Network Working Group M. Eder
Request for Comments: 3387 H. Chaskar
Category: Informational Nokia
S. Nag
September 2002
Considerations from the Service Management Research Group (SMRG)
on Quality of Service (QoS) in the IP Network
Status of this Memo
This memo provides information for the Internet community. It does
not specify an Internet standard of any kind. Distribution of this
memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2002). All Rights Reserved.
Abstract
The guiding principles in the design of IP network management were
simplicity and no centralized control. The best effort service
paradigm was a result of the original management principles and the
other way around. New methods to distinguish the service given to
one set of packets or flows relative to another are well underway.
However, as IP networks evolve the management approach of the past
may not apply to the Quality of Service (QoS)-capable network
envisioned by some for the future. This document examines some of
the areas of impact that QoS is likely to have on management and look
at some questions that remain to be addressed.
1. Introduction
Simplicity above all else was one of the guiding principles in the
design of IP networks. However, as IP networks evolve, the concept
of service in IP is also evolving, and the strategies of the past may
not apply to the full-service QoS-capable network envisioned by some
for the future. Within the IP community, their exists a good deal of
impetus for the argument that if the promise of IP is to be
fulfilled, networks will need to offer an increasing variety of
services. The definition of these new services in IP has resulted in
a need for reassessment of the current control mechanism utilized by
IP networks. Efforts to provide mechanisms to distinguish the
service given to one set of packets or flows relative to another are
well underway, yet many of the support functions necessary to exploit
these mechanisms are limited in scope and a complete framework is
non-existent. This is complicated by the fact that many of these new
services will also demand some form of billing framework in addition
to a control one, something radically new for IP.
This document intends to evaluate the network and service management
issues that will need to be addressed, if the IP networks of the
future are going to offer more than just the traditional best effort
service in any kind of significant way.
2. Background
The task of defining a management framework for QoS will be difficult
due to the fact that it represents a radical departure from the best
effort service model that was at the core of IP in the past, and had
a clear design strategy to have simplicity take precedence over
everything else [1]. This philosophy was nowhere more apparent than
in the network and service management area for IP [2]. Proposed
changes to support a variety of QoS features will impact the existing
control structure in a very dramatic way. Compounding the problem is
the lack of understanding of what makes up a "service" in IP [3].
Unlike some other network technologies, in IP it does not suffice to
limit the scope of service management simply to end-to-end
connectivity, but the transport service offered to packets and the
way the transport is used must also be covered. QoS management is a
subset of the more general service management. In looking to solve
the QoS management problem it can be useful to understand some of the
issues and limitations of the service management problem. QoS can
not be treated as a standalone entity and will have its management
requirements driven by the general higher level service requirements.
If the available transport services in IP expand, the result will be
the further expansion of what is considered a service. The now
de-facto inclusion of WEB services in the scope of IP service, which
is remarkable given that the WEB did not even exist when IP was first
invented, illustrates this situation well. This phenomenon can be
expected to increase with the current trend towards moving network
decision points towards the boundary of the network and, as a result,
closer to the applications and customers. Additionally, the argument
continues over the need for QoS in IP networks at all. New
technologies based on fiber and wavelength-division multiplexing have
many people convinced that bandwidth will be so inexpensive it is not
going to be necessary to have an explicit control framework for
providing QoS differentiation. However uneconomical it is to
engineer a network for peak usage, a major argument in this debate
certainly is the cost of developing operational support systems for a
QoS network and deploying them in the existing networks. Just the
fact that customers might be willing to pay for additional service
may not be justification for implementing sweeping architectural
changes that could seriously affect the Internet as it is known
today. The IP community must be very concerned that the equality
that characterized the best effort Internet may be sacrificed in
favor of a service that has a completely different business model.
If the core network started to provide services that generated more
revenue, it could easily come at the expense of the less revenue
generating best effort service.
