Professional Documents
Culture Documents
Service
Support
QoS #1
QOS in IP Networks
q IETF groups are working on proposals to provide
QOS control in IP networks, i.e., going beyond
best effort to provide some assurance for QOS
q Work in Progress includes RSVP, Differentiated
Services, and Integrated Services
q Simple model
for sharing and
congestion
studies:
QoS #2
Principles for QOS Guarantees
QoS #3
Principles for QOS Guarantees (more)
QoS #4
Principles for QOS Guarantees (more)
QoS #5
Principles for QOS Guarantees (more)
QoS #6
QoS #7
Building blocks
q Scheduling
m Active Buffer Management
q Traffic Shaping
m Leaky Bucket
m Token Bucket
q Modeling
m The (σ,ρ) Model
m WFQ and delay guarantee
q Admission Control
m QoS Routing
QoS #8
Scheduling: How Can Routers Help
q Packet dropping:
m not drop-tail
m not only when buffer is full
• Active Queue Management
q Congestion signaling
m Explicit Congestion Notification (ECN)
QoS #9
Buffer Size
q Small buffers:
m often drop packets due to bursts
m but have small delays
q Large buffers:
m reduce number of packet drops (due to bursts)
m but increase delays
q Basic premise:
m router should signal congestion when the queue first
starts building up (by dropping a packet)
m but router should give flows time to reduce their sending
rates before dropping more packets
m Note: when RED is coupled with ECN, the router can
simply mark a packet instead of dropping it
QoS #11
RED
q FIFO scheduling
q Buffer management:
m Probabilistically discard packets
m Probability is computed as a function of average queue
length (why average?)
Discard Probability
0
min_th max_th queue_len Average
Queue Length
QoS #12
RED (cont’d)
Discard
Discard Probability (P)
0
min_th max_th queue_len Average
Queue Length
Enqueue
Discard/Enqueue
probabilistically
QoS #13
RED (cont’d)
avg_len - min_th
P = max_ P
max_th - min_th
Discard Probability
max_P 1
P
0
min_th max_th queue_len Average
avg_len Queue Length
QoS #14
Average vs Instantaneous Queue
QoS #15
RED and TCP
q Fairness in drops
m Bursty versus non-Bursy
m Probability of drop depends on rate.
q Disadvantages
m Many additional parameters
m Increasing the loss
QoS #16
RED Summary
q High network utilization with low delays when flows are long lived
q Does not work well for short lived flows (like Web traffic)
m Dropping packets in an already short lived flow is devastating
QoS #17
Traffic Shaping
QoS #18
The Leaky Bucket Algorithm
(a) A leaky bucket with water. (b) a leaky bucket with packets.
QoS #20
Token Bucket Algorithm
QoS #21
The Token Bucket Algorithm
5-34
p3 (1) p2 6-5=1 p1
4-2-1=1 p3,p2
QoS #23
Leaky Bucket vs Token Bucket
QoS #24
The (σ,ρ) Model
q Parameters:
m The average rate is ρ.
m The maximum burst is σ.
q (σ,ρ) Model:
m Over an interval of length t,
m the number of packets/bits that are admitted
m is less than or equal to (σ+ρt).
m Composing flows (σ1,ρ1) & (σ2,ρ2)
• Resulting flow (σ1+ σ2,ρ1+ρ2)
q Token Bucket Algorithm:
m σ = MaxTokens & ρ=1/r per time unit
q Leaky Bucket Algorithm
m σ = 0 & ρ= =1/r per time unit
QoS #25
Using (σ,ρ) Model for admission Control
q What does a router need to support
streams: (σ1,ρ1) … (σk,ρk)
m Buffer size B > Σ σi
m Rate R > Σ ρi
QoS #26
Delay Bounds: WFQ
q Corollary:
m For any packet p and link rate R
• Let Time(p,S) be its completion time in policy S
• Then Time(p,WFQ)-Time(p,GPS) ≤ Lmax/R
QoS #27
Parekh-Gallagher theorem
Suppose a given connection is (s,r) constrained,
has maximal packet size L, and passes
through K WFQ schedulers, such that in the
ith scheduler
m there is total rate r(i)
m from which the connection gets g(i).
Let g be the minimum over all g(i), and suppose
all packets are at most Lmax bits long. Then
s
L k
L k
end - to - end delay £ + å +å
g i =1 g (i ) i =1 r (i )
QoS #28
P-G theorem: Interpretation
s L k k
L
end - to - end delay £ + å +å
g i =1 g (i ) i =1 r (i )
Delay of last packet of
a burst. Only in
bottleneck node
GPS term
store&forward penalty WFQ lag behind
GPS: each node
QoS #30
Fine Points
QoS #31
Approaches to QoS
m RSVP
QoS #32
IETF Integrated Services
QoS #33
Intserv: QoS guarantee scenario
q Resource reservation
m call setup, signaling (RSVP)
m traffic, QoS declaration
m per-element admission control
request/
reply
m QoS-sensitive
scheduling (e.g.,
WFQ)
QoS #34
Call Admission
QoS #35
RSVP request (T-Spec)
QoS #36
RSVP request (R-spec)
QoS #38
Example of QoS Routing
5 D=
=5 10, B
W W=
4, B 20
D=
A
90
D
=5 B
W=
,B
W 0 D = 14, BW = 40
5, B
=9 =6
0 W
5, B
D=
D=
W = 90
D = 7, B
D
=5
,B
D = 10, BW = 90
W
=9
= 45
, BW
0
D=3
Diffserv approach:
q simple functions in network core, relatively
complex functions at edge routers (or hosts)
q Don’t define service classes, provide functional
components to build service classes
QoS #40
Diffserv Architecture
Edge router:
- per-flow traffic management
r marking
scheduling
- marks packets as in-profile
and out-profile b ..
.
Core router:
- per class traffic management
- buffering and scheduling
based on marking at edge
- preference given to in-profile
packets
QoS #41
Edge-router Packet Marking
q profile: pre-negotiated rate A, and token bucket size B
q packet marking at edge based on per-flow profile
Rate A
User packets
QoS #42
Classification and Conditioning
QoS #43
Classification and Conditioning
QoS #44
Forwarding (PHB)
QoS #45
Forwarding (PHB)
QoS #46
DiffServ Routers
DiffServ
Edge
Router
Classifier Marker Meter Policer
DiffServ PHB
Select PHB PHB Local
Core PHB
PHB conditions
Router
Extract Packet
DSCP treatment
QoS #47
IntServ vs. DiffServ
IP
IP
IntServ
network
DiffServ
network
Intserv Diffserv
Granularity of service Individual Flow Aggregate of
differentiation flows
State in routers(e.g. Per Flow Per Aggregate
scheduling, buffer
management)
Traffic Classification Several header fields DS Field
Basis
Type of service Deterministic or Absolute or
differentiation statistical guarantees relative
assurance
Admission Control Required Required for
absolute
differentiation
Signaling Protocol Required(RSVP) Not required for
relative schemes
QoS #49
Comparison of Intserv & Diffserv
Architectures
Intserv Diffserv
Coordination for End-to-End Local (Per-Hop)
service differentiation
Scope of Service A Unicast or Multicast Anywhere in a
Differentiation path Network or in
specific paths
Scalabilty Limited by the number Limited by the
of flows number of classes
of service
Network Accounting Based on flow Based on class
characteristics and QoS usage
requirement
Network Management Similar to Circuit Similar to existing
Switching networks IP networks
Interdomain Multilateral Bilateral
deployment Agreements Agreements
QoS #50