March 1998
By Victor R. Parente, America Online
Packet Over Sonet
Packet Over Sonet: Ringing Up Speed
POS shunts LAN traffic onto the WAN a whole lot faster than ATM
|
|
|
Scalable and super-reliable: Say Sonet and those are the advantages
that spring to mind. They're also the reasons net managers using it as
a foundation for ATM MANs (metropolitan-area ne
tworks) are ordering
rings as fast as carriers can pull fiber. Now think of overhead and
complexity: Those are the problems with using ATM to shunt LAN traffic
onto Sonet WANs.
Since 5 bytes of every 48-byte ATM cell go to overhead, net
managers waste nearly 60 Mbit/s of bandwidth for every OC12 (622
Mbit/s) they order. And everyone knows how hard it is to design,
implement, and manage an ATM infrastructure.
So why not just skip ATM altogether? That's the idea behind POS
(packet over Sonet), a high-speed WAN transport that leaves LAN
traffic in its native format. In a sense POS is simply a serial link
like any other—only much more reliable and a whole lot faster. It also
makes much better use of Sonet capacity, since there's no ATM
overhead. And because it lets net managers take the
circuit-provisioning decisions out of the carrier's hands, it's also
more versatile.
On top of that, net managers don't have to wait to take advantage
of the POS possibilities. More and more carriers
are offering it, and
vendors like Ascend Communications Inc. (Alameda, Calif.), Cisco
Systems Inc. (San Jose, Calif.), and Ipsilon Networks Inc. (Palo Alto,
Calif.) already are rolling out gear. Yes, POS is just as pricy a
proposition as ATM over Sonet. But by eliminating the ATM middleman
(along with all the overhead and management hassles), it's a much
better value.
Sonet From the Top
To get a sense of Sonet (synchronous optical network), it helps to
look at the specifics. The physical-layer transport for fiber optic
nets was conceived by MCI Communications Corp. (Washington, D.C.) and
developed by Bellcore (Morristown, N.J.) in the mid-1980s. Like other
physical-layer transports, it describes transmission speed, line
encoding, and signal multiplexing. It also boasts built-in fault
tolerance: Unlike many other Layer 1 technologies, it can reroute
around failed links. Sonet also was the first WAN transport capable of
carrying both legacy (like DS-0 [56-kbit/s] circuits) and broadband
servic
es (like ISDN and ATM); in fact, ATM was designed around Sonet's
native transport capacity.
Here's how the technology works: Sonet divides a fiber path into
multiple logical channels called tributaries. A tributary's basic unit
of transmission is an STS-1 (synchronous transport signal level 1) or
OC1 (optical carrier level 1) signal. Both operate at 51.84 Mbit/s;
STS describes electrical signals, and OC refers to the same traffic
once it's been converted into optical signals. Sonet also allows
channels to be multiplexed, so an OC12 circuit, for instance, might
carry traffic from four OC3 links. A circuit also can carry a single
channel, in which case the line is said to be concatenated; circuits
are described as OC3c, OC12c, and so on.
|
Table 1: Comparing Capacity
|
Sonet line rates are currently
defined up to OC192 (9.953 Gbit/s)
over a single fiber pair, with OC768 (39.813-Gbit/s) capacity expected
soon (see
Table 1
). With rates like these, Sonet can be positioned as
a very scalable transport for any data service.
Almost identical to Sonet is SDH (synchronous digital hierarchy),
defined by the ITU (International Telecommunication Union) as G.707
shortly after ANSI formally ratified the T.105 spec for Sonet.
Although interconnection of Sonet and SDH networks is still relatively
rare, several new transoceanic telecommunications projects make use of
such links, including Flag (Fiber-optic Link Around the Globe, a joint
venture led by Bell Atlantic Corp. [Philadelphia]), Gemini (a joint
venture of Worldcom Inc. [Jackson, Miss.] and Cable and Wireless PLC
[London]), and Atlantic Crossing-1 (whose backers include Deutsche
Telekom AG [Bonn, Germany] and Global Telesystems Group [Vienna,
Va.]).
One of Sonet's key features is APS (automatic protection
switchin
g), which guards against fiber cuts and other link-layer
failures. Sonet rings with APS use redundant strands of fiber: If one
breaks, traffic shifts to another and transmission resumes within
microseconds. A variant called line switching offers even greater
redundancy by routing around failed Sonet equipment. Carriers
provision so-called bidirectional line-switched rings (BLSRs), which
are perhaps the most fault-tolerant WAN services available.