3. IP Management Standardization
Management standardization efforts in the IP community have
traditionally been concerned with what is commonly referred to as
"element management" or "device management". Recently, new efforts
in IP management have added the ability to address service issues and
to look at the network in more abstract terms. These efforts which
included a logical representation of services as well as the
representation of resources in the network, combined with the notion
of a user of a service, has made possible the much talked about
concept of 'policy'. Notable among these efforts are the Policy work
in the IETF and the DMTF work on CIM and DEN. Crucial elements of
the service management framework are coming into perspective, but
point to a trend in IP that is a quite radical departure from the
control mechanisms of the past. As the service model evolves from
being what was sufficient to support best effort to being able to
support variable levels of service, a trend towards a centralized
management architecture has become quite apparent.
This is becoming increasingly apparent for two reasons. QoS
mechanisms need network wide information [4], and for them to
succeed, they must not require a tremendous amount of support from
the core network. It is becoming increasingly accepted that only at
the edge of the network will there be sufficient resources to provide
the mechanisms necessary to admit and control various QoS flows.
A question often asked these days is if "the architectural benefits
of providing services in the middle of the network outweigh the
architectural costs"[5]. This same question should be asked of
service management. As new network elements are needed to support
service management, even if they are not contributing directly to the
forwarding of packets, the cost both in the increased complexity and
the possibility of destabilizing the networks needs to be considered.
An analyses of this issue will be made by the SMRG when we start to
look more in detail at some of the issues raised in this survey
document.
4. Telecommunications Service Management
One place to start an effort to define service management in IP
networks is by looking at what has been done previously in
telecommunications networks. The telecommunications standards for a
service management framework have not received wide scale acceptance
even in an environment in which the service is fairly constrained.
Many proprietary protocols still dominate in the market even though
regulation has made it necessary for network operators to open their
networks sufficiently to allow for multiple vendor participation in
providing the service. This indicates that some formalized
boundaries exist or the markets are sufficiently large to justify the
development of interfaces. International telecommunications
management standards look at the complete management problem by
dividing it into separate but highly related layers. Much of the
terminology used to describe the management problem in IP has
diffused from the telecommunications standards [6]. These standards
were designed specifically to address telecommunications networks and
services, and it is not clear how applicable they will be to IP
networks. Service management is defined in terms of the set of
services found in telecommunications networks and the management
framework reflects the hierarchical centralized control structure of
these networks. The framework for service management is based on the
Telecommunications Management Network (TMN) layered approach to
management. Current IP standards are heavily weighted towards the
element management layer and especially towards the gathering of
statistical data with a decentralized approach being emphasized. In
the TMN architecture a dependency exists between layers and clear
interfaces at the boundaries are defined. To what extent service
management, as defined in the TMN standards, can be applied to IP
where there would likely be resistance to a requirement to have
formalized interfaces between layers [6] must be further
investigated.
TMN concepts must be applied carefully to IP networks because
fundamental differences exist. Control of IP networks is highly
distributed especially in the network layer. Management is non-
hierarchical and decentralized with many peer-to-peer relationships.
A formal division of management into layers, where management
dependencies exist at the borders of these layers, may not be
applicable to IP. Any effort to define service management in IP must
be constantly vigilant that it does not assume the telecommunications
concepts can be applied directly to IP networks. The most basic
abstraction of the network management problem into element, network,
and service management has its origins in the telecommunications
industry's standardization work and the IP management framework might
not have made even these distinctions if it where not for the
telecommunications legacy.
5. IP Service Management: Problem Statement
In defining the Service Management Framework for IP, the nature of
services that are going to need to be managed must be addressed.
Traditionally network management frameworks consist of two parts, an
informational framework and the framework to distribute information
to the network devices. A very straight forward relationship exists
in that the distribution framework must support the informational
one, but also more subtle relationships exists with what the
informational and distribution frameworks imply about the management
of the system. The informational framework appears to be the easier
problem to address and the one that is principally being focused on
by the IP community.
Efforts like the DMTF CIM are currently trying to define network, and
to a lesser extent service, information models. These efforts show a
surprising similarity to those of the telecommunications industry to
define information models [7]. What has not emerged is a standard
for defining how the information contained in the models is to be
used to manage a network.
