MORE INFO
ATM Advocates
|
Plenty of promise, a paucity of choices. After years of false
starts and volumes of marketing hype, that was pretty much the story
on ATM services. Now they're finally here, and PTTs and emerging
operators are wasting no time making up for lost time. They're
publicizing pricing. They're trumpeting their offerings. They're
churning out enough information to overwhelm even the most intrepid
network manager. Why not let Data Comm sweat the details? To help
corporate networkers negotiate the ATM service selection process, we
issued our own RFP (request for proposal). Designed in close
cooperation with Nokia Telecommunications (Espoo, Finland), which runs
an ATM net in Finland, it calls for a six-site national network. And
because there's more to choosing an ATM service than price and
possible configuration, we also consulted with a couple of net
managers who know t
he technology inside and out: Kari Makela, manager
of network services at Nokia Telecommunications, and Kalle Helameri,
manager of automated data processing operations at Kela (Helsinki),
the Finnish social security administration. Each is a seasoned
networker with thousands of man hours of experience designing,
running, and building multisite ATM networks—and each was ready
with recommendations and warnings that can further help network
managers (see "
ATM Advocates
").
|
|
In the end, we learned a lot about building an ATM net, not the
least of which is that ATM services can be cheaper than leased
lines—as much as 35 percent cheaper, depending on a company's
traffic profiles. But as we investigated further, we also found that
pricing is all over the map. There are other differences, as well,
ranging from availability issues to the handling of data bursts to
SLAs (service-level agreements). All of which proves that when it
comes t
o picking an ATM service, it helps to know what to look
for.
The Network
We asked more than 20 operators to price out a network that
reflected Nokia's own needs. It comprised six sites, linking four
remote offices with two data centers via identical PVCs (permanent
virtual circuits) (see
Figure 1
). The circuits could sustain traffic
at 2 or 4 Mbit/s, with bursts up to 5 or 10 Mbit/s, for a theoretical
total of 30 Mbit/s at each data center. Servers in the data center ran
a mix of Web, telnet, and FTP (file transfer protocol) apps, as well
as Notes from Lotus Development Corp. (Cambridge, Mass.). They also
hosted client-server databases.
|
Figure 1: The ATM Configuration
|
Pricing covered installation and monthly charges for a three-year
period.
Overall costs reflected compliance with three SLAs: Monthly
backbone availability of at least 99.8 percent; around-the-clock
customer support; and four-hour response time to network problems.
Discounts typically offered by the operator would also have to be
reflected. Finally, each office would have to handle current traffic
loads but have plenty of room for growth. So the data centers were
connected to the ATM network via 155-Mbit/s links (15 kilometers
long), and the remote offices via 34-Mbit/s lines (5 or 30 kilometers
long).
Helameri recommends making other stipulations—management and
reporting requirements chief among them. "One should really ask about
visibility into the operator network," he says, "like the ability to
read the MIBs [management information bases] from the access devices.
It doesn't really hurt if operators' devices would provide some alerts
as well." And some networkers may also want to work in the costs of
pulling fiber and renting CPE (customer premises equipment), someth
ing
our RFP didn't cover.
The Bids
In the end, only seven operators came up with bids for commercially
viable ATM services. Some of those that declined to
participate—like Deutsche Telekom AG (Bonn, Germany)—said
the survey was too general, or that customer needs are so unique they
can only be evaluated case by case.
|
MORE INFO
The Money Matters
|
Nevertheless, there were enough responses to make a couple of
things clear—both having to do with price. First, ATM can be a
lot cheaper than a leased-line equivalent. Leased-line proposals from
Cable & Wireless Communications Ltd. (C&W, London) and Telia AB
(Farsta, Sweden), for instance, were 25 percent to 35 percent more
expensive (see "
The Money Matters
"
).
|
|
Second, pricing for ATM services can vary widely. While the bid
from Telecom Italia (Rome), for example, topped US$6 million, the one
from Telia was only US$775,455 (see
Table 1
). What's behind the
variety? Simple: Different operators charge differently for different
things. Take monthly charges, which account by far for the bulk of the
project cost. Even in the case of British Telecommunications PLC (BT,
London), whose installation charge of US$363,880 was the highest we
saw, the monthly charge still represented 78 percent of the total
cost.
|
BUILD YOUR OWN CUSTOM TABLE
Table 1: Carrier Response to the RFP
|
And sometimes those monthly charges refl
ect access line costs. Take
Telecom Italia, whose bid included the highest monthly charges:
US$162,893, or nearly eight times those of Telecom Finland Ltd.
