Internet Engineering Task Force Olivier Duroyon
Rudy Hoebeke
Hans De Neve
Dimitri Papadimitriou
Internet Draft Alcatel
Document: draft-duroyon-te-ip-optical-01.txt November 2000
Expiration Date: May 2001
G.LSP Service Model framework in an Optical G-MPLS network
<draft-duroyon-te-ip-optical-01.txt>
Status of this Memo
This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts.
Internet-Drafts are draft documents valid for a maximum of six
months and may be updated, replaced, or obsoleted by other
documents at any time. It is inappropriate to use Internet- Drafts
as reference material or to cite them other than as "work in
progress."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt.
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html.
1. Abstract
The objective of this draft is to propose an IP service model for a
non-packet switch capable optical network where G.LSPs are
dynamically triggered by the IP layer and subsequently advertised
for IP routing. The business model assumes that several IP service
domains, some of which represent different administrative entities,
share the same optical backbone and focuses therefore primarily on
an overlay model. G-MPLS signaling (refer to [g-mpls]) with UNI
support is assumed as underlying control plane protocol.
2. Conventions used in this document
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
Duroyon et al. Expires May 2001 1
draft-duroyon-te-ip-optical-01.txt November 2000
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in
this document are to be interpreted as described in RFC2119.
3. Introduction
This draft introduces an end-to-end IP service model enabling the
dynamic management of Generalized Label Switched Paths (G.LSP) by
means of G-MPLS signaling with User-to-Network Interface (UNI)
support. A G.LSP is a point-to-point connectivity with specified
attributes (some of which are mandatory, while others are optional)
established between two termination points in the optical network.
A G.LSP could be a fiber switched path, a lambda switched path, a
TDM switched path (circuit) or a packet-switch capable G.LSP. The
scope of this draft is restricted to optical networks, which are by
definition non-packet switch capable. Consequently, G.LSPs are
restricted to non-packet switch capable G.LSPs, which we hereafter
refer to as G.LSPs.
For reasons of definiteness, the optical devices are always
referred to as Optical Network Elements (ONE) and the IP devices as
Client Network Elements (CNE). Boundary CNEs and boundary ONEs are
interconnected through an UNI signaling and routing interface.
The owner of the UNI interface in the optical domain (UNI-Network
or UNI-N) is referred to as in the Boundary ONE Controller (ONE-C).
Its counterpart in the Client Network (UNI-Client or UNI-C) is
referred to the Boundary CNE Controller (CNE-C).
The terminology used in the draft attempts to be in line with the
definitions found in [ip-optical], [ouni-framework] and
[OIF2000.125.2].
An overlay use of G-MPLS (UNI support) is appropriate for an
untrusted environment where several IP service domains,
representing different administrative entities, share the same
optical backbone. Moreover, this model seems well suited for a
network architecture including non-IP devices, e.g., legacy TDM or
ATM equipment, that interface with the same optical backbone. This
is however beyond the scope of this draft.
To distinguish between trusted and untrusted peers, a separate
definition for a Trusted and Untrusted network interfaces is
proposed:
- An Untrusted interface is defined when UNI (respectively NNI)
interfaces belongs to distinct administrative authorities. For
instance an UNI interface between a client network element and an
optical network element belonging to distinct ISPs defines an
untrusted relationship between the client and the optical network
element.
- A Trusted interface is defined when UNI (respectively NNI)
interfaces belongs to the same administrative authority. For
instance an NNI interface between two ONEs belonging to the same
Duroyon et al. Expires May 2001 2
draft-duroyon-te-ip-optical-01.txt November 2000
optical carrier defines a trusted relationship between these
optical networks elements.
The service model further assumes that a decision point in the IP
service domain, e.g., a Traffic Engineering tool (TE tool), will
trigger a boundary CNE to issue a G.LSP request towards the optical
domain. The decision point determines the need for a G.LSP on the
basis of IP Service Level Agreements (IP SLAs) and related
information, such as for instance load measurements in the IP
service domain.
The same TE-tool may also decide about the configuration of Traffic
Engineering LSPs (TE-LSP), which are by definition Packet-switch
capable LSPs. For the purpose of IP traffic engineering, TE-LSPs
are in this case created on top of non-Packet-switch capable
G.LSPs.
In the remainder of the document, the terms TE tool or decision
point are used interchangeably and refer to the IP service domain
device, capable of triggering G.LSP requests.
4. IP service model description
This section outlines the sequence of events that characterize our
proposed IP service model.
(1) Configuration
Configuration consists of installing and configuring interfaces of
the boundary ONEs and boundary CNEs.