The number of elements to be managed in these networks will require
this information to be highly distributed. Highly distributed
directories would be a prime candidate for the information that is of
a static nature. For information that is of a dynamic nature the
problem becomes far more complex and has yet to be satisfactorily
addressed. Policy management is a logical extension of having
distributed directories services available in the network. The IETF
and DMTF are looking to Policy management to be a framework to handle
certain service management issues. Much of the current policy
efforts are focused on access and traffic prioritization within a
particular network element and only for a single administrative
domain [8]. Classifying traffic flows and enforcing policies at the
edge with the intent of focusing on admission issues, without
addressing the end-to-end nature of the problem, leaves some of the
most complex QoS management issues still unanswered. Providing a
verifiable commodity level of service, in IP, will effect every facet
of the network and a management solution to the problem will have to
address the scale and the dynamics by which it operates.
5.1 Common Management Domain
Standardization efforts need to concentrate on the management
problems that are multi-domain in character. The test for multi-
domain often centers around there being a many-to-one or a one-to-
many relationship requiring the involvement of two or more distinct
entities. Domains could reflect the administrative domain, routing
domain, or include agreements between domains. Unlike the
telecommunications network in which traffic traverses only a
relatively small number of domains, traffic in IP networks is likely
to traverse numerous domains under separate administrative control.
Further complicating the situation is, that unlike the
telecommunications network, many of these domains will be highly
competitive in nature, offering and accommodating varying service
level agreements. Telecommunications traffic, even with
deregulation, passes from the access providers network to a core
network and then, if it is an international call, across
international boundaries. The number of domains is relative to IP
small, the service supported in each is virtually identical, and yet
each domains is likely to have a different business model from the
other. In contrast IP will have many domains, many services, and
domains will likely be highly competitive. To be successful IP will
need to model the domain problem in a way that reduces the complexity
that arises from having many independent networks each having a
different service model being responsible for a single flow.
Addressing service management issues across domains that are direct
competitors of each other will also complicate the process because a
solution must not expose too much information about the capabilities
of one domains network to the competitor. Solutions may require a
3rd party trusted by both to provide the needed management functions
while at the same time insuring that sensitive information does not
pass from one to the other.
5.2 Service Management Business Processes
A service management framework must address the business processes
that operate when providing a service. A service can be separated
into two fundamental divisions. The first is the definition of the
service and the second is the embodiment of the service. While this
division may seem intuitive, a formal process that addresses these
two aspects of a service needs to be in place if management of the
service is to be actually realized.
In specifying a service it must be possible to map it onto the
capabilities of the underlying network architecture. The service
needs to be specified in an unambiguous way so that mechanisms can be
put in place to enable the control of the service. It can be a
useful tool to view the relationship of the definition of a service
to an instance of that service to the relationship between the
definition of an object to the instantiation of that object in object
oriented modeling. As networks evolve it is going to be necessary to
logically describe the network capabilities to the service and
because IP networks are so fragmented specific service
classifications will need to be made available that transcend the
individual regions and domains. An interface that defines and
controls the network capabilities, abstracted for the service
perspective, allows for the administration of the network by the
service management systems.
Services are often designed with management capabilities specific to
them. These services have tended to not rely on the service aspects
of the network, but only on its transport capabilities. As services
become more dependent on the network, Management over a shared
framework will be required. Operators have recognized the business
need to allow the user to have as much control over the management of
their own services as possible. IP services will be highly diverse
and customizable further necessitating that the management of the
service be made available to the user to the extent possible.
In the IP environment where they may be many separate entities
required to provide the service this will create a significant
management challenge.
5.3 Billing and Security
Paramount to the success of any service is determining how that
service will be billed. The process by which billing will take place
must be defined at the service inception. It is here that the
network support necessary for billing should be addressed.
Analogously, security must also be addressed in the most early stages
of the service definition. It is not practical to assume that the
billing and the security services will be hosted by the same provider
as the service itself or that it will be possible to have the billing
and security functions specifically designed for every service.
These functions will have to be a generic part of the network.
5.4 Standards
Given the limited success of the telecommunications standards bodies
efforts to formalize the relationship between different management
support functions it is highly suspect that such efforts would
succeed in IP networks which have an even more diverse concept of
network and services. If the IP network is to be made up of peer
domains of equal dominion it will be necessary to have management
functionality that is able to traverse these domains. Of course the
perspective of where management responsibility lies is largely
dependent on the reference point. A centric vantage point indicates
responsibility shared equally among different domains. From within
any particular domain management responsibility exists within that
domain and that domain only. For a management framework to succeed
in IP networks logical management functions will have to be
identified along with an extremely flexible definition language to
define the interface to these management functions. The more the
management functionality will have to cross boundaries of
responsibility, the more the network management functions have to be
distributed throughout the network.
