Lecture 7 - Quality-Of-Service Routing

Quality-of-Service Routing
Course 34355 Routing in Data Networks

March 14, 2016
www.com.dtu.dk
Take home points
After the lectures you should:
Be familiar the concept of QoS
Understand some implication of QoS requirements on routing
Be familiar with QoS issues in both traditional unicast routing and
other scenarios
Know about QoS support in the Internet today
Know about some QoS routing algorithms and protocols
Key terms
QoS, Constraints, ISA, Diffserv, OSPF-TE
2 DTU Fotonik, Technical University of Denmark
www.com.dtu.dk
Outline of lecture
What is Quality-of-Service (QoS)?
Local and global state information
QoS Routing
Distribution of state information
Path finding algorithms
Using QoS in the Internet
Selected QoS protocols
Lecture material:
Huitema: Chapter 14
D. Katz et al, "RFC 3630 - Traffic Engineering (TE) Extensions to OSPF
Version 2", IETF, September 2003
www.com.dtu.dk
Example Applications setting new requirements to the
network
Videoconferencing
Interactive service
Video on Demand
Control signals should experience low delay
Distributed music recording
Low latency a must
High Quality video production with remote studios
Platform independent gaming transmission of display information.
Backup and restore
Huge bandwidth pipes required on demand
www.com.dtu.dk
Network requirements for typical research
application
Application Quality BW Latency Packet loss Mode
jitter
Storage and High 120-600 200 ms / - < 0.1% Unidirectional
backup Mbps
Accelerated VoD Very high 50-100 Mbps 500 ms / < 1% Unidirectional
streaming 50 ms
Uncompressed Very high 300-1500 150 ms / 1 < 1% Bidirectional
video production Mbps ms
Multipoint video High 2-13 Mbps 150 ms / < 1% Bidirectional
conferencing pr. partner 50 ms
Best effort IP networks cannot satisfy

these requirements as no
guarantees for delay are provided
www.com.dtu.dk
Schemes for providing QoS
Increasing bandwidth
Bandwidth affordable
Overprovisioning
No guarantees issued for any applications
Restricting access
Prioritisation of data
Guaranteed bandwidth supported
Delay bounds introduced
Dynamic bandwidth allocation
Improve utilisation while prioritising access
Integration of the applications with the network
Clear interfaces between heterogenous network domains
www.com.dtu.dk
Quality of Service (QoS)
What is network Quality-of-Service (QoS)?
Very general concept No simple/unique definition!
Here today:
Network service, where users can specify requirements of service
(e.g., in terms of different parameters)
Parameters can be: Throughput, Delay, Jitter, Reliability, Cost,
Service can be guaranteed or approximate
QoS must be measurable
Traditional best-effort service is opposite of QoS
www.com.dtu.dk
Qualification of QoS parameters
www.com.dtu.dk
Why QoS?
Applications that require QoS support:
Real-time applications, e.g., VoIP and video/audio-streaming
Bounds on delay, delay-jitter and throughput
Applications that can benefit from QoS support:

Interactive applications, e.g. remote login
Distributed databases
In general: All applications can benefit from QoS, but not all are worth it!
QoS more expensive than best-effort
www.com.dtu.dk
History of QoS
IPv4 specification defines Type-of-Service field in header:
Version IHL Type of Service Total length

Identification Flags Fragment offset
Time to live Protocol Header checksum
Source IP address
Destination IP address
The type of service is used to specify the treatment of

the datagram during its transmission through the
internet system
TOS definition from RFC-791 :
www.com.dtu.dk
IP TOS Type of Service
Problem with QoS support using TOS:
Use of TOS field requires extra processing in routers

Early applications (Email, FTP, ) did not require QoS
Routers ignored TOS field!
Early attempt at QoS in the Internet failed!
Still no guarantees
www.com.dtu.dk
History of QoS (2)
Next attempt at QoS: ATM and ATMForum
www.com.dtu.dk
History of QoS (3)
ATM research resulted in many papers, book, conferences, ...
Many different companies were started on ATM
A lot of knowledge on QoS was accumulated
Experience used by MPLS
Knowledge is starting to flow into IP community
www.com.dtu.dk
QoS support in routers
Example from Integrated Services Architecture (see later)
www.com.dtu.dk
QoS Support in routers (2)
Different queues for different service categories
Intelligent scheduling queues are served in order of their QoS