The POSsibilities
ATM has been an attractive option for corporate users ever since
carriers began offering Sonet services in the early 1990s. With
bandwidth of anywhere from 1.5 Mbit/s to 10 Gbit/s (40 Gbit/s in the
future), ATM's appeal lies in its flexibility and scalability. ATM
also allows aggregation of different traffic types—like time-sensitive
voice and video traffic as well as data—over the same network, and it
delivers different levels of QOS for each. Further, there are various
ATM standards defining path routing, signaling, and
switching.
Great—but is there such a thing as being over-spec'd? Debate
over the standards and the complexity of implementing them have
resulted in immature ATM software—especially in terms of
configuration, management, and troubleshooting tools. Emerging
protocols like Lane (LAN emulation) and MPOA (multiprotocol over ATM),
intended to address the complexity of moving switched and routed
traffic over ATM networks, are even less mature. Want to move LAN
traffic over ATM? Start by using Lane to build a Layer 2
network—and then take on all the hardware, software, and
configuration hassles of building a router network. By any standard,
that's not easy.
But POS technology uses familiar link-layer protocols to achieve
the same high bandwidth and scalability as ATM. With POS, IP traffic
(or any other Layer 3 protocol) runs over PPP (point-to-point
protocol) or HDLC (high-level data link control), just as it would
with other types of WAN circuits like leased lines. These link-layer
pr
otocols in turn run directly over Sonet. (Actually, IP over Sonet
uses HDLC framing and PPP's link control protocol. The spec is defined
in IETF RFC 1619.)
Initial POS implementations treat Sonet circuits as unidirectional
rings with only two nodes—and a ring with only two nodes is
really just a point-to-point circuit. Point-to-point links represent
the simplest type of network design—they're either up or they're
down.
And it's a lot easier to run IP over Sonet than IP over ATM. With
POS, net managers assign IP addresses to each interface and define
routing mechanisms like OSPF (open shortest path first) or RIP
(routing information protocol), just as with any other type of IP
network. IP-over-ATM configurations, in contrast, require not only IP
setup but also the definition of ATM VCs (virtual circuits) and
mapping of IP broadcast domains onto these VCs. Since the number of
VCs increases geometrically as switches are added, management and
complexity grow along with the network.
|
MORE INFO
The Art of The Deal: Sonet RFP
|
POS has an even more obvious advantage when it comes to transport
efficiency. Because of ATM overhead, an OC12c circuit delivers only
563.39 Mbit/s of bandwidth, or nearly 59 Mbit/s less than the
circuit's raw capacity—and that's just for the cell header.
There's more overhead for NSAP (network service access point)
addressing, Lane signaling, and whatever type of NNI
(network-to-network interface) signaling is used. All told, ATM eats
up more like 15 percent to 20 percent of bandwidth as overhead.
There's no such loss of capacity with an OC12c POS circuit—it
uses all 622 Mbit/s available.
POS networks also can be just as cost-effective as ATM over Sonet.
ATM backers like to point out how different traffic types
can be
consolidated onto one network, reducing operational costs. As it turns
out, provisioning space- or time-division multiplexed Sonet circuits
rather than cell-switched services for every link saves just as much.
It's also easier to design, implement, and manage separate
networks and consolidated Sonet WANs than it is to build and maintain
an ATM infrastructure. Keeping data, voice, and video circuits
logically separate while taking advantage of Sonet multiplexing makes
efficient use of WAN capacity—and spares net managers from the immature products, complex administration, and hardware and engineering costs of maintaining ATM.
Getting Started With Sonet
Companies design and implement Sonet rings in cooperation with a
LEC (local exchange carrier). LECs typically won't provision Sonet in
the local loop until the customer requires two or more DS-3
(45-Mbit/s) or equivalent circuits. Even there, companies are only
buying circuits, not Sonet rings. For example, a carrier might
use a
Sonet OC3 ring to provision an order for two DS-3 circuits and a
number of T1 and fractional T1 lines. But the ring is
"unprotected"—that is, it doesn't have automatic protection
switching to guard against circuit failure. Further, the company won't
realize the cost savings of consolidating multiple services onto the
same local loop; it's still paying for separate circuits. That means
it's better to ask specifically for Sonet rings when bandwidth demands
dictate.
Customized Sonet rings come in two basic forms. Some carriers lease
the actual ring for a given monthly cost, with additional monthly
charges for each circuit provisioned. Others provision the ring but
sell only the active circuit services on the ring. With both
arrangements, carriers can provide protection switching, monitoring,
and reporting services.
Which is better? It's hard to say. Leasing a customized Sonet ring
becomes cost-effective only after the ring is filled to 50 percent of
capacity (that's because of current
tariffs and that at 50 percent
utilization, at least two circuits can be fitted onto each Sonet
tributary). One advantage of leasing a ring is that net managers can
reprovision circuits on the fly—a 30-second change at a network
management console vs. a two-month wait for a new link from the
carrier.
When selling services rather than the ring itself, carriers count
on the ring to be at least 40 percent utilized to pay for the costs of
building and operation. Carriers may also be willing to negotiate deep
discounts to drum up Sonet business. It ultimately comes down to
bandwidth requirements, current tariff levels, and network managers'
creativity in placing the maximum number of circuits possible on each
Sonet ring.
Designer Rings
Corporate networkers can begin the design process by documenting
all current WAN requirements for voice, data, and video services. The
key issues here are bandwidth requirements, number of locations, and
distance between locations.
|
Figure 1: Candidate for Change
|
To help illustrate Sonet network design, imagine a company that
uses a mix of T1 and T3 circuits to link offices in a metropolitan
area, along with a T3 link to its ISP (Internet service provider) and
56-kbit/s connections to branch offices (see
Figure 1
). Assume that
data circuits are running at full capacity and additional bandwidth is
needed—about a T3 on each link between major sites.
|
Figure 2: The ATM Approach
|
One way to simplify the network while adding capacity would be to
design an ATM network that carries voice and data over
OC1 and OC3c
circuits (see
Figure 2
). With this design, three OC3c circuits would
be required to meet anticipated growth needs between the main
locations. Such a network would also require buying and deploying ATM
switches, integrating voice and data networks, and installing new
PBXs. The LEC still determines the design of the Sonet services, and
it might not offer protection switching for the OC3c circuits.
Further, the T3 link to the ISP and the 56-kbit/s circuits also remain
unprotected.
A customized POS network would deliver the same (or more) capacity,
greater reliability, and better scalability. Assume the POS network is
built around an OC12 ring; then the company could deliver more T3s'
worth of capacity among all major sites—for the same circuit cost
as the ATM network.
|
Figure 3: The POS Payoff
|
Assuming an OC12 (BLSR) is used, there will be 12 STS-1 tributaries
between each node. Such a ring has 622 Mbit/s of capacity between
adjacent nodes. Each node on the ring is an ADM (add/drop multiplexer)
that, as the name suggests, adds and drops traffic onto the STS-1
tributaries (see
Figure 3
). Taking advantage of the ADM function
allows the STS-1 tributary capacity to be reused between each set of
nodes. All told, this ring has an aggregate of 48 STS-1 (2.488 Gbit/s)
of capacity (four times OC12 capacity, or four groups of 12 STS-1
tributaries).
Configured for the hypothetical company, the ring is about
one-sixth full. Pricing structures depend on how the services are
tariffed. For those priced as circuits only, ring utilization doesn't
directly affect the total cost of the network. Higher utilization
means more circuits in use—and the possibility that a better
price per circuit can be negotiate
d. For services tariffed as Sonet
networks, there's a flat rate for the ring plus additional charges for
each circuit provisioned.
Bargaining Tactics
To determine which setup makes the most sense, select the sites
that require enough capacity now and in the next year to justify at
least one OC3. Most carriers require at least one central office (CO)
on each ring, giving telcos not only the ability to manage the ring
from the CO but also to connect circuits from IECs (interexchange
carriers) to the corporate network.
A ring with one CO is a good way to keep costs down. But network
designers concerned about reliability may want to include a second CO
(or even more) for the entrance of the IEC's circuits; thus, there's
no one CO point of failure. Further, setting up two or more COs gives
net managers greater flexibility in acquiring long-haul circuit
capacity.
Based on the network's needs and the services available, write an
RFP. The local RBOC should always be allowed to par
ticipate in this
process, since it's the only LEC that will deliver circuits from any
IEC. Competitive LECs (CLECs) also should be encouraged to
participate; they generally offer lower-priced services, SLAs
(service-level agreements), custom configurations, and more advanced
technologies. Other LECs may choose not to carry circuits on the ring
from competing IECs.
Even if carriers don't offer the kind of services desired, net
managers still should ask for exactly what they want. Most carriers
have made very substantial investments to build Sonet services, and as
a rule they desperately need more customers. After all, they know it's
better to customize for a new network than to leave fiber sitting in
the ground unused. Still, carrier proposals in response to the RFP
will depend in part on the amount of business at stake and the degree
to which the carrier's current services meet the RFP's
requirements.
In evaluating the total costs, be creative in negotiating and keep
in mind all the costs inv
olved. It may be possible to negotiate a deal
that includes savings on long-haul circuits. For larger networks,
consider getting multiple smaller rings from different providers in
the same areas. This gives net managers more flexibility in ordering
circuits and greater protection against failures.
Victor R. Parente is network architect for America Online
(Vienna, Va.). His e-mail address is network192@aol.com
|
|
|
 |
|
|