5.5 Core Inter-domain Functions
The service management paradigm for IP must address management from a
perspective that is a combination of technical solutions as well as a
formula for representing vendor business relationships. Currently
services that need support between domains require that the service
level agreements (SLAs) be negotiated between the providers. At some
point these agreements will likely become unmanageable, if the number
of agreements becomes very large and/or the nature of the agreements
is highly variable. This will result in there being sufficient need
for some form of standardization to control these agreements.
Bandwidth Brokers have been conceived as a method for dealing with
many of the problems between the domains relating to traffic from a
business perspective. The premise of the Bandwidth Brokers is to
insure agreement between the network domains with regards to traffic,
but security and billing issues, that are not likely to be as
quantifiable, will also need to be addressed. Service providers have
traditionally been reluctant to use bandwidth broker or SLA types of
functions as they fear such tools expose their weaknesses to
competitors and customers. While this is not a technical problem, it
does pose a real practical problem in managing a service effectively.
Looking at the basic requirements of the QoS network of the future
two competing philosophies become apparent. The network providers
are interested in having more control over the traffic to allow them
to choose what traffic gets priority especially in a congested
environment. Users desire the ability to identify a path that has
the characteristics very similar to a leased line [9]. In either
situation as IP bandwidth goes from being delivered on an equal
basis, to being delivered based on complex formulas, there will
become an increasing need to provide authentication and validation to
verify who gets what service and that they pay for it. This will
include the ability to measure that the service specified is being
provided, to define the exact parameters of the service, and to
verify that only an authorized level of service is being provided.
Some of the earlier work on an architectural framework for mixed
traffic networks has suggested that bilateral agreements will be the
only method that will work between administrative domains [10].
Multilateral agreements may indeed be complex to administer, but
bilateral agreements will not scale well and if the traffic needs to
traverse many administrative domains it will be hard to quantify the
end-to-end service being provided. Instability in the ownership and
administration of domains will also limit the usability of bilateral
agreements in predicting end-to-end service.
As the convergence towards all IP continues it will be interesting to
understand what effects existing telecommunications regulations might
have on IP networks as more regulated traffic is carried over them.
Regulation has been used in the telecommunications world to open the
network, but it has had mixed results. A regulated process could
possibly eliminate the effects competitive pressures will have on
bilateral types of agreements and make it possible to get a truly
open environment, but it could also have an opposite effect.
Unfortunately the answer to this question may not come in the form of
the best technical solution but in the politically most acceptable
one. If traffic agreements between the boundaries of networks is not
standardized a continuing consolidation of network providers would
result. Providers unable to induce other providers to pair with them
may not be able to compete if QoS networks become commonplace. This
would be especially visible for small and midsize service providers,
who would be pressured to combine with a larger provider or face not
being able to offer the highest levels of service. If this
phenomenon plays out across international boundaries it is hard to
predict what the final outcome might be.
5.6 Network Services
The majority of current activity on higher level management functions
for IP networks have been restricted to the issue of providing QoS.
Many service issues still remain to be resolved with respect to the
current best effort paradigm and many more can be expected if true
QoS support is realized. Authentication, authorization and
accounting services still inadequate for the existing best effort
service will need additional work to support QoS services.
It is reasonable that services can be classified into application
level services and transport level services. Transport services are
the services that the network provides independent of any
application. These include services such as Packet Forwarding and
Routing, QoS differentiation, Traffic Engineering etc. These might
also include such functions as security (Ipsec) and Directory
services. In IP networks a distinction is often made between QoS
transport services that are viewed as end-to-end (RSVP) or per-hop
(Diffserv). From a management perspective the two are very similar.
Transport level services are not very flexible, requiring application
level services to fit into the transport framework. An application
that needs additional transport level services will need to be a
mass-market application where the investment in new infrastructure
can be justified. Because of the effort in altering transport
services, applications that need new ones will have a longer time to
market and the effort and cost to develop a framework necessary to
support new transport services should not be underestimated.