During this stage of end-points configuration, physical attributes
of the end-point (as protection attributes of the drop side) are
also configured. Configuration of the NNI interfaces of the ONEs is
out of the scope of this draft.
(2) Neighbor Discovery, Endpoint Registration and Service Discovery
The objective of Neighbor Discovery is to provide the information
needed to identify the neighbor identity and neighbor connectivity
over each link interconnecting a boundary CNE to a boundary ONE.
Endpoint registration is concerned with registering boundary CNE
endpoints to the optical network. The registration information
includes the resource capabilities, closed user group (CUG)
identification, port reachability information, UNI protection
capabilities etc. This set of information is critical to enable
dynamic G.LSP services at the UNI. The endpoint registration
mechanism enables the end system to register its critical set of
information so that other end systems can identify its existence
and network properties.
Duroyon et al. Expires May 2001 3
draft-duroyon-te-ip-optical-01.txt November 2000
Service discovery is concerned with obtaining the essential
information about services from the attached optical network that
are available for the CNE. This information is used by CNEs to
establish the service environment. The service discovery mechanism
allows the network element to convey information about available
services to the end system.
After finishing neighbor discovery, endpoint registration and
service discovery, each end system should establish the service
environment and have sufficient information to generate G.LSP
service request. These mechanisms complement each other and they do
not depend on the establishment and use of signaling channels
between the two parties.
(3) Optical Service Level Agreements
Next, the service model consists of negotiating Optical SLAs (O-
SLA) at optical network-client network boundaries, or between
optical networks.
In case of an untrusted peering relationship (i.e. untrusted UNI,
respectively NNI), each G.LSP request is authenticated and
validated against the O-SLA. The validation process against an O-
SLA includes checking whether the request is conforming to the
restrictions (e.g., on scope) defined in the O-SLA.
O-SLAs may also be defined at trusted interfaces as the optical
domain to provision resources that could use them. Trusted in this
context refers to the fact that you don't expect the CNE to violate
this O-SLA, and as such requests received from trusted neighbors
don't need to be validated against the O-SLA.
(4) G.LSP service request
The decision point of the client network determines the required
connectivity through the optical domain based on service
requirements as per the IP SLAs. It then triggers the boundary CNEs
to send a G.LSP service request towards the associated boundary
ONEs, using G-MPLS signaling with UNI support. This process is
dynamic and may involve, amongst others, the creation of additional
G.LSPs, the deletion of existing G.LSPs or the modification of
existing G.LSPs.
(5) Address resolution
At the UNI, the G.LSP service request sent by the CNE needs to
include the ONE source and destination termination-point
identifiers (in case of trusted UNI interface) or the CNE source
and destination termination-point identifiers (in case of untrusted
UNI interface). CNE termination-points should also be considered
when the G.LSP is established through several optical networks
belonging to different administrative authorities.
Duroyon et al. Expires May 2001 4
draft-duroyon-te-ip-optical-01.txt November 2000
Consequently, the source client needs to send an address resolution
request to obtain the ONE destination termination-point ID or CNE
termination-point ID corresponding to the CNE destination logical-
address of the G.LSP service request.
(6) Optical path selection
In case a dedicated instance of an IGP is used inside the optical
transport network, each boundary ONE learns the complete topology
of the optical domain. A Constraint-based Shortest Path First
(CSPF) algorithm can then be implemented in the boundary ONEs to
calculate a route for the G.LSP in line with the constraints
specified in the request. As an example, the route of a G.LSP may
depend on the protection requirements or routing constraints
specified in the G.LSP request. The latter may indicate that the
requested G.LSP should be routed diversely from other G.LSPs. This
CSPF algorithm is expected to be quite different from an IP CSPF
algorithm because of optical networking specific considerations.
(7) G.LSP advertisement to the IP layer
As soon as the G.LSPs are lit up, they are advertised to the client
network. The involved boundary CNEs will either create a new IP
link and start to exchange routing information (using IGP or eBGP)
or modify the characteristics of the existing IP link.
(8) Traffic engineering for optimization of the optical domain
Optionally, the optical domain may have its own off-line Optical
Traffic Engineering (O-TE) tool. This tool may be used for
optimization of resource utilization in the optical network by
rearranging some G.LSPs.
5. UNI discovery and registration services
In order to provide a flexible and end-to-end IP Service model,
with a minimum set of local provisioning, specific mechanisms and
procedures have to be defined at the boundary between the client
and the optical network:
- to discover neighbors identity and connectivity
- to register client end-point identity
- and to discover the supported UNI and network services.
Transport mechanisms used for the UNI discovery and registration
services are referenced in [OIF2000.125.2] and [OIF2000.200].