(Helsinki, Finland), whose at US$10,762 were the lowest. Italy has
long been known for having some of the highest access line charges
around—and 75 percent of Telecom Italia's monthly charge is
related to access line fees.
There also were dramatic differences in setting up those access
lines. The installation charges of BT and C&W were nearly 35 times
higher than those of Telecom Finland. That's because the U.K. carriers
assumed fiber would have to be pulled to the customer premises. Telia
and Telecom Finland didn't factor in fiber pulls—and for good
reason, according to Makela. "A large portion of the buildings in
Sweden and Finland with more than 500 workers already have fiber
pulled to the premises," he says.
And even if fiber is installed, there can be big differences in
what operators charge. Telecom Finland, for instance, calc
ulated its
bid by running native ATM over a single pair of access fibers. There
was no low-level SDH (Synchronous Data Hierarchy) transmission
equipment involved—unlike the bids of BT and C&W. These operators
not only included the cost of low-level SDH gear, but also they used
twice as much fiber as Telecom Finland.
Factors To Be Figured
Of course, other factors can skew operators' bids. Take distance,
performance parameters, and such intangibles as the customer's
relationship with the operator.
Start with distance. With access lines accounting for so large a
part of the price quote, even small changes in lengths could
dramatically increase or decrease the cost of the overall network. In
Finland and Sweden, for instance, ATM is pretty extensive—which
means customer sites are relatively close to a switch, often within 5
km. And this proximity is reflected in the pricing: Reducing access
line lengths from 30 km to 10 km knocks 65 percent off Telia's monthly
charges and 35 perce
nt off the total project cost. But in the U.K.,
distance plays less of a factor: C&W charges a single rate regardless
of length; BT charges US$4,135 per month access lines under 40 km and
US$413.50 per kilometer for anything over 40 km.
The length of the ATM circuit also has big implications. When
offices are located farther from the data center, bids are generally
higher. How much higher depends on the operator and the
characteristics of the circuit. BT, for example, charges US$2,689.46 a
month for a 65-km 4-Mbit/s circuit that can burst up to 10 Mbit/s for
200 cells. But increase the distance to 100 km and expect to pay
US$1,198.30 more per month.
And once that circuit crosses international borders, prices can really jump. Depending on the configuration, for example, BT charges about the same for a half circuit to Norway as it does for a full circuit within the U.K. All operators offer international connections of some kind, except for Telecom Italia—which offers none.
Bit-Rate Re
alities
|
MORE INFO
Want to participate in an RFP?
|
Then there's the type of circuit—which also can boost or
reduce bids. Our RFP called for comparatively inexpensive VBR
(variable bit-rate) circuits to handle LAN-based traffic. If we'd
called for CBR (constant bit rate), bids would have been higher; if
we'd been in the market for UBR (unspecified bit rate), the bids would
likely have been lower.
What makes VBR cost effective? It lets companies pay only for what
they use, which suits applications that generate intermittent bursts
of traffic. VBR services are thus a lot like frame relay: Net managers
can match a VBR circuit's SCR (sustained cell rate) to the amount of
data that normally travels between two sites. Bursts are spec'd as
VBR's PCR (peak cell ra
te) and duration of those bursts as the MBS
(maximum burst size).
CBR circuits, on the other hand, are more like leased lines. A CBR
PVC boasts guaranteed bandwidth and low delay, which makes it suitable
for voice and other time-sensitive apps, but a poor choice for bursty
traffic. So how does that all affect price? BT, for example, offers a
2-Mbit/s VBR connection with unlimited bursting up to 5 Mbit/s for
US$12,028—about the same price as a 2.7-Mbit/s CBR connection and
43 percent less than a 5-Mbit/s CBR link.