Application level services are those specific to the application.
Many service management functions occur between the application
supplier and the application consumer which require no knowledge or
support by the existing network. By keeping service management
functions at this level time to market and costs can be greatly
reduced. The disadvantages are that many applications need the same
functionality causing inefficient use of the network resources.
Services supplied by the network are able to be built more robustly
and can provide additional functionality, by virtue of having access
to information that applications can not, providing additional
benefit over application level services. An example of an
application level service that could benefit from a Network service
is the AAA paradigm for Web based E-Commerce, which is largely
restricted to user input of credit card information. Sometimes
application level service requirements have the disadvantages of both
transport service and application service level. For instance, in IP
telephony, this may include services provided by a gateway or other
network device specific to IP telephony to support such services as
call forwarding or call waiting. The mass appeal of IP telephony
makes it possible to suggest considerable infrastructure changes, but
the nature of this kind of change has contributed to the slow
penetration of IP telephony applications.
6. The Way to a QoS Management Architecture
An overview of some of the problems in the previous sections shows a
need for a consolidated framework. Transport level QoS will demand
traffic engineering that has a view of the complete network that is
far more comprehensive than what is currently available via the
Routing protocols. This view will need to including dynamic network
congestion information as well as connectivity information. The
current existing best-effort transport control may become more of a
hindrance to new services and may be of questionable value if the IP
network will truly become a full service QoS network. Both IntServ
and DiffServ QoS schemes require network provisioning to adequately
support QoS within a particular domain and agreements for traffic
traversing domains. Policy management, object oriented information
models, and domain gateways are leading to a more centralized
management structure that provides full service across domains and
throughout the network. Given the probable cost and complexity of
such a system failure to come up with a standard, even if it is a de
facto one, will have serious implications for the Internet in the
future.
6.1 Point to Point QoS
For the current trends in QoS to succeed, there will need to be
harmonization across the new and existing control structures. By
utilizing a structure very similar to the existing routing control
structures, it should be possible develop functionality, not in the
data path, that can allocate traffic within a domain and use inter-
domain signaling to distribute between domains. Additional
functionality, necessary to support QoS-like authorization and
authentication functions for edge devices admitting QoS traffic and
administering and allocating traffic between administrative domains
could also be supported. While meeting the requirements for a
bandwidth broker network element [10], additional functionality of
making more general policy decisions and QoS routing could also be
performed. Given that these tasks are interrelated it makes sense to
integrate them if possible.
The new service architecture must allocate traffic within a
particular administrative domain and signal traffic requirements
across domains, while at the same time it must be compatible with the
current method for routing traffic. This could be accomplished by
redirecting routing messages to a central function, which would then
calculate paths based on the entire network transport requirements.
Across domains, communication would occur as necessary to establish
and maintain service levels at the gateways. At the edges, devices
would provide traffic information to billing interfaces and verify
that the service level agreed to was being provided. For scalability
any central function would need to be able to be distributed in large
networks. Routing messages, very similar in content to the existing
ones, would provide information sufficient to support the traffic
engineering requirements without changing the basic forwarding
functions of the devices. Having routes computed centrally would
simplify network devices by alleviating them from performing
computationally intensive routing related tasks.
Given the number of flows through the network the core can not know
about individual flow states [11]. At the same time it is not
practical to expect that the edge devices can determine paths that
will optimally utilize the network resources. As the information
needed to forward traffic through the network becomes related to
complex parameters that can not be determined on a per hop basis and
have nothing to do with the forwarding of packets, which routers do
best, it might make sense to move the function of determining routes
to network components specifically designed for the task. In a QoS
network routing decisions will become increasingly dependent on
information not easily discernable from the data that routers could
logically share between themselves. This will necessitate the need
to for additional functionality to determine the routing of data
through the network and further suggests that all the information
needed to allow a router to forward packets might not be better
provided by a network element external to the packet forwarding
functions of a router.