5.1 Neighbor discovery at the UNI
The key objective of Neighbor Discovery at the UNI is to provide
the information needed to identify the neighbor identity (IP
address associated to the corresponding UNI) and neighbor
Duroyon et al. Expires May 2001 5
draft-duroyon-te-ip-optical-01.txt November 2000
connectivity over each link connecting the boundary CNE to the
boundary ONE.
Neighbor discovery process which is also referred to as the
Termination-port identity process, provides the following
information to the boundary CNE and ONE:
- the ONE discovers the identity of the client CNE by
automatically discovering the IPv4 address assigned to the UNI-C
and the identity of each physical port connected to the CNE
- the CNE discovers the identity of to the connected ONE by
automatically discovering the IPv4 address assigned to the UNI-N
and the identity of each physical port connected to the ONE
If the signaling transport mechanism is not explicitly configured,
the neighbor discovery process ends by the bootstrapping of the
signaling control-channel used to exchange the information during
the end-point registration and the service discovery processes.
5.2 End-point Registration and UNI Service Discovery
The end-point registration process includes the exchange of
information between the CNE and ONE for each of the ports and
logical ports connecting the CNE to the ONE. A logical port defines
the structure of a physical port by identifying for a given port
each of the channels included within this port and sub-channels
included within this channel.
The UNI Service discovery process includes the exchange of
resource-related information of the Framing and Bandwidth capacity
of each of the ports and logical ports connecting the CNE to the
ONE. Additional parameters, such as the UNI drop-side protection
attributes (UNI client-side protection and UNI network-side
protection) and the G.LSP Directionality support could also be
exchanged during the resource discovery process. For SDH/Sonet
interfaces, the Transparency levels (STE-C, LTE-C), the client
support of Virtual Concatenation (VC) and the levels of Continuous
Concatenation (CC).
The end-point registration process includes also the address
registration process [OIF2000.261.1]. The address registration
process allows the CNE to register the CNE logical addresses
attached to the CNE Termination-point ID to which corresponds an
unique ONE Termination-point IDs. A CNE Termination-point ID
includes the unique IP address associated with the client element
and the logical-port ID. The logical-port ID comprises the port-ID,
Channel-ID and Sub-channel-ID as defined in [OIF2000.125.2].
When the address registration is part of the end-point registration
process, the CNE associates the CNE Termination-point ID with the
corresponding logical address and ONE termination-point ID. When
the CNE does not associate logical addresses with their interfaces,
Duroyon et al. Expires May 2001 6
draft-duroyon-te-ip-optical-01.txt November 2000
the CNE termination-point ID resolution implies that the boundary
ONE knows the mappings between the CNE termination-point ID and the
ONE termination-point ID. This case is considered as a particular
case where the CNE logical address fields are empty. In this case,
the value of the logical address could correspond to the user-group
identifier to which the G.LSP belongs; however, in this particular
case, the address resolution is always based on the CNE
termination-point ID.
Other client identifiers could be exchanged during end-point
registration process:
- the CNE registers the Contract ID attached to a specific element
and/or interface
- the CNE registers the Closed User-Group (CUG) IDs (i.e. User-
Group ID) attached to a specific client end-point or port
5.3 Network Service Discovery
The network service Discovery consists of the G.LSP service-related
discovery process and a policy related service discovery process.
During the G.LSP-related service discovery process, the CNE
registers and/or discovers the parameters related to
- SDH/Sonet related services, i.e., the SDH/Sonet Transparency
levels supported and the Continuous Concatenation levels
supported
- G.LSP Directionality support
- Network-side Protection, i.e., the Protection-levels services
provided by the internal optical network (Unprotected, Dedicated
1+1 Protection, Dedicated 1:1 and Shared Protection)
- G.LSP Priority classes and Preemption levels supported by the
optical network
- G.LSP Diversity options supported by the optical network
- Security levels support (IPSec or other authentication
mechanism) within the signaling used on the control-plane
throughout the optical network
The discovery of the Policy-related service may include the
following parameters:
- Service-levels offered by optical network
- Scheduling-related service supported by the optical transport
network and/or the scheduling desired by the client
- Billing-related service supported by the optical transport
network and/or the billing method desired by the client
- Vendor-related and Optional parameters
6. Address resolution
As stated in section 4.5, the source client needs to send an
address resolution request to obtain the ONE destination
termination-point ID (trusted UNI interface) or CNE termination-
Duroyon et al. Expires May 2001 7
draft-duroyon-te-ip-optical-01.txt November 2000
point ID (untrusted UNI interface) corresponding to the destination
CNE logical-address.
