1. Integrated and Differentiated
Services Architecture
for Multimedia Applications
over the Internet

2. • Approaches for providing QoS on the
Internet
• Two different service types IntServ and
DiffServ : implementation and problems
• Two-bit differentiated services architecture
• Problems with end-end bandwidth
allocation based on level of marked traffic
• Discussion
Presentation Plan

3. Presentation Plan
• Focus on concepts and reasoning
• Discuss issues & challenges
• Offer personal opinions stimulating
thinking

4. Two Approaches for Providing
QoS on the Internet
1. “Freeway model” – Integrated Services
(IntServ)
– Build a dedicated highway or “circuit”
between communicating points
2. “Doctor’s model” – Differentiated Services
(DiffServ)
– Mark a doctor’s vehicle (e.g. ambulance) or
“packet” to get priority on the road and
limit the percentage of such high-priority
vehicles in the total traffic mix

5. Integrated Services (IntServ)
• A connection-oriented solution
• QoS on a per-flow basis
• Depends on resource reservation
(RSVP)

6. 7
Information Network : Scale of Network
LAN
SD2
SD1
RC1
Ingres Egress
BW1 BW2 BW3
S S
RC2
SD = sender, RC = receiver, S = server, LAN = Local Area Network, BW = Bandwidth
Node1
Node2
Node3
ack
Ingress

7. 8
Quality of Service: View at a glance
Multimedia applications:
network audio and video
network provides
application with level of
performance needed for
application to function.
QoS

8. 9
Multimedia QoS Requirements
• live sources, stored sources
• requirements: deliver data in timely manner
– short end-end delay for interactive multimedia
• e.g., IP telephony, teleconf., virtual worlds
– in time for “smooth” playout
• relaxed reliability
– 100% reliablity not always required

9. 10
Why is QoS so hard?
need session’s input traffic
• must know app’s traffic
demand
(resources, delivery-time constraint,
reliability of information transfer)
To provide performance (delay, jitter, loss) guarantees:
compute session’s output
• Priority of scheduling
(scheduling discipline)

10. 11

11. 12
RSVP – Reservation Protocol
• Reservation is done in one direction
• Receiver-initiated
• The sender sends QoS wanted to the
receiver which sends an RSVP message
back to the sender
• The sender does not need to know the
capabilities along the path or at the
receiver

12. 13
RSVP : logical frame work
LAN
SD2
SD1
RC1
Ingres Egress
BW1 BW2 BW3
S S
RC2
SD = sender, RC = receiver, S = server, LAN = Local Area Network, BW = Bandwidth
Node1
Node2
Node3
Forward resource
Request along the
Communication path
Backward response
To resource request
Ingress

13. 14
Intserv: QoS guarantees
• Resource reservation
– call setup, signaling (RSVP)
– traffic, QoS declaration
– admission control
– QoS-sensitive
scheduling
(e.g., WFQ)
request/
reply

14. 15
QoS Routing
• QoS Routing = Multiple parameter
routing subject to constraints
– Link metrics are vectors
– NP-complete (good heuristics needed)
A
B
C D
E
F
G
H J
K
delay: 10 ms
bandwidth :100
Mb/s
cell loss ratio: 1.0e-6
I

15. Difficulties with IntServ & RSVP
• Scalability
– Keeping a state (and using it!) for each flow
overloads the router
– Periodic messages to refresh the states creates
more traffic
• Router complexity increases (more
computing capabilities required)
• How to satisfy heterogeneous QoS
requirements for different receivers
• Ambitious signaling is not practical

16. Differentiated Services (DiffServ)
• Motivations: no end-to-end signaling
and per-flow state, for scalability,
complexity and quick-to-deploy
reasons.
• Diff-Serv approach: use the TOS field
to sort packets into classes and treat
them differently.
– e.g., one TOS bit to indicate the delay
requirement of the packet, and another to
indicate the drop precedence

17. IP Header (V4) 8 bits

18. 19
Type of Service (TOS) Routing
“low delay”
“high throughput”
Originally does not support real QoS

19. Principles of DiffServ
• Connectionless mode
• QoS for aggregates of traffic, such as
primary/premium, assured and best-effort
• Keep the forwarding path simple to allow
easy and early deployment; push complexity
to network edge
• Avoid strong assumptions about traffic
types; assume the dominant Internet traffic
will be the best-effort traffic in a massive
amount, so there is ample headroom to
provide priority for a relatively small set of
packets

20. How to describe a service
• What is provided to the customer
– E.g., 1 Mbps, continuously available
• To where is this service provided
– A single destination
– A group
– All nodes on local provider
– Everywhere
• Level of assurance provided to
service
– What level of performance uncertainty can
user tolerate

21. Premium service
• Provides guaranteed peak bandwidth service
with negligible delay/jitter
• “Virtual leased lines”: don't waste
bandwidth when circuits are unused
• Useful for voice
• is given its own high-priority queue in
routers; takes a small portion of network
bandwidth
• Shaped and hard-limited to its provisioned
peak rate

22. Mechanism for premium
service
Host First-hop
Intra-network
Router
H-Q: premium, no dropping
L-Q: best effort, dropping on
congestion

23. Assured service
• Provides an expected level of bandwidth
with delay
• Characteristic slightly better than best-
effort
• Useful for e.g., connections to search
engines
• Permits flows to use any additional
available bandwidth
• Requires buffer management, e.g., RED

24. Mechanism for assured service
Host First-hop
Counter
Counter
Out- and in- dropper
RIO scheme, packets
are treated preferentially
Marking packets
according to the service profile

25. RIO algorithm
• RED - Random Early Detection
– Packets dropped with low but
increasing probability as queue grows;
instead of waiting until it is full and
dropping all new packets
• RIO
– Run two RED algorithms for “in” and
“out” with different dropping
frequencies

26. Usage Example of DiffServ

27. Components of DiffServ
(1) Edge node algorithms
Classification, policing, shaping,
metering, marking, etc.
(2) Router algorithms
Packet discard algorithms: RED-like
methods for in-profile and outof-
profile traffic
Priority, low-latency queueing

28. Forwarding Path

29. Access Router
• Packet Classifier: classifies the packets based on
their flow-id
• Marker:
– being implemented as tocken bucket
– checks the conformance of arriving packets and marks
them accordingly

30. Marker Implementation
Premium Marker Assured
Marker

31. Core Router

32. Illustration of Discard Policy

33. Closure
• DiffServ is sort of a gamble. It may
actually work in the market place. We
are all learning!
• Initially, probably only few traffic classes
will need to be deployed. Deployment can
start when ISP routers have been made
TOS aware
• Access and boundary routers need
processing power per packet
• Network management is needed for
resource control (on-going work)

34. Assignment
• Compare ATM’s CBR with DiffServ’s
Premium Service
• Compare ATM’s ABR with DiffServ’s
Assured Service

35. We are all learning, so keep
improving your knowledge