Transcript EURESCOM - SALTAMONTES

QoS Provision in an MPLS/DiffServ Network
Χάρης Κωνσταντινίδης
Νοέμβριος 2004
Summary

MPLS Architectural description and basic concepts

QoS Management general aspects
MPLS Architectural description and
basic concepts
Definition
An improved method for forwarding packets
through a network using information, contained
in labels attached to IP packets.
It combines the performance and capabilities of
Layer 2 switching with the proven scalability of
Layer 3 routing, thus creating flexible networks
that provide performance and stability.
MPLS Architectural description and
basic concepts
Why MPLS?

MPLS addresses the main concerns with traditional
IP routing concerns:
 Winner-takes-all
 Rely on coarse attributes picking the best path
 Forwarding process can be rather complex processing the entire IP header
Host
A
Router
B
Router
Router
A
Host
B
Router
C
Router
Router
C
Router
F
Path 1
Path 2
MPLS Domain
Router
G
Router
D
Router
Host
C
MPLS Architectural description and
basic concepts
MPLS Operation [1]
Central concept behind MPLS: the label
Packets are assigned a label when they enter an
MPLS network and the network uses that label,
rather than an IP address to deliver packets to
the destination.
Forwarding
based on
IP Address
Sender
Forwarding
based on
Labels
Forwarding
based on
IP Address
Ingress LSR
LSR
Egress LSR
MPLS Domain
Label
Label
IP Packet
IP Packet
IP Packet
Label
IP Packet
Receiver
IP Packet

MPLS Architectural description and
basic concepts
MPLS Operation [2]

Label vs IP address
 Labels are numbers. Numbers that are used to
forward packets. They are little like IP
addresses. What is then the difference?
 Their scope: A legitimate IP address is unique in
all the world while an MPLS label has only local
significance. A given label value is only
significant on a particular link between two
LSRs. Label values can change as a packet
traverses an MPLS network.
MPLS Architectural description
and basic concepts
MPLS Operation [3]


Label size: 32 bits
Schematic representation
LSR D
Ingress
LSR A
Ingress
18
Z
8
LSR B
22
X
37
LSR C
Ingress
Y
MPLS Architectural description and
basic concepts
MPLS Operation [4]

Forwarding Equivalence Class (FEC)
A group of IP packets which are forwarded in the
same manner (over the same path with the same
forwarding treatment)
 Characteristics:
 Set of IP packets
 Is eventually encoded as the label
 Is not a route or path. However, packets in a
FEC and originating at a given point follow a
route (or one set of routes)
 FEC gives greater control over the forwarding
behavior of the network.
MPLS Architectural description and
basic concepts
MPLS Operation [5]

Unlike traditional IP forwarding which is generally
based strictly on IP addresses and possibly the
diffserv Codepoint, forwarding equivalence
classes can take into account many different
factors:
 Packet’s application protocol
 Packet’s source host
 Link on which the packet arrived
 Quality of service constraints
 Service levels agreements
 Current network conditions
 Virtual private network requirements
MPLS Architectural description and
basic concepts
MPLS Operation [6]


Label Switched Routers (LSRs):
Mapping <incoming interface, label> to <outgoing
interface, label>
Label Switched Paths (LSPs):
Even though the actual label value may change as
a packet travels across a network,the packet’s
path through the network is completely
determined by the initial label the ingress LSR
assigns it. This complete path is known as the
label switched path (LSP).
MPLS Architectural description and
basic concepts
MPLS Operation [7]


Mapping table:
Each router along the path maintains a mapping
table. The table takes an incoming interface and
label value, which then maps it to an outgoing
interface and label value.
Selecting Labels:
When the ingress router assigns an initial label to
a packet, that label determines the packet’s full
path through the MPLS network. Ingress routers
select a label by determining the packet’s
forwarding equivalence class or FEC.
MPLS Architectural description and
basic concepts
MPLS Operation [8]

Distributing labels
 Label distribution is the process by which the
upstream and the downstream router reach the
agreement on the meaning of all MPLS labels
they exchange.
 General principles of label distribution
protocols:
The most important principle is that the
downstream router picks the label value
because it is the only way to ensure that a label
value for an incoming link is unique.
MPLS Architectural description and
basic concepts
MPLS Operation [9]

Even though downstream routers pick label
values, the trigger that generates a new label
can come from either router:
 Downstream unsolicited label distribution
 Downstream on demand label distribution
MPLS Architectural description and
basic concepts
MPLS Operation [10]