Consequently, at a trusted UNI interface, the G.LSP create message
sent by the CNE to the ONE includes the source and destination ONE
(or CNE) termination-point IDs requested for this G.LSP. This
implies that the source ONE must perform a internal address-lookup
toward a directory service or a local mapping table, in order to
find the mapping between the destination CNE termination-point ID
and the destination ONE termination-point ID.
So, the optical network client only needs to know the CNE source
and destination logical address and termination-point ID in order
to request a G.LSP creation; any other topological information
concerning the optical network termination-point identification is
transparent for the client.
This mechanism is also adapted for inter-domain G.LSP (cf. Section
9.1) since in this case only the autonomous-system (AS) boundary
ONE termination-point to CNE termination-point mapping-list has to
announced to the neighboring BGP AS's.
7. Optical Service Level Agreements
An optical domain-IP service domain boundary coincides with a UNI
with its associated O-SLA. Similarly, if there are multiple optical
sub-networks in the optical domain, there will be O-SLAs negotiated
at each optical sub-network boundary. An optical sub-network
boundary then corresponds to an optical Network-to-Network
Interface (NNI). In this draft, we limit the discussion to O-SLAs
at the level of UNIs.
As mentioned before, G.LSP requests issued by a boundary CNE are
only accepted within the constraints of an O-SLA. This means that
in case of an untrusted peering relationship, each G.LSP request is
authenticated and validated against the O-SLA. It was already
indicated that O-SLAs may also be defined at trusted interfaces.
However, G.LSP requests received from trusted neighbors don't need
to be validated against the O-SLA.
In the scope of this draft, we only discuss the technical aspects
of an O-SLA. Borrowing from the terminology introduced in
[diffserv-framework], we refer to the technical part of an O-SLA as
an Optical Service Level Specification (O-SLS). An O-SLS is
considered to be unidirectional and to specify performance
expectations (i.e., the service level) for the IP service domain as
well as imposed reachability constraints, e.g., CUG.
O-SLS parameters could for example include:
1. Capacity constraints
Duroyon et al. Expires May 2001 8
draft-duroyon-te-ip-optical-01.txt November 2000
An ingress O-SLS may contain limits on the maximum number of G.LSPs
that can be established from a specific ingress point, possibly as
a function of time of day, as well as bandwidth constraints (OC-48,
OC-192, etc.).
An egress O-SLS may put capacity constraints on the G.LSPs that the
receiving IP service domain is willing to terminate.
2. Service performance parameters
Examples are G.LSP setup and/or recovery admitted latency,
supported protection/restoration options, availability, supported
routing constraints, accessibility (i.e., G.LSP request blocking
probability), responsiveness (specifying upper limits on the
processing time of G.LSP requests), etc.
3. Constraints on the 'scope' of G.LSP request
This may be viewed as an extension to the concept of CUGs, which by
nature already exhibit reachability limitations. Scope constraints
are intended to additionally restrict the topological extent of
G.LSPs. In its simplest form, the O-SLS offers to accept any G.LSP
request issued by the IP service domain over a specific O-UNI up to
a maximum capacity without any scope constraint within the CUG (so-
called hose O-SLS). Conversely, the agreement may be constrained by
the egress point of a G.LSP. For example, the optical domain
service provider might agree to the setup of G.LSPs, up to a
certain maximum capacity, but only if these G.LSPs are destined to
a specific set of egress points within the CUG.
Part of the purpose of O-SLSs is to protect resources in the
optical domain by validation of submitted G.LSP requests. If the
optical domain and the IP service domain are under control of the
same administrative authority, then there is likely to be a trusted
peering relationship between both domains. Conversely, in case of
an untrusted peering relationship, the optical domain service
provider validates incoming G.LSP requests as per the O-SLS. This
validation process can be implemented in the ONE-C. In this case,
there exist several mechanisms to install an O-SLS in an ONE-C,
e.g., CLI, SNMP, LDAP or COPS. Alternatively, the O-SLS enforcement
may be outsourced to another policy entity.
An O-SLS offers to accept G.LSP requests at the service level
agreed with the IP service domain. The optical domain service
provider will provision the optical domain accordingly. A broad
range of optical services could be envisioned. As an example,
services could be defined with different levels of accessibility,
depending on the probability that a G.LSP establishment request is
blocked. Moreover, services could also be categorized as protected
or non-protected, depending on the offered protection level. All of
these service level characteristics influence the resource
provisioning process in the optical backbone.
Duroyon et al. Expires May 2001 9
draft-duroyon-te-ip-optical-01.txt November 2000
For each G.LSP request, the optical domain service provider may
also need to identify the O-SLS for which the request is submitted.