Makela says that customers interested in CBR might simply be better
off with leased lines. "The ATM backbone is too complicated to be used
only that way," he says. "Consider ATM if only 40 percent or so of the
traffic is CBR-based."
And just because we didn't look for UBR or ABR (available bit-rate)
services doesn't mean net managers shouldn't. Both are good choices
for carrying LAN traffic, and both might prove to be low-cost
options.
UBR is a best-effort service
—there aren't any guarantees on
delays or minimum bandwidth. Cells are tagged for discard, but when
there's congestion in the network they're dropped. A technique called
early-packet discard can correct this problem: Rather than dropping
cells from different packets, cells from a single packet only are
dropped. This way, only one packet has to be sent again, not an entire
string of them.
And when it comes to UBR, it sometimes helps to have a good
relationship with the carrier—or one that involves a lot of
money. Ask Helameri, whose contract runs between US$5 million and
US$10 million a year. "The argument for UBR is that I have the
influence with the operator so my UBR traffic always gets through," he
says.
And net managers shouldn't lose sight of the fact that UBR can be a
lot cheaper. Consider Telecom Finland, whose bid came in at US$677,160
when UBR was specified—16 percent less than its discounted VBR
offering. "We offer VBR but we don't have any customers using it,"
says Arto U
rataipale, ATM product manager. "They're all using
UBR."
As for ABR, the ATM Forum hails it as the ideal service for
LAN-to-LAN traffic. ABR circuits include a flow-control mechanism, so
they can dynamically adjust transmission rates. Net managers only need
to specify the MCR (minimum cell rate) and PCR. Telecom Italia offers
ABR—but only up to 4 Mbit/s, too low for the requirements of our
RFP.
Price Follows Function
|
MESSAGE SERVER
|
Presented with a wide range of pricing, we suspected there might be
some differences in functionality. And we were right.
Part of the big attraction of VBR circuits is their ability to
handle sporadic data bursts. As with frame relay, operators engineer
their ATM backbones with additional capacity to allow for
oversubscription. T
he more capacity, the better the chance a user will
be able to send data above the SCR.
But as is the case with frame relay, operators don't divulge the
exact amount of additional capacity. Still, users can get a read on
the backbone by looking at the maximum burst size defined for a given
PVC.
Data Comm learned that each operator handles bursts differently.
Most cap bursts at either 50 or 200 cells—and start dropping
cells once that limit is reached.
C&W offers a different twist. It allows customers to burst
indefinitely as long as there's capacity on the ATM network. The cells
sent over the MBS are simply tagged for discard. BT offers the same
feature with its VBR+ service—for a 13 percent premium.
Guaranteed Differences
There also were significant differences in the service-level
guarantees operators included in their bids. Remember, we stipulated
99.8 percent monthly availability and four-hour response time to
problems.
All operators claim that network
availability far exceeds the
requirements outlined in our RFP—and most say they can top 99.9
percent. Yet only Telecom Finland could guarantee 99.8 percent
availability (1.4 hours of downtime) on a monthly basis. Tele Danmark
A/S (Aarhus, Denmark) and Telecom Italia guarantee higher
availability—but not by the month. Tele Danmark sets quarterly
levels, so downtime has to exceed 2.92 hours before the rebates kick
in. Telecom Italia does it by the year—so downtime can reach
13.14 hours before networkers users see any refunds.
|
CONTACT AUTHOR
dgreenfield@data.com
|
Telecom Finland attributes availability to its decision not to use
low-level SDH transmission equipment in the local loop. "The problem
with SDH is that the network management tools and practices are
nonexistent
," says Helameri. The upshot is that "in Telecom Finland's
case, the availability of the SDH infrastructure is worse than that of
the ATM backbone built on top of it."
As for response time, Telia guarantees to be on site and working
within two hours. On the other hand, C&W guaranteed four-hour response
time only if the problem was reported during the normal working day,
or within four hours of the start of the next working day.
David Greenfield is international technology editor at Data Communications International and is based in Jerusalem. He can be reached at
dgreenfield@data.com
.
|
|