requirements
Example: Queues for flow with delay bounds must be kept short, i.e.,
queue is served often
QoS-support requires some form of admission control
Scheduling: Oldest Cell First, Fairness etc.
www.com.dtu.dk
Local state
Example of Link state
Bandwidth (B)
Delay (D)
Cost (C)
Link attribute: (B,D,C)
Example of node state
Queueing delay
CPU cost
Processing delay
Network state
All links include QoS states
All nodes include QoS state information
www.com.dtu.dk
From local states to routing decision
(1,5,...)
2 4
(1,1,...)
(4,1,...)
1 6
(6,1,...)
3 5
Given:
Graph representation of links (local state)
Multiple metrics per link
(Bandwidth, delay, ... , .. )
Find:
Path from 1 to 6 with lowest delay and min BW 2 ???
www.com.dtu.dk
QoS Routing Some definitions
Link constraints: Constraints on the use of links
Example: Every link must have capacity > R
Path constraints: Constraints on the end-to-end path

Example: End-to-end delay must be smaller than D
Feasible path: Path that satisfies all link constraints and path constraints
Goal of QoS routing: Find at least one feasible path, given a set of
constraints, specified by users
www.com.dtu.dk
QoS Routing Some definitions (2)
Types of QoS routing:
Link-constrained routing
Example: Find path with throughput > R
Link-optimised routing
Example: Find path with maximum throughput
Path-constrained routing:
Example: Find path where delay < D
Path-optimised routing
Example: Find path with minimum delay
Combinations also possible, e.g., path-constrained link-optimized routing
www.com.dtu.dk
QoS Routing Link metrics
QoS link metrics can be:
Additive: Delay, Jitter, Cost, (Packet loss probability), ...
Non-additive: Available capacity, Policies, ...
Non-additive link metrics are simple: Remove links that do not conform to
(link) constraints topology filtering
Additive link metrics are harder

Single (additive) constraint Traditional shortest-path routing
Multiple (additive) constraints Possibly NP-complete problem
www.com.dtu.dk
Local state to global state what you did
Conversion of problem to weighted graph
Examining non-additive constraints
Finding those not satisfying constraints
Removing those
Result: New weighted graph
Shortest path routing techniques based on one additive constraint
Problems?
Scalability
Convergence time
www.com.dtu.dk
Hierarchical routing
Network divided into subdomains

Other domains (except parents) abstracted
Advantage: Scalable
Drawback: Imprecision
www.com.dtu.dk
QoS Routing Multiple Constraints Problem
(1)
Given graph with m additive metrics (w1, w2, ..., wm) on every link
Given constraints Li for i = 1 ... m
Objective: Find path P such that:
Fori 1m: w (link) L

alllinksin P
i i
In general: More than one path may be feasible

MCOP problem: From the set of feasible paths, determine the
optimal path
For instance, the path (among the set of feasible path) that minimizes the
cost
www.com.dtu.dk
QoS Routing MCP problem (2)
MCP problem can be NP-complete
Depends on the range and relationship between metric components
Exact algorithms only useful for small networks
Approximative
algorithms,
e.g., Jaffes
or Fallback
or
www.com.dtu.dk
Optimisation Programming Language (OPL)
int nbLinks = ...;
int nbFlows = ...;
int nbNodes = ...;
5
5 range Links 1..nbLinks;
3
2 3 range Flows 1..nbFlows;
3 4
4 range Nodes 1..nbNodes;
2 6
2 3
2 2 3
1
1 2 // Topology of Network:
2 int+ u[Links] = ...;
1 1
1
1 4 5 int+ v[Links] = ...;
1
float+ linkbw[Links] = ...;
float+ cost[Links] = ...;
// Flows:
float+ effbw[Flows] = ...;
int+ s[Flows] = ...;
int+ d[Flows] = ...;
float+ h[Flows] = ...;
var int x[Flows,Links] in 0..1;

OPL constraints Minimize
sum(l in Links)
cost[l] * ( sum(i in Flows) effbw[i] * x[i,l] )

Constraint (1) ensures that link
capacity is not exceeded. subject to
{
forall(l in Links) // (1)
Constraint (2) puts a maximum sum(i in Flows) effbw[i] * x[i,l] <= linkbw[l];
on the number of hops in the
path of a given flow. forall(i in Flows) // (2)
sum(l in Links) x[i,l] <= h[i];
Constraints (3) and (4) bind the forall(n in Nodes) // (3)
flows to their source and forall(i in Flows : s[i] = n)
sum(l in Links : u[l] = n) x[i,l] = 1;

destination, respectively.
forall(n in Nodes) // (4)
Finally, constraint (5) ensures forall(i in Flows : d[i] = n)

sum(l in Links : v[l] = n) x[i,l] = 1;
that a flow, which enters a
node, will also leave that node forall(n in Nodes) // (5)
as long as the node is neither forall(i in Flows : s[i] <> n & d[i] <> n)
source nor destination for that sum(l in Links : u[l] = n) x[i,l] =

sum(l in Links : v[l] = n) x[i,l];
flow. };