Some authentication may be included in the request in order to
verify that the rightful IP service provider issued the request. In
some cases, this customer might be implicitly derived from the
signaling channel on which the G.LSP request was received. The
authentication mechanism must be specified in the O-SLS.
Although it can be assumed that O-SLSs will be static for the
foreseeable future, this draft does not preclude dynamic O-SLSs.
These would necessitate some automated form of interaction between
the IP service domain and the optical domain. In case of an O-SLS
at the O-UNI, this may for instance require the interaction between
a Bandwidth Broker (BB) in the IP service domain and a Lambda
Broker (LB) in the optical domain. At the level of an O-NNI, this
would be between different LBs, acting on behalf of the different
optical sub-networks. This automated (re-)negotiation of O-SLSs
would in turn call for an automated O-SLS admission control
function. Note that this admission control function is different
from the validation of G.LSP requests as per the negotiated O-SLS,
referred to as O-SLS enforcement.
8. G.LSP triggers
As stated in [ip-optical], the G.LSP request triggering process
should be part of a stable traffic engineering tool in the IP
service domain as opposed to a data-driven shortcut approach,
similar to the schemes proposed for IP over ATM networks.
Henceforth, an integrated TE-LSP and G.LSP triggering process is
proposed at the end of this section to alleviate the shortcomings
of the former method and is elaborated below.
8.1 Data-driven shortcut approach for G.LSPs
The data-driven shortcut model would imply that the boundary CNEs
use traffic measurements to autonomously control the number of
G.LSPs that connect them with a set of remote boundary CNEs across
the optical domain. For example, boundary CNE A could detect that
some of its traffic is reaching boundary CNE B in a multi-hop way.
If this traffic trunk is large enough, boundary CNE A might decide
to set-up a G.LSP to boundary CNE B, relieving the IP forwarding at
all intermediate CNEs on the multi-hop path. In an overlay model,
once a G.LSP has been established to a new destination, it should
be announced as a (new) IP link in the IP service domain routing
database. As such, it can be used by any TE entity in the IP
service domain and this IP link may carry several TE-LSPs. This
implies that the TE entity in the IP service domain would then be
constantly reacting to decisions of the boundary CNEs that are
continuously changing the IP topology.
Such a layered scheme of G.LSP requests and TE-LSP requests is
inefficient and could also break the TE service model, when the
Duroyon et al. Expires May 2001 10
draft-duroyon-te-ip-optical-01.txt November 2000
only available G.LSP between two boundary CNEs would be torn down.
Such a decision might be based on the valid observation that the
traffic pattern has changed such that the existing G.LSP is under-
utilized and may be re-directed towards another boundary CNE.
However, the G.LSP might still carry TE-LSPs. Turning off the G.LSP
has the effect of a link failure and will hence trigger the TE
entity in the IP service domain to recover from this failure.
Depending on whether the TE-LSP was protected or not, one of the
following scenarios will take place.
8.1.1 Protected TE-LSPs
TE-LSPs can be used to carry mission critical traffic requiring a
fast recovery scheme in case of link failures. Upon such an event,
the traffic of the working TE-LSP can be switched to a protect TE-
LSP that has been pre-configured along a node- and link-disjoint
path. Depending on whether G.LSP is protected or not throughout the
optical network, the following alternative is considered:
- Protected G.LSP: if the turned-off G.LSP was protected within the
optical domain, the TE-LSP path calculation might have selected
this IP link for both the working and the protect path of the TE-
LSP. In that case, the TE-LSP protect path will not be available
and connectivity will be lost.
- Unprotected G.LSP: in this case the problem would not arise since
the route diversity TE-LSP protect scheme would have selected
another IP link for the protect path.
8.1.2 Unprotected TE-LSPs
If the TE-LSP was not protected, the source nodes of the TE-LSPs
running over the turned-off G.LSP will start a CSPF calculation to
find an alternative path. As all source nodes will be competing for
the same resources, some G.LSP requests will be blocked and it
might take a while before all G.LSPs have been restored.
The above scenario equally pertains to the case of any link failure
in an IP service domain. However, link failures in an IP service
domain may be considered as rare events. This is however not the
case when this link failure behavior is the result of a data-driven
shortcut approach across an optical backbone.
