CS 552

Computer Networks
Quality Of Service
Richard Martin
Credit slides by B. Nath, I. Stoica

1
 Outline

• What is Quality of Service

• Basic mechanisms
– Leaky and token buckets
• Integrated Services (IntServ)
• Differentiated Services (DiffServ)
• Economics and Social factors facing QoS
• QoS Vs. Over Provisioning

2
 Best Effort vs. QoS

• Best Effort:
• You get a link to the Internet with at most B bits/sec.
• If you don’t like it, switch to another provider.
• Quality of Service (QoS)
– We provide you some kind of guarantees for:
• Bandwidth
• Latency
• Jitter
– I.e., network is engineered to provide some
Quality beyond “Not to exceed B bits/s”

3
 QoS’s Quest

The Holy Grail of computer networking is to

design a network that has the flexibility and
low cost of the Internet, yet offers the end-to-
end quality-of-service guarantees of the
telephone network.
--S. Keshav

4
 Two Styles of QoS

• Worse-case
– Provide bandwidth/delay/jitter guarantee to every
packet
– E.g., “hard real time”
• Average-case
– Provide bandwidth/delay/jitter guarantee over
many packets
– Statistical in nature
– E.g. “Soft real time”

5
 Resource Reservation: Example

Src Router Router Dest

10 Mbps 4 Mbps 6 Mbps
available available available

Case 1: Source attempts to connect to destination, and attempts to

reserve 4 Mbps for the connection

Result: Connection accepted. There is enough bandwidth

available. Available link bandwidths updated.

Case 2: Source attempts to connect to destination, and attempts to

reserve 5 Mbps for the connection

Result: Failure. There is not enough bandwidth available on

one of the links.
6
 Resource Reservation (cont’d)

• Once a connection is accepted, the host must use

only the amount of resources reserved. It may not
use more than that.
• What if the host is malicious and attempts to use
more network resources than it reserved?

7
 Leaky Bucket

• Used in conjunction with resource reservation

to police the host’s reservation
• At the host-network interface, allow packets
into the network at a constant rate
• Packets may be generated in a bursty
manner, but after they pass through the leaky
bucket, they enter the network evenly spaced

8
 Leaky Bucket: Analogy
Packets from host

Leaky
Bucket

Network 9
 Leaky Bucket (cont’d)

• The leaky bucket is a “traffic shaper”: It changes the

characteristics of packet stream
• Traffic shaping makes more manageable and more
predictable
• Usually the network tells the leaky bucket the rate