QoS Routing Multicast
Multicast QoS routing introduces additional issues:
Bottleneck link may only affect some of the receivers
Delays usually different from receiver to receiver
Tree analogy to shortest path: Steiner Tree

Tree with branches to every receiver that minimizes the sum of the
metric of the included branch
QoS examples:
Constrained S.T.: Bounded delay from source to every node
www.com.dtu.dk
QoS Routing Ad-hoc networks
QoS in ad-hoc networks difficult because:
Dynamic topology means imprecise information for routing
Overhead is likely to be higher than in wireline networks
Topology changes too fast QoS might not be possible
Topology changes slowly QoS might be possible

except for transient periods
Soft QoS instead of hard QoS
Soft QoS might be tolerated if applications can compensate in
transient periods, i.e., by reducing encoder quality
Problem of resiliency
www.com.dtu.dk
QoS Routing Interdomain QoS routing
Problem with interdomain/hierarchical QoS Routing:
Domain hides internal topology State information is incomplete
Accurate QoS routing not possible
Possible solution (from RFC-2386):

End-to-end QoS is composed from QoS in visited ASs
Border router must determine available QoS for internal destinations
Can be used directly in advertisements of internal destinations
Advertisements of external destinations uses information from other
border routers
QoS metrics should not depend on instantaneous values but instead
on the expected traffic flows service
www.com.dtu.dk
Example of interdomain issues
Client
network
Client
network UNI1.0 A
D Optical network
A
R2
Optical network
B
UNI1.0
R2
I-NNI
I-NNI
E-
NNI UNI
Carrier domain
Client UNI1.0 UNI1.0 Client
network R2 network
C
R2 B
www.com.dtu.dk
QoS Routing Related issues
Admission control: New connection/flow rejected if QoS constraints can
not be met
QoS negotiation: QoS parameters may be negotiated

User requests one set of QoS parameter. Network cant establish path
with required QoS; suggest less stringent set of QoS parameters.
User may accept or reject
Resource reservations
Traffic contract / Policing
www.com.dtu.dk
QoS in the Internet
Two approaches directly aimed at QoS in the Internet:
Integrated Services Architecture (ISA) (Intserv)
Differentiated Services (DiffServ)
Other approaches:
MPLS-TE
Keep adding capacity
www.com.dtu.dk
Integrated Services Architecture (1)
Uses (soft-state) resource-reservation at routers
Resources are reserved per flow
Signalling using the RSVP protocol
Path messages from sender to receiver
Resv messages from receiver to sender
Receiver initated reservations
www.com.dtu.dk
Integrated Services Architecture (2)
Advantages and disadvantages:
Application requirements are matched (assuming that resources are
available!), i.e., the application gets what it asks for
Reservations can be shared in multicast groups (reservation merging)
Requires state-information per flow in routers, i.e., scalability issues
in core routers
RSVP protocol requires capacity for RSVP messages
Routers must determine for each IP packet the flow it belongs to. This
classification can be performance intensive in routers
Easier in IP version 6 compared to version 4
www.com.dtu.dk
Diffserv
Users have Service-Level Agreement (SLA) with network provider
Inside DS domains, the network provider provides a number of

different per-hop behaviors (PHB)
www.com.dtu.dk
Diffserv (2)
Customer selects specific PHB for packet treatment, indicated in the DS
field of IP packets
DS field was the original ToS field
Routers use contents of DS field to determine PHB (Simple lookup)

Traffic classification and conditioning performed at access routers, i.e.,
at the edge of the DS domain
PHBs may be different in different DS domains, i.e., translation between
different DS values may be required at egress routers
QoS is provided for aggregated traffic
www.com.dtu.dk
Diffserv (3)
Advantages and disadvantages:
Better scalability in core part
Simplified processing in router
Only finite number of PHBs Application requirements not exactly
matched
www.com.dtu.dk
Intserv and Diffserv
ISA has scalability problems, Diffserv has granularity
problems
Idea: Why not combine ISA and Diffserv
Border routers intercept RSVP messages

RSVP messages are dropped if combined flows would exceed SLA
RSVP messages traverses DiffServ domain like normal traffic
If accepted, packets in flow are marked with corresponding DSCP
www.com.dtu.dk
OSPF Traffic Engineering
Opaque Link State Advertisement (LSA)
Includes traffic metric
Maximum bandwidth
Reservable bandwidth
Unreserved bandwidth
Flooded when changes occur
E.g. when unreserved bandwidth changes by e.g. 10%
Why not every 2%?