8.2 Integrated TE-LSP and G.LSP triggering process
Given the above shortcoming, boundary CNEs should not autonomously
decide to tear down a G.LSP. Yet, it may not always be appropriate
to maintain an under-utilized G.LSP. However, a G.LSP should not be
turned off until the TE-LSPs it carries, have been re-routed along
an alternative path. This might even require an additional G.LSP
setup between two other boundary CNEs. All of this calls for a
coordinated TE-LSP and G.LSP triggering process, integrated in the
Duroyon et al. Expires May 2001 11
draft-duroyon-te-ip-optical-01.txt November 2000
same decision point. This is possible since both responsibilities
reside within the IP service domain.
The ability to dynamically establish G.LSPs adds an extra dimension
to the TE capabilities of an IP service domain. In addition to
forwarding packets along non-shortest paths, it is now also
possible to (re-)configure the topology of the IP service domain by
means of adding, deleting or modifying G.LSPs across the optical
backbone.
This integrated decision point will use the set of IP SLAs and the
derived traffic trunk requirements across the IP service domain
(possibly complemented with traffic measurements) to determine the
optimal set of G.LSPs and TE-LSPs.
Several setup optimization strategies are possible depending on the
business model in use between the IP Service domain and the optical
domain, and also the assumptions taken on the pre-existing optical
topology.
The TE decision point has the complete knowledge of the IP
Topology, all optical end-points, including their logical, and
physical attributes (granularity, protection attributes, _).
The different strategies may be chosen among the following:
1- Minimize the number of G.LSPs to be lit up
This strategy fits in business models where the optical domain
doesn't belong to the service domain, and as such each
additional network G.LSP is an additional cost to the service
domain. The TE decision point optimizes the number of G.LSPs
to set up through the optical domain for a given IP traffic
pattern.
2- Add capacity without rearranging optical topology
Before triggering new G.LSPs, the TE decision point tries to
rearrange TE-LSPs without rearranging the underlying optical
topology.
3- Add capacity with specific explicit constraints
Some environment may lead to some specific constraints to be
taken into account during route computation.
One simple example is a mixed ATM / IP network. In this
example TE-LSP used by ATM and their underlying G.LSP must not
be rearranged during the computation to add optical capacity.
The TE decision point optimizes the number of G.LSP (and
subsequently TE-LSP topology) with the possibility of pinning
down some components (TE-LSP, G.LSP, _)
Duroyon et al. Expires May 2001 12
draft-duroyon-te-ip-optical-01.txt November 2000
4- Optimize IP topology without any optical constraint
TE decision point optimizes the IP topology without taking any
constraint on number of G.LSPs setup. The only constraints
taken are in this case coming from the end-points attributes.
In addition to the computation algorithm strategy, the TE decision
point also takes into account the sort of IP services to be
achieved, in order to achieve a consistent restoration between
protocol layers.
One simple way is to define a linear hierarchy between IP services.
1.- Layer 1 protection - Non-PSC Level Protection
This service only applies for IP link built between two PSC-
capable end-points. The G.LSP connecting both end-points is
totally protected. It means that it will be chosen from a pool
of G.LSPs with source and destination drop-side protection
(1+1, 1:1, Shared Protection). And in addition the G.LSP will
request a network-protected path via the optical network.
This service will be mainly seen in a traditional environment
where the service domain lies on a very reliable transport
layer, dedicated to any fast restoration mechanism.
2.- Diverse Layer 2 _ PSC Level Protection
This service also only applies for a G.LSP built between two
PSC-capable end-points (for instance, an IP link connection).
Two G.LSPs are requested to the optical cloud via the same
CNE-to-ONE interface, using source and destination drop-side
protected G.LSPs.
No optical or SDH/Sonet network protection are required for
the G.LSPs. But diverse optical paths are requested for both
G.LSPs.
This service makes sense in a network architecture where the
CNE is locally connected to an ONE, and so the diverse path
routing must start at the first ONE of the optical network.
3. Diverse Layer 2 - Network Level Protection
This service applies indifferently in a mixed PSC-capable (and
particularly for IP services) and non-PSC capable optical
environment, and not necessarily at the boundary CNE.
Two G.LSPs are requested from the IP service domain using two
diverse paths. In this case when the G.LSP request reaches the
optical cloud boundary, there is no specific protection
requirements towards the optical cloud.
Duroyon et al. Expires May 2001 13
draft-duroyon-te-ip-optical-01.txt November 2000
4. No G.LSP protection
This service applies when the restoration mechanisms don't
rely on pre-existing backup paths. In this case on protection
constraints have to be taken into account at the optical
layer.
As described in this paragraph, in order to create a consistent
end-to-end IP Service Model, network devices and TE decision point
have to synchronize each other to setup and maintain the adequate
and optimal set of G.LSPs and TE-LSPs. The resulting topology is
based on IP services requirements (Protection scheme, _) and
computation strategies (Business models, _).