Label stacks:
 Label stacks allow the creation of nested label
switched paths, in which one large LSP uses
several smaller LSPs on the way to the
destination. MPLS supports LIFO (last in first
out) for label stacks.
 However, now LSRs have to do some more than
mapping <incoming interface, label> to
<outgoing interface, label>.LSRs must take into
account stack processing.
 “Penultimate hop popping”
QoS Management
Quality of service (QoS):
QoS is defined as those mechanisms that give
network administrators the ability to manage
traffic’s bandwidth, delay and congestion throughout
the network. To realize true QoS, its architecture
must be applied end to end, and not just at the end
or at selected network devices. It is that feature of
the network by which it can differentiate between
different classes of traffic and treat them differently.
QoS Management
Resilience Capabilities [1]




The quality that a customer should receive when
using a service is specified by SLA
Typical SLA QoS-parameters for packet switched
networks:
 packet or cell loss, delay, delay-jitter
 availability of the service
Deterioration of service due to failures of network
equipment (IP/MPLS routers, SDH equipment).
During that time service is unavailable.
Today’s customers put high demands.
QoS Management
Resilience Capabilities [2]


Availability of a network:
The percentage of time that it actually can be used.
 Network congestion availability (NCA)
The percentage of time that the network
between two points is available.
 Service availability
The percentage of time that the service can be
used.
Gradations of availability:
 complete, partial availability, not available
QoS Management
Resilience Capabilities [3]

Survivability of a network:
The ability of providing essential services in the
presence of failures and recover full services in
a timely manner.
Availability is the result of survivability.
 Goal in network design:
To provide end-to-end IP services with high
availability at the lowest possible cost.
QoS Management
Resilience Capabilities [4]

Protection mechanisms:
Used to increase the availability
 Physical protection (use protected physical links)
Consists of routing each of the protected IP links
over two disjoint physical paths (primary and
protection path with the required capacity).
 Duplicated physical required capacity.
 Low cost but provides only protection against
link failures due to fibre cut (not against routers
or router’s interfaces failures).
QoS Management
Resilience Capabilities [5]

IP layer protection (duplication of routers and
physical links)
 Requires that the two IP links be routed over
non-protected but disjoint physical paths.
 Same physical capacity as the previous
method but duplication of routers and router’s
interfaces as well.
 Significantly higher cost due to the cost of
router’s interfaces (full protection implies
duplication of transit routers).
 Drawbacks (not efficient utilization of the
network, long reaction times – IP layer
protocols).
QoS Management
Resilience Capabilities [6]

MPLS protection (by using redundant topology
and MPLS Tunnels for link protection).
 Pre-establish backup MPLS tunnels to protect
critical links and to enable MPLS link protection
with fast restoration on those links.
 Very fast reaction times (comparable to
detection time of IP protocols).
 Keep the effect of the failure within a small
portion of the network.
QoS Management
Resilience Capabilities [7]


Without MPLS the failure would cause updating of
the routing tables in the whole network.
With MPLS full de-loading the LD2_LD1 path, which
would take place in case of IP layer protocol, will be
avoided.
LD1.1
LD2.1
Failed
link
LD1.2
T1.1
LD2.2
Backup MPLS tunnels
(one for each direction)
T2.1
QoS Management
Resilience Capabilities [8]

Global Repair Model (backup LSP utilization)
 The ingress node is responsible for resolving the
restoration.
 One backup path per working path (cost in
terms of recovery time-continuity test for
detection)
LSR2
LSR1
Working path
LSR3
Recovery path
LSR4
LSR5
LSR6
QoS Management
Resilience Capabilities [9]

Local Repair Model
 The restoration procedure starts from the point
of failure.
 Multiple backup paths and a priori reservation of
resources leads to inefficient utilization.
LSR2
LSR4
Recovery path
LSR1
Working path LSR3
LSR5
LSR6
QoS Management
Resilience Capabilities [10]

Reverse Backup
 Redirection of traffic back to the sender and use
of alternate LSP.
 Suitable in network scenarios where the traffic
streams are very sensitive to packet losses.
 Drawback the time needed to reverse.
LSR2
LSR1
Working path LSR3
Recovery path LSR4
LSR5
LSR6
QoS Management
Resilience Capabilities [11]



MPLS vs optical protection
Drawbacks
 MPLS protection switching uses more IP ports
which is expensive.
Benefits
 Better utilization of the fibre capacity.
 More equipment is protected.
QoS Management
Network Dimensioning[1]:


Refers to that part of the network planning process
responsible for the evaluation of resources required
in the network to support the expected amount of
traffic with the requested QoS.
Network elements taken into account:
 Routers
 Switches
 Buffers
 Transmission capacity
QoS Management
Network Dimensioning[2]:

Design issues taken into account:
 Protection scheme to be applied
 Traffic demand
 Routing scheme to be applied
 Traffic classifications
 …
QoS Management
Traffic and QoS measurements[1]


How and which parameters should be monitored to
provide QoS in an MPLS network.
QoS deployment intends to provide a connection
with certain performance bounds from the network
by measuring the following key parameters:
 Bandwidth
 End-to-end delay
 Packet Delay and Jitter
 Packet Loss
QoS Management
Traffic and QoS measurements[2]




