Hi lml and threaders:
Something from Robertson Stephens - Networking Hardware Research Group
Subject: The First Mile - Release 2.3 - GUIDE to BUILDING A Next
Generation Network in 1999
Date: Wed, 26 May 1999 13:17:49 -0700
The First Mile - Release 2.3
"Every journey starts with the first mile." - Anonymous
GUIDE to BUILDING A Next Generation Network in 1999
Introduction
We have found that perhaps the easiest way to understand the
intricacies of future communications networks is to assume we are a 'greenfield' network provider and then explain how we would build our network this year and upgrade it next year. To put next generation transport networks into perspective, we will first summarize our current thesis on the build-out of data-optimized service provider networks and then delve into how we believe the traditional SONET and DWDM equipment layers will evolve over time.
For those of our readers who care more about stocks and less about such things as STS-1 provisioning models, our general conclusion is that the best way to play transport networking from the systems OEM level is via some of the start-ups that should be having IPOs in late 1999. But we also believe
that some of today's established component vendors make for incredibly
compelling investments. We are especially enthusiastic about the
opportunity
available to E-Tek (ETEK $44, NR), JDS Fitel (TSE:JDS $91, NR), SDL
(SDLI
$120, NR), and Uniphase (UNPH $127 NR).
Investment Thesis
In 1998 we witnessed a banner year for telecom equipment, and in 1999
we are
in the midst of one once again. Service providers of all shapes and
sizes
have recognized the tremendous growth in demand for wide-area bandwidth
as
well as the shift toward a data-oriented service mix with announcements
of
not only new network deployments, but new service provider business
models
as well. Qwest started last year with a planned 19,000 miles network
and now
has approximately 80% of that fiber lit, to be completed in May. (Qwest
buried two conduits along their rights-of-way: one conduit has 96
fibers-48
are for Qwest and 48 are sold wholesale-and the other conduit is left
empty.) Williams announced $2.7 billion in network spending on a new
20,000
miles network and just recently upped the ante to $4.7 billion and
26,000
miles-most of which will be lit by the end of 1999. IXC has finished
lighting 9300 route miles and will be at 17,000 by year-end. UUnet,
with
900,000 aggregate ports in 1998, estimates it will have 50 million
deployed
in five years (includes modem ports, switch virtual interfaces and
router
sub-interfaces).
Interestingly, while MCI, AT&T, and Sprint worry about driving the cost
of a
DS3 mile, the above carriers are examples of a new breed which are
instead
focusing on the cost of a STSc mile-the new unit of currency in the
world of
bandwidth. (An STS is the lowest common denominator in the SONET
synchronous
hierarchy and represents one time slot (frame) of capacity or 51.84
Mbps. A
STSc represents the concatenation of multiple STS-1 time slots and so
an
STS-3c, the electrical equivalent of an optical OC-3c, represents 155
Mbps
of capacity).
Not only was the supply-side of the bandwidth equation growing in leaps
and
bounds, but demand wasn't too shabby either: Frontier Globalcenter is
witnessing growth of 3.5% per week and doubles in offered load every 3
to 4
months. GTE Internetworking, which provides modems for AOL, is
currently
upgrading from 250,000 to 400,000 modem ports for them. UUnet deploys a
new
DS3 customer at the rate of 3 per day and new OC3 PoS customers at the
rate
of 7-10 per month. Williams reports strong demand for wholesale OC48
circuits from ISP customers. Finally, with Qwest's new pricing
structure the
cross-over between DS1 frame relay and a DS3 ATM UNI is now 2x-$1,595
versus
$3,190-and down from a more traditional 5x a year ago, meaning that a
45Mbps
connection can be had for not much more than a 1.5Mbps connection.
All of this activity in 1998 led to a slew of Request For Quotes and
Request
For Information (RFQ/RFI) by service providers for equipment within a
couple
distinct segments of their networks. In particular, for Layer 2
switching we
witnessed almost every major carrier award and begin to deploy core and
edge
ATM switches, primarily the Ascend Gx550/500 combination. We now
believe
that the majority of large ATM vendor decisions are done and going
forward
we will see both continued execution on these awards as well as a whole
host
of new core switching RFPs, but with a twist: they will be for IP/MPLS
solutions. As we described in detail in our October 1998 report on IP,
we
believe that with the advent of IP traffic engineering and DiffServ
service
classes, MPLS will be the Layer 2 of choice going forward and that
traditional ATM solutions will be pushed to the edge of networks. We
think
that popular applications of ATM will be for DSLAM aggregation and in
next-generation ADMs for building VP rings.
For instance, UUnet, which is probably a pretty good leading indicator
of
cutting-edge network deployment, and which currently lays claim to the
largest ATM network in the world, will in 1999 upgrade to an MPLS core
running on Cisco and Juniper routers. UUnet will push its current ATM
switches to the edge in order to re-deploy ATM as an access
technology-replacing TDM and deriving statistical gain at the earliest
entry
point into the network. With an IP/MPLS over SONET core, UUnet gets the
double benefit of better scaling as well as improved coordination of
protection and restoration.
IP/MPLS Core Networks
The IP/MPLS trend began to pick up steam in late 1998 with the
availability
of a Cisco alternative, namely the Juniper M40, as well as increased
RFP
activity. Telstra, the Australian PTT, recently put out a huge RFP for
IP
equipment. Closer to home, the AT&T and British Telecom joint venture
will
be based on a common IP core network. With Juniper's M40 shipping for
volume
revenue and other router developments slated to ship for alpha or beta
testing in the first half of 1999, service providers such as both IXC
and
Williams Communications are now exploring router bake-offs to weed out
the
hype surrounding next generation IP platforms. As they 'kick the
tires', we
believe that despite the marketing propaganda, terabits of capacity in
the
near term is probably over-kill in most situations. We believe that
what is
needed in 1999 is 8 or so ports of OC48 and what will be needed in 2000
will
be 16 ports of OC192. If one looks to router development over the last
decade, the trend has been for each router port to go faster, but the
number
of ports per platform has stayed approximately the same.
Furthermore, if one looks to the state of network deployment, we saw
the
first OC48 IP circuits go up in late 1998 and anticipate N x OC48 (with
N
less than 4) in 1999-easily supported by a scaleable line-speed gigabit
router. If a router has dozens to hundreds of ports across multiple
chassis,
it has to perform bandwidth management functionality that we believe
carriers will leave to their transport platforms. In other words, a
router
with terabits of aggregate capacity would essentially route across
wavelengths and thus would serve to eliminate the need for a separate
device
to perform backbone grooming and aggregation. We do not see even
IP-centric
carriers such as Level 3 communications as willing to rely on a router
to do
this kind of bandwidth management in the backbone. Keep in mind that
although UUnet expects to deploy over the next few years hundreds of
OC48s
to connect each of their major metro areas to their backbone; in a CO
or
PoP, 80% of the traffic is pass-through and does not need to transit a
router-only the approximately 20% which is dropped from the node does.
To read the rest of this research report, please navigate your web
browser
to our web page:
nextgenerationnetworks.com
<http://www.nextgenerationnetworks.com/first_mile/first_mile_archive.htm>
|