This leads to needs for potential new standardization items, as
information exchange between routers and TE decision point (in case
of G.LSP setup failure, _). This will be tackle via subsequent
studies.
9. G.LSP advertisement to the IP layer
The decision point may thus trigger the set-up of an additional
G.LSP to an already connected boundary CNE. Alternatively, it may
trigger a rearrangement of existing G.LSPs, or the establishment of
a G.LSP to a boundary CNE that could previously not be reached
through the optical domain. It might very well be that the decision
point triggers a boundary CNE to drop a G.LSP if its capacity is no
longer needed to meet the requirements of the IP SLAs.
In order to make efficient use of the dynamicity of the G.LSP
create requests, the routing protocol parameters should be
dynamically configurable as well. This section outlines a proposal
to achieve a seamless integration of a new G.LSP within the IP
service domain for the overlay model by means of automatic
configuration.
As soon as a G.LSP to a particular boundary CNE has been lit up, it
is assumed that it is promoted to an operational IP link. In case
of, e.g., regular SDH/SONET framing, this is achieved by running
PPP protocol on the newly established G.LSP.
9.1 G.LSP set-up to a previously unreachable boundary CNE
The draft [ompls-ospf] defines the different facets of the creation
of an IP link in case of a peer-to-peer model and proposes to use
the newly established IP link as a forwarding adjacency in the IP
service domain.
The overlay model imposes different requirements on the IP layer of
the boundary CNEs. Indeed, once the first G.LSP is established
between two boundary CNEs and promoted as IP link, it is to be
advertised as a point-to-point link for IP routing in order to
Duroyon et al. Expires May 2001 14
draft-duroyon-te-ip-optical-01.txt November 2000
initiate the IP connectivity between the two boundary CNEs. And
subsequently it will allow IP reachability between the associated
IP service domains.
Two cases must be considered. The G.LSP is promoted to an IP link
connecting:
- two boundary CNEs of the same Autonomous System (IGP peering),
or,
- two boundary CNEs of different Autonomous Systems (eBGP peering).
The IGP and BGP peering cases do not require the same kind of
configuration and are described separately.
Note that in case of an IGP peer, it is necessary that the G.LSP be
bi-directional, because IGP protocols require a bi-directional
transport layer. Bi-directional G.LSP setup is further detailed
within [g-mpls] and [onni-framework].
From an addressing point of view, the Packet switch capable end-
points can be unnumbered (and, e.g., identified by the Router Id of
the boundary CNE), or numbered through initial configuration, but
different from the IP address assigned to the UNI signaling agents
(UNI-Client and UNI-Network) terminating the signaling channel.
It has to be noticed again that within the overlay model, the
signaling channel identification is neither known nor advertised
throughout the IP service domain.
9.1.1 IGP support
Once the first IP link is established between two boundary CNEs and
configured to support an IGP peer, the boundary CNEs need to get
the proper information:
- The first requirement is to select IS-IS or OSPF for the newly
formed IP link.
- In addition, link routing parameters such as timers and area
numbers might have to be specified. For instance, timers in OSPF
should be consistent at both ends of the IP link.
- Also, link metrics (e.g., resource classes, etc.) need to be
inherited or configured for use by IP routing.
- Finally, the IGP protocol is enabled and the IP link is
advertised.
These parameters have to be accessible and are automatically
configured (prior to or at the time of G.LSP establishment) by the
Duroyon et al. Expires May 2001 15
draft-duroyon-te-ip-optical-01.txt November 2000
decision point of the IP service domain to efficiently deploy the
IP service on top of the G.LSP.
However, some of the routing parameters (e.g., OSPF timers) may be
defaulted to pre-determined values. Those values must be defined
network-wide and be consistent between all possible boundary CNE
pairs. The decision point should be allowed to overwrite those
parameters at the setup time of the G.LSP.
9.1.2 eBGP peering
In the case of an inter-domain G.LSP, static route configuration
specifying the BGP peer (most probably a virtual interface of the
remote boundary CNE) should be configured in the local boundary CNE
in order to set up the TCP session used in BGP.
In addition, an IP SLA is going to be negotiated between the
autonomous systems and routing policies are going to be configured
on both ends of the G.LSPs.
As opposed to the IGP peering case, triggering of inter-domain
G.LSPs will very likely not arise from an automated process because
of the BGP peering negotiation procedure.
9.2 Set-up of an additional G.LSP
In addition to the option described in 9.1 (creation of a new
stand-alone IP link with the new G.LSP and advertisement to the
routing protocol), a second option is now possible, which is to
create an additional G.LSP to an existing IP link that forms a
bundled link [mpls-bundle]. In this case there is no new
configuration necessary for the IGP or BGP routing layer but only
interactions internal to the boundary CNEs.
