG Switching) is a methodology for assigning and distributing
labels for each destination address in an IP routing table. Essentially, the label is an index which points
to a table entry and can be passed to other routers so that the receiver does not have to look up an
address - it knows it as part of the label. This is all very important, because each of these labels
represent the entrance to what are essentially IP-based virtual circuits. Basically, whereas in the ATM
world a cell `sees' a number of circuits at the outgoing port and in the traditional IP world a packet
"sees" just the interface and a destination address, now the IP packet sees a "slot," which is a
combination of the destination address and a QoS level - essentially something that looks very much
like ATM VCs. The result is that through the use of MPLS, it is possible to run ATM or frame relay on
top of IP and not just IP on top of a switching layer. Through the use of these technologies it becomes
much easier to build a hybrid ATM switch/IP router that does not care if it's routing IP packets or
switching ATM VPIs/VCIs, and so allows networks to be built with IP over ATM or ATM over IP.
This is the direction that leading vendors are moving. Ascend already has an IP Navigator card for the
CBX500 and by the end of the year will ship a card for the GX550 that will have an OC48 IP port on it
which runs IP Navigator. This is the first step to the hybrid switch just described. In addition, there are
three startups working on similar developments: Argon Networks, NetCore, and Nexabit. So while
these IP add-ons do enhance the value of a PoS network, it will take a few years for the software to
4 BancAmerica Robertson Stephens
stabilize and get to the level of readiness and maturation that ATM is at today. Meanwhile, leading
ATM vendors will simply add these enhancements to their ATM switches - offering both on one
platform. In the future, carriers want to lower the potential for trouble in their networks and one of the
easiest ways to do this is to lower the number of separate boxes - essentially consolidate technologies
into one platform. We talked to one carrier who currently has one IP engineer for each router in their
network, but only had 4 ATM engineers for the 38 installed ATM switches - it becomes much easier to
cut costs, in our view, when one box functions as the ATM switch and the IP router.
Continuing on the theme of technology consolidation, the second trend we see occurring in 1998 and
1999 is ATM switch vendors implementing SONET capability into their switches. Already, there are a
few vendors who have added SONET Automatic Protection Switching (APS) protection and Ascend's
GX550 is shipping with SONET add/drop functionality built into it, eliminating the need for a stand-alone
Add-Drop Mux (ADM). Throughout 1998 we expect to see additional SONET capabilities, first
linear SONET then later ring-based SONET, migrated into ATM switches. While carriers that sell
circuits at less than OC3 rates will continue to require the add/drop and cross-connect functions, we
believe these will be services offered by a SONET interface on leading ATM switches and not via a
stand-alone cross-connect or add-drop multiplexor.
Furthermore, there are a number of vendors (mainly start-ups such as Atmosphere, Omnia, Mainsale,
and Pipelinks) who are designing hardware for the access segment of Metropolitan-Area Networks
(MANs) that combine a SONET ADM with a CPE IP access router and/or a CPE ATM access switch.
By combining networking technologies with the add/drop function, the need for TDM is eliminated.
TDM is a great solution for DS0 voice, but makes little sense for bursty data. With ATM or IP
replacing the TDM structure in SONET, there is more efficient bandwidth utilization from the SONET
metro ring and it is possible for the CLEC or RBOC to realize statistical bandwidth gain.
Essentially, we believe that over time SONET will split into two layers - the network layer and the link
layer. ATM switches and IP routers will continue to use SONET OC-x interfaces as a network port, but
the SONET network layer - the TDM structure and DS0 channelization - is not required in a world
where data traffic dominates over voice traffic. Voice circuits and leased lines require TDM, but IP
requires a fat pipe - not a channelized pipe. Therefore, traditional voice networks will continue to run
over SONET ADMs and cross-connects, but future data networks will simply run ATM switches and IP
routers - eventually a hybrid of the two - directly connected to WDM multiplexors. It is quite
interesting that while ATM vendors are moving down the OSI stack with integrated SONET
functionality, SONET vendors are attempting to do the opposite - integrate ATM switches into their
cross-connects and voice switches. For example, Tellabs (TLAB $72 1/8) has announced their
intentions to build an ATM matrix into future versions of their Titan 5500 product.
In summary, we would like to offer a few of the conclusions that we are drawing from the above
discussion. Currently, IP-centric vendors argue for the deployment of PoS technology via IP routers
and ATM-centric vendors advocate for build-out of ATM backbones that run IP as one service on top
of the switching layer. While it is true that there is a visible trend towards IP-based data services and
therefore why in theory it may make sense to some to eliminate the switching layer, here's why we
continue to anticipate major deployments of ATM switches in the foreseeable future. First, businesses
will continue to demand ATM and frame relay services to connect their offices together and there is no
way of tunneling frame and ATM PVCs through a PoS backbone - the best example of this is both
Qwest's and Level 3's decisions to deploy an ATM and frame relay switching layer. Second, with IP
becoming more connection oriented and more like ATM, it is starting to pick all of the complexities of
ATM that IP advocates use as an argument against ATM. Because of this, while ATM is currently
witnessing the third generation of ATM hardware and software, it will take a couple of years for these
new IP technologies to get to where ATM is today. Additionally, in this same time frame we anticipate
that ATM switch vendors will incorporate IP routing engines and PoS ports in their ATM switches -
leading to the hybrid switch described above. Finally, switching vendors are already offering limited
built-in SONET capabilities, and we expect to see more announcements in this space throughout the
year. As such, the overarching theme we are closely following is the consolidation of hardware and
software which run at layers 1, 2, and 3 of the OSI stack combining into one switching platform. We
believe the vendors that dominate today on each plane will build this box and will continue to drive
network build-outs of the future.
Finally, we offer one last thought. Due to what we have described in this report, for facilities-based
carriers the costs to provide data services are falling. At the same time, demand for bandwidth and new
differentiated services is so strong that in many cases prices are rising. What could be better for the
equipment vendors who are enabling this?
BancAmerica Robertson Stephens maintains a market in the shares of Ascend Communications,
CIENNA Corp., Cisco Systems and Tellabs and has been a managing or comanaging underwriter for or
has privately placed securities of Ascend Communications within the past three years. |