At the edges of the network where the traffic is admitted it will be
necessary to have mechanisms that will insure the traffic is within
the bounds of what has been specified. To achieve this it will be
necessary to buffer and control the input traffic. Second the
traffic would need to be marked so the other network elements are
able to identify that this is preferred traffic without having to
keep flow information. Conversely, a path could be chosen for the
traffic that was dedicated to the level of service being requested
that was per flow based. A combination of the two would be possible
that would allow a reservation of resources that would accommodate
multiple flows. Both methods are similar from a management
perspective and are really identical with regards to route
determination that could be performed centrally in that one method
represents just a virtual path based on the handling of the packets
by the device in the network and the second would be a pre-reserved
path through the network. Existing best effort routing will not
provide the optimum routes for these new levels of service and to
achieve this it would be necessary to have either routing protocols
that supported optimum path discovery or mechanisms to configure
paths necessary to support the required services. In addition to
specific service parameters reliability will also be a potential
service discriminator. It is unlikely using traditional path
determination methods that in the event of a failure a new path could
be determined sufficiently quickly to maintain the agreed service
level. This would imply the need for multiple path reservations in
some instances. Because Per flow reservations are too resource
intensive virtual trunks would provide a good way to reduce the
amount of management traffic by reserving blocks of capacity and
would provide stability in the event of a failure in the resource
reservation and route selection functions.
There are implications of providing shaping at the network
boundaries. Shaping would include both rate and burst parameters as
well as possible delay aspects. Having to provision services with
specific service parameters would present both major business and
technical problems. By definition, packet data is bursty in nature
and there exist periods of idleness during the session that a
provider could reasonable hope to exploit to better utilize the
network resources. It is not practical to expect a consumer paying a
premium for a service would not check that the service was truly
available. Such a service model seems to be filled with peril for
the existing best effort Internet, because any significant amount of
bandwidth that was reserved for exclusive use or a high priority flow
would not be available for best effort data.
With respect to traffic within the network itself there will be the
need to pre-configure routes and to provide the ability to have
routes be dynamically configured. Some of the problems with pre-
configured traffic include the basic inconsistency with the way
traffic is currently engineered through the Internet and the
difficulty in developing arrangements between administrative domains.
The current Internet has been developed with one of the most
egalitarian yet simplistic methods of sharing bandwidth. Supporting
the existing best effort service, in an unbiased way, while at the
same time providing for other classes of service could potentially
add a tremendous amount of complexity to the QoS scheme. On the
other hand, if the reserved bandwidth is not shared it could result
in a significant impact on the availability of the bandwidth in the
Internet as we know it today. QoS could potentially contribute more
to their being insufficient bandwidth, by reserving bandwidth within
the network that can not be used by other services, even though it
can be expected that this bandwidth will be underutilized for much of
the time. Add to that the motivation of the service providers in
wanting to sell commodity bandwidth, and there could be tremendous
pressures on the availability of Internet bandwidth.
Current work within the IP community on defining mechanisms to
provide QoS have centered on a particular few architectures and a
handful of new protocols. In the following sections, we will examine
some of the particular issues with regards to the current IP
community efforts as they relate to the previous discussions. It is
not the goal of this document to serve as a tutorial on these efforts
but rather to identify some of the support issues related to using
particular technologies that support some form of classifiable
service within an IP network.
6.2 QoS Service Management Scope
One can restrict the scope of a discussion of QoS management only to
the configuration of a path between two endpoints. Even within this
limited scope there still remains many unresolved issues. There is
no expectation that a QoS path for traffic between two points needs
to be, or should be, the same in both directions. Given that there
will be an originator of the connection there are questions about how
billing and accounting with be resolved if the return path is
established by a different provider then that of the originator of
the connection. To facilitate billing a method will need to exist
that permits the application originating the call to pay also for the
return path and also for collect calls to be made. 3rd party
providers will need to be established that are trusted by all parties
in the data path to insure billing and guaranteed payment. Utilizing
the service of a virtual DCN that is built upon both IETF and non-
IETF protocols, messages between service providers and the 3rd party
verification system can be secured. A signaling protocol will be
necessary to establish the cost of the call and who will be paying
for it, and each provider will need a verifiable method to bill for
the service provided. As pointed out earlier this functionality will
be similar to what is used in the existing telephone network, but
will be at a much larger scale and potentially involve providers that
are highly competitive with each other.