Why not every 50%?
www.com.dtu.dk
OSPF-TE Lower limit of setup times (modelling
study)
Dependent on convergence times of routing protocols
Routing protocols for constraint based routing: (among others)
CSPF
OSPF-TE
Flooding procedure
When release/occupation of 10% of available link bandwidth
www.com.dtu.dk
Convergence time in large network
Small networks: 1 second convergence

Large network: 9-10 second convergence
=> Lower limit of LSP scales: 10 seconds.
www.com.dtu.dk
OSPF-TE/MPLS/RSVP Modelling study
Enabling traffic engineering in
the network
For two large traffic flows
Packet loss reduced
Delay reduced
Delay wo/w TE
Network scenario
42 DTU Fotonik, Technical University of Denmark Packet loss wo/w TE

www.com.dtu.dk
Summary
Definition of QoS and some QoS history
QoS routing:
Definition of local states
From local states to a network global state
QoS routing for multicast, ad-hoc networks and interdomain
ISA and Diffserv approaches to QoS in the Internet
OSPF TE a case study
www.com.dtu.dk

Lecture 7 - Quality-Of-Service Routing

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Lecture 7 - Quality-Of-Service Routing

Transféré par

Droits d'auteur :

Formats disponibles

Quality-of-Service Routing

Course 34355 Routing in Data Networks

2 DTU Fotonik, Technical University of Denmark

3 DTU Fotonik, Technical University of Denmark

4 DTU Fotonik, Technical University of Denmark

Best effort IP networks cannot satisfy

6 DTU Fotonik, Technical University of Denmark

Very general concept No simple/unique definition!

7 DTU Fotonik, Technical University of Denmark

8 DTU Fotonik, Technical University of Denmark

Applications that can benefit from QoS support:

9 DTU Fotonik, Technical University of Denmark

Version IHL Type of Service Total length

The type of service is used to specify the treatment of

10 DTU Fotonik, Technical University of Denmark

Use of TOS field requires extra processing in routers

Routers ignored TOS field!

Early attempt at QoS in the Internet failed!

11 DTU Fotonik, Technical University of Denmark

12 DTU Fotonik, Technical University of Denmark

Knowledge is starting to flow into IP community

13 DTU Fotonik, Technical University of Denmark

14 DTU Fotonik, Technical University of Denmark

Intelligent scheduling queues are served in order of their QoS

QoS-support requires some form of admission control

Scheduling: Oldest Cell First, Fairness etc.

15 DTU Fotonik, Technical University of Denmark

16 DTU Fotonik, Technical University of Denmark

17 DTU Fotonik, Technical University of Denmark

Path constraints: Constraints on the end-to-end path

18 DTU Fotonik, Technical University of Denmark

Combinations also possible, e.g., path-constrained link-optimized routing

19 DTU Fotonik, Technical University of Denmark

Additive link metrics are harder

20 DTU Fotonik, Technical University of Denmark

21 DTU Fotonik, Technical University of Denmark

Network divided into subdomains

22 DTU Fotonik, Technical University of Denmark

Objective: Find path P such that:

Fori 1m: w (link) L

In general: More than one path may be feasible

23 DTU Fotonik, Technical University of Denmark

24 DTU Fotonik, Technical University of Denmark

var int x[Flows,Links] in 0..1;

25 DTU Fotonik, Technical University of Denmark

cost[l] * ( sum(i in Flows) effbw[i] * x[i,l] )

sum(l in Links) x[i,l] <= h[i];

Constraints (3) and (4) bind the forall(n in Nodes) // (3)

flows to their source and forall(i in Flows : s[i] = n)

sum(l in Links : u[l] = n) x[i,l] = 1;

Finally, constraint (5) ensures forall(i in Flows : d[i] = n)

source nor destination for that sum(l in Links : u[l] = n) x[i,l] =

26 DTU Fotonik, Technical University of Denmark

Tree analogy to shortest path: Steiner Tree

27 DTU Fotonik, Technical University of Denmark

Topology changes too fast QoS might not be possible

Topology changes slowly QoS might be possible

28 DTU Fotonik, Technical University of Denmark

Possible solution (from RFC-2386):

29 DTU Fotonik, Technical University of Denmark

30 DTU Fotonik, Technical University of Denmark

QoS negotiation: QoS parameters may be negotiated

Traffic contract / Policing

31 DTU Fotonik, Technical University of Denmark