Bandwidth: describes the rated throughput capacity
of a given medium, protocol or connection. It
describes the required “size of the pipe”.
End to end delay: is the average time it takes for a
network packet to traverse the network from one
endpoint to the other and is consisted of serialization
delay, propagation delay and switching (queuinginfluence when network is congested) delay.
Jitter: is the variation in the end-to-end delay of
sequential packets.
Packet loss: is measured as the percent of
transmitted packets that never reach the intended
destination.
QoS Management
Traffic and QoS measurements[3]

MMC (measuring, monitoring, control) framework in
the QoS field.
It is the means to provide differentiated service and
to ensure that traffic profiles and SLAs are followed.
Traffic monitoring is the process of observing traffic
characteristics at a given point in the network and
collect traffic information for analysis.
 Investigates which metrics and properties of the
network are the most vital.
 Find appropriate way of measuring these
properties without getting misleading results.
 Evaluate the results and apply appropriate
policies.
QoS Management
Control Actions[1]


Real time QoS management by analyzing the
different control actions that can be activated when
congestion is detected.
Control actions can be invoked for various reasons:
 High load on the link
 New LSPs with higher priority are set up over a
shared resource path pre-empting existing LSPs
with lower priority.
 Equipment or link failure
QoS Management
Control Actions[2]

Possible control actions:
 Protection switching (switching to a backup LSP in
case of failure) ~ms
 Automatic LSPs Rerouting ~sec
 Manually controlled LSPs rerouting ~min
 OSPF weights reconfiguration ~min/hours
 LSPs characteristics modification ~min
QoS Management
Control Actions[3]

Information required for performing control actions.
Control actions could be triggered by one or more
congestion indicators crossing a threshold value.
Control action is useful only if the duration of the
congestion is significantly longer than the control
reaction time.
Two main questions:
 When is the network congested?
 How long this situation is likely to continue?
QoS Management
Control Actions[4]
which congestion
indicator ??
Congestion
threshold
congestion??
congestion Time-span
first detection time

time
Parameters used for Congestion detection
 Packet loss ratio
 Maximum packet delay (for real time traffic)
 Individual flow throughput (for data traffic)
QoS Management
Control Actions[5]

Estimation of congestion duration
 External information: congestion appears after
automatic protection switching and apply of
rerouting mechanisms → equipment failure.
 On the basis of the present and past status of
the network using some predictive models.
 Exponential smoothing techniques
 Predictive models (short term trends)
QoS Management
Control Actions[6]

Suitability of the different control actions:
 LSPs re-routing: move some traffic from the
congested link to under utilized links.
 LSPs policing activation: if the overload is merely
due to MPLS tunnels exceeding their administrative
bandwidth.
 LSPs characteristic modification: modification of
the administrative bandwidth of an MPLS tunnel
(useful to find the actual traffic before rerouting).
 Schedulers re-configuration: tuning of the link
bandwidth to the actual characteristics of the
offered traffic (severe congestion conditions of
valuable traffic).
Requirements for QoS management
systems
MPLS VPNs specific requirements


A VPN is a set of administrative policies that
control both connectivity and QoS among sites.
Area of QoS: the challenge is to support a wide
range of VPN customers:
 Multiple classes of service per VPN
 Decision on which classes of service per VPM
 A class of service provided to an application in
a VPN could be different from the class of
service that the same application uses in
another VPN.
Conclusions
QoS Aspects to be considered:


Resilience:
Resilience is an important aspect of the network.
Besides that a network should provide the
promised QoS when all network elements are
functioning and should also be able to provide
service while failures occur.
Proper network dimensioning of network resources
is the first step required to ensure that the network
is able to fulfill the QoS requirements of the
different services under different operating
conditions.
Conclusions
QoS Aspects to be considered:


Traffic and QoS measurement:
Investigation of the most important metrics and
properties of the network is vital.
Control actions:
Congestion, defined as a situation in which some of
the supported services experience a certain level of
performance degradation. Several control actions
exist to detect and handle these situations.
Conclusions
Why MPLS?






Speed
Scalable
Simple
Traffic engineering
QoS
Support of services
References




IP switching and routing essentials,
Stephen A. Thomas [WILEY,2002].
MPLS and Label Switching Networks,
Uyless Black [Prentice Hall PTR, 2001].
Selected QoS provision in an MPLS/DiffServ
Internet – Saltmamontes, [Eurescom, 2003].
QoS Online Routing and MPLS Multilevel
Protection: A Survey.
Jose L. Marzo, Eusebi Calle, Caterina Scoglio,
Tricha Anjali, IEEE Communications Magazine ,
October 2003.
Ερωτήσεις