This bundle is advertised as a single IP link. The G.LSP in itself
may be unidirectional, and hence the bundle could have an
asymmetric bandwidth.
- The boundary CNE upgrades the bandwidth of the bundle link based
on the characteristics of this new G.LSP.
- The new G.LSP is included in the load balancing mechanism that
distributes the traffic amongst the component G.LSPs of the bundled
link, e.g., proportional to their bandwidth.
- The addition of a new G.LSP to a bundle does not impact the
routing topology. Only the new bandwidth of the IP link is
advertised within the IP service domain. The other characteristics
of the IP link, e.g., the resource classes, remain unchanged.
Duroyon et al. Expires May 2001 16
draft-duroyon-te-ip-optical-01.txt November 2000
In the case described in this section, the only mandatory
information to be automatically configured by the decision point is
the bundle identifier to which the G.LSP is to be added.
9.3 Rearrangement of an existing G.LSP
Within the optical network, two alternatives could be considered
for the rearrangement of an existing G.LSP:
- Either the non-destructive modification of an already established
G.LSP. In this case, the source and destination termination-points
of the G.LSP can not be changed, but other parameters such as
bandwidth and network protection could be modified without
disrupting the working G.LSP.
- Or the destructive modification of an already established G.LSP.
This case is the straightforward combination of G.LSP tear-down
followed by a new G.LSP set-up towards a different destination.
10. References
[diffserv-framework] Y.Bernet et al, "A Framework for
Differentiated Services", Internet Draft, draft-ietf-diffserv-
framework-03.txt, February 1999.
[g-mpls] Peter Ashwood-Smith et al., "Generalized MPLS _ Signaling
Functional Description", Internet draft, draft-ietf-mpls-
generalized-signaling-00.txt, October 2000.
[ip-optical] James Luciani et al., " IP over Optical Networks _ A
Framework", Internet draft, draft-ip-optical-framework-00.txt,
February 2000.
[ompls-isis] K. Kompella et al., "IS-IS Extensions in Support of
MPL(ambda)S", Internet Draft, draft-kompella-isis-ompls-extensions-
00.txt, July 2000.
[ompls-ospf] K. Kompella et al., "OSPF Extensions in Support of
MPL(ambda)S", Internet Draft, draft-kompella-ospf-ompls-extensions-
00.txt, July 2000.
[mpls-bundle] K. Kompella et al., "Link Bundling in MPLS Traffic
Engineering", Internet Draft, draft-kompella-mpls-bundle-02.txt,
July 2000.
[onni-framework] D.Papadimitriou et al,. "Optical NNI Framework and
Signaling Requirements", Work in progress, draft-papadimitriou-
onni-framework-00.txt, November 2000.
[ouni-framework] B.Rajagopalan et al., `Signaling Requirements at
the Optical UNI', Internet Draft, draft-bala-mpls-optical-uni-
signaling-00.txt, July 2000.
Duroyon et al. Expires May 2001 17
draft-duroyon-te-ip-optical-01.txt November 2000
[OIF2000.125.2] B. Rajagopalan et al., "User Network Interface
v1.0 Proposal", OIF Contribution 125.2, October 2000.
[OIF2000.200] D. Pendarakis et al., "Signaling Transport
Mechanisms for UNI 1.0", OIF Contribution 200, September 2000.
[OIF2000.261.1] D. Papadimitriou et al., "Address Registration and
Resolution", OIF Contribution 261, November 2000.
11. Acknowledgments
The authors would like to thank Emmanuel Desmet, Sitaram
Kalipatnapu and Gert Grammel for their constructive comments.
12. Author's Addresses
Olivier Duroyon
Alcatel USA
15036 Conference Center Drive
Chantilly, VA, 20151
Phone: (703) 679 6415
Email: olivier.d.duroyon@usa.alcatel.com
Rudy Hoebeke
Alcatel Bell
Francis Wellesplein 1
2018 Antwerp, Belgium
Phone: (32) 3/240.84.39
Email: rudy.hoebeke@alcatel.be
Hans De Neve
Alcatel Bell
Francis Wellesplein 1
2018 Antwerp, Belgium
Phone: (32) 3/240.76.94
Email: hans.de_neve@alcatel.be
Dimitri Papadimitriou
Alcatel NSG-NA
Francis Wellesplein 1
2018 Antwerp, Belgium
Phone: (32) 3/240.84.91
Email: dimitri.papadimitriou@alcatel.be
Duroyon et al. Expires May 2001 18