7. The DiffServ Architecture
The DiffServ management problem is two pronged. First there is the
management within the administrative domain that must be addressed,
and then the management between the domains. There has been little
actual work on the second in the architecture. What work there has
been anticipates that service level agreements will be reached
between the administrative domains, and that end-to-end service will
be a concatenation of these various service level agreements. This
is problematic for many reasons. It presumes that agreements reached
bilaterally could be concatenated and continue to provide a level of
end-to-end service the customer would be willing to pay a premium
for. Problems discussed earlier, with trying to maintain large
numbers of these agreements between competitive networks would also
apply, and tend to limit the effectiveness of this approach. To
efficiently establish the chain necessary to get end to end service
it might take an infinite number of iterations.
Guaranteeing a class of service on a per hop basis is in no way a
guarantee of the service on an end-to-end basis. It is not likely
that a customer would be willing to pay for an improved level of
service if it did not include guarantees on the bandwidth and the
quantitative bounds on delay and error rates guaranteed end-to-end.
This would necessitate engineering the paths through the network so
as to achieve a desired end-to-end result. While it is very likely
that an initial attempt at providing this kind of service will
specify only a particular ingress and egress border, for robustness
and flexibility it will be desirable to have a network that can
support such service without such limitations. The Intserv approach,
as opposed to the DiffServ architecture, would require per flow
information in the core network and may as a result of this prove not
to be scalable [11]. A DiffServ type architecture, with a limited
number of service classes, could be pre-provisioned, and as network
circumstances warranted, be modified to support the actual dynamics
of the network.
The high level functional requirements for edge routers has been
quite well defined in the DiffServ architecture, but the true scope
of the effort to implement this functionality has not been well
recognized. While interesting differences exist between the QoS
architecture of the Internet and the circuit switched network used
for telecommunications much of the lessons learned in
telecommunications should, even if they might do little else, provide
some insight into the level of effort needed to implement these kinds
of requirements. Ironically, given the Internet community in the
past has rejected the level of standardization that was proposed for
management of telecommunications networks, it may be the full service
internet where it becomes actually imperative that such requirements
be completed if the desired services will ever be offered.
8. A Summary of the QoS Functional Areas
The management of QoS will need to provide functionality to the
application and/or at the access, at the core, and at the boundaries
to administrative regions.
QoS traffic functions will need to include admission control,
authentication and authorization, and billing. Verification that
traffic is within agreed parameters and programmatic interfaces to
advise when the service is outside the agreed limits. Interfaces
that provide service verification, fault notification, and re-
instantiation and termination will also be necessary.
Core functions will include traffic engineering, network device
configuration, fault detection, and recovery. Network devices will
need to inform the management system of their available resources and
the management system will need to tell devices how and where to
forward data.
Between administrative regions accounting, service signaling, and
service verification will be needed. At the administrative
boundaries of the network functions similar to those provided at the
edge will be necessary. Peer entities in different administrative
domains would signal their needs across the boundary. Verification
at the boundary could then occur consistent with the verification at
the edge. Actual traffic through the boundaries could be measured
and billing information be transferred between the domains. The
central management function would be responsible for re-routing
traffic in the event of a failure or to better utilize the existing
network resources.
Billing requirements suggest the need for 3rd party verification and
validation functions available to each provider of QoS service within
the flow. On one side of the transaction functionality is needed to
approve pricing and payment and on the other side there will need to
be an interface to provide the pricing information and make payment
request for payment demands.
These requirements will raise a host of issues not the least of which
is security. For the most part security considerations will be
addressed both by securing the protocols (like with IPsec) and by
establishing a dedicated network for control information [6]. While
it will be in most instances too costly to create a physically
separated DCN it will be possible to create a virtually separated
network that will provide the same security benefits. Future work in
the IRTF Service Management Research Group intends to look in detail
at these requirements.
9. Security Considerations
For an issue as complex as a Service Management architecture, which
interacts with protocols from other standards bodies as well as from
the IETF, it seems necessary to keep in mind the overall picture
while, at the same time, breaking out specific parts of the problem
to be standardized in particular working groups. Thus, a requirement
that the overall Service Management architecture address security
concerns does not necessarily mean that the security mechanisms will
be developed in the IETF.
This document does not propose any new protocols, and therefore does
not involve any security considerations in that sense. However,
throughout this document consideration of the security issues raised
by the architectural discussions are addressed.
10. Summary
The paradigm for service management in IP networks has been adopted
from that of telecommunications networks. Basic differences between
the service models of these networks call into question if this is
realistic. Further analysis is needed to determine what is the
proper paradigm for IP service management and to define a common
vocabulary for it.
The IP community is currently very active in solving problems
relating to transport QoS issues. These activities are illustrated
by the work of the Diffserv, Intserv, and Policy working groups. In
contrast not enough effort is being focused on service issues
relating to applications. The present solution is for applications
to build in their own service management functionality. This is
often an inefficient use of network resources, but more importantly
will not provide for access to transport level services and the
functionality that they offer.
The IP community needs to focus on adding service functionality that
is flexible enough to be molded to specific application needs, yet
will have access to service information that will be necessary to
provide superior application functionality. Principal needs to be
addressed relate to developing transport level services for billing
and security. Directory services and extending the work done to
define AAA services are promising starting points for developing this
needed functionality.
11. References
[1] L. Mathy, C. Edwards, and D. Hutchison, "The Internet: A Global
Telecommunications Solution?", IEEE Network, July/August 2000.
[2] B. Leiner, et. al., "A Brief History of the Internet version
3.31", revised 4 Aug 2000.
[3] Eder, M. and S. Nag, "Service Management Architectures Issues
and Review", RFC 3052, January 2001.
[4] Y. Bernet, "The Complementary Roles of RSVP and Differentiated
Services in the Full-Service QoS Network", IEEE Communications
Magazine, February 2000.
[5] Floyd, S. and L. Daigle, "IAB Architectural and Policy
Considerations for Open Pluggable Edge Services", RFC 3238,
January 2002.
[6] Recommendation M.3010 "Principles for a telecommunications
management network", ITU-T, February 2000.
[7] Recommendation M.3100 "Generic network information model",
ITU-T, July 1995.
[8] Moore, B., Ellesson, E., Strassner, J. and A. Westerinen,
"Policy Core Information Model -- Version 1 Specification", RFC
3060, February 2001.
[9] V. Jacobson, "Differentiated Services for the Internet",
Internet2 Joint Applications/Engineering QoS Workshop.
[10] Nichols, K., Jacobson, V. and L. Zhang, "A Two-bit
Differentiated Services Architecture for the Internet", RFC
2638, July 1999.
[11] Mankin, A., Baker, F., Braden, B., Bradner, S., O'Dell, M.,
Romanow, A., Weinrib, A. and L. Zhang, "Resource ReSerVation
Protocol (RSVP) Version 1 Applicability Statement Some
Guidelines on Deployment", RFC 2208, September 1997.
12. Authors' Addresses
Michael Eder
Nokia Research Center
5 Wayside Road
Burlington, MA 01803, USA
Phone: +1-781-993-3636
Fax: +1-781-993-1907
EMail: Michael.eder@nokia.com
Sid Nag
PO Box 104
Holmdel, NJ 07733, USA
Phone: +1-732-687-1762
EMail: thinker@monmouth.com
Hemant Chaskar
Nokia Research Center
5 Wayside Road
Burlington, MA 01803, USA
Phone: +1-781-993-3785
Fax: +1-781-993-1907
EMail: hemant.chaskar@nokia.com
13. Full Copyright Statement
Copyright (C) The Internet Society (2002). All Rights Reserved.
This document and translations of it may be copied and furnished to
others, and derivative works that comment on or otherwise explain it
or assist in its implementation may be prepared, copied, published
and distributed, in whole or in part, without restriction of any
kind, provided that the above copyright notice and this paragraph are
included on all such copies and derivative works. However, this
document itself may not be modified in any way, such as by removing
the copyright notice or references to the Internet Society or other
Internet organizations, except as needed for the purpose of
developing Internet standards in which case the procedures for
copyrights defined in the Internet Standards process must be
followed, or as required to translate it into languages other than
English.
The limited permissions granted above are perpetual and will not be
revoked by the Internet Society or its successors or assigns.
This document and the information contained herein is provided on an
"AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Acknowledgement
Funding for the RFC Editor function is currently provided by the
Internet Society.
|
Comment about this RFC, ask questions, or add new information about this topic: