P1: JDV
DeNoia WL040/Bidgolio-Vol I WL040-Sample.cls June 20, 2003 17:57 Char Count= 0
PROVIDERS ANDSERVICES 789It is important to remember that “the Internet” is not
really a single entity, but rather an interconnected set of
autonomous networks whose owners have agreed to co-
operate and use a common set of standards to ensure in-
teroperability.Peeringis a form of interconnection where
ISPs agree to exchange traffic for their respective cus-
tomers, based on a specific set of business terms. Peering
points are where the networks actually connect to effect
this exchange. The number and location of peering points
and partners is decided by each ISP according to customer
demand and its own business criteria. Subscribers may
need to be aware of these agreements in order to under-
stand fully the performance they can expect end to end
across the Internet.
Just as the background and emphasis of traditional
voice and traditional data service providers differ, so do
their business models and their choices of technology.
Some offer only transport for traffic, either between sub-
scriber sites or to the Internet. Others offer access to ap-
plications or management services. Local telecommuni-
cation carriers tend to offer MAN services over an ATM
and SONET infrastructure, while data providers would
be more likely to offer IP services or simply Ethernet ac-
cess and transport. Cable television and wireless service
providers also offer access services according to the char-
acteristics of their infrastructure technologies. The op-
tions available will likely continue to grow as technology
progresses.Class of Service, Quality of Service
As interest in carrying multimedia or multiple-service
traffic (i.e., voice, data, video) over MANs and WANs
has grown, managing the traffic to provide perfor-
mance appropriate to each application has become more
important. Quality of service techniques are expected to
guarantee performance and delivery, usually in terms of
bandwidth allocation, timeliness of delivery, and minimal
variation in delay (e.g., ATM service categories). Class of
service (CoS) techniques do not make such guarantees,
but rather attempt to meet user requests on a best-effort
basis. Typically CoS works by grouping together traffic
with similar requirements (e.g., voice or streaming video)
and using a priority queuing system so that switches and
routers forward the traffic accordingly. Connectionless
network services such as IP offer CoS traffic management,
while connection-oriented services such as ATM provide
QoS.
QoS cannot really be guaranteed unless it is available
all the way from end to end of the connection. This creates
a challenge for MAN and WAN environments where mul-
tiple technologies from one or more service providers may
be involved in delivering user traffic, and especially when
the traffic originates or terminates in a LAN of yet another
different technology. Several groups are involved in devel-
oping standard techniques for CoS and QoS. The problem
is making sure that appropriate translation mechanisms
can carry user application requirements across network
and SP boundaries:IEEE 802.1p is a Layer-2 tagging mechanism to specify
priority using 3 bits in the Layer-2 frame header.IETF’s differentiated services (DiffServ) indicates how
packets are to be forwarded using per-hop behavior
(PHB) queuing, or discarded if there is not sufficient
bandwidth to meet performance requirements.
ATM traffic management defines service categories and
traffic classes.Virtual Private Networks
A virtual private network (VPN) is a special service that
amounts to establishing a closed user group capabil-
ity over a shared or public network infrastructure. This
means that access is restricted to authorized users only,
privacy of data content is assured, traffic belonging within
the VPN does not get out or become visible to unautho-
rized users, and outside traffic does not get in. VPNs are
becoming a very attractive way for organizations to re-
duce the cost of private WANs while improving the secu-
rity for traffic that travels over public networks. Where
high-speed MAN and WAN services are available, long-
distance performance can even be kept reasonably close
to what the remote users would experience if they were
directly connected to the LAN. VPNs may also be built to
send traffic across the Internet, with one or more SPs pro-
viding the access links between the Internet and various
geographically dispersed customer sites. Internet VPNs
can be significantly less expensive than the private lines
or networks they replace.Management
The OSI model for network management encompasses
five functional areas: configuration management, perfor-
mance management, fault management, accounting man-
agement, and security management. A MAN or WAN ser-
vice provider must cover these from the perspective of
both operating the entire network effectively and balanc-
ing the needs and expectations of paying customers who
could always choose to take their business elsewhere. Op-
eration must be reliable, there must be sufficient capacity
to meet traffic needs and performance expectations, and
privacy must be maintained not only for the content of
the traffic carried but also for data about the customers.
At the same time, subscribers typically want the ability
to manage the performance and flow of their own traffic
through their allotment of SP resources. SP operation sys-
tems must be capable and sophisticated to meet all these
requirements.
A primary mechanism used to establish and manage
expectations between customers and providers is theser-
vice level agreement(SLA). SLAs are the defining docu-
ments (contracts) that spell out what services and levels
of support will be provided to the customer at a specified
price. Successful SLAs are built on a solid, shared un-
derstanding of business priorities and service impact, for
both the service user and the service provider. Detail about
roles and responsibilities, metrics and reporting, added
cost for incremental services or enhancements, escala-
tion procedures, and change management are just some
of what should be covered in an SLA. Many customers
also build in penalties in case the provider fails to de-
liver services at the level specified in the SLA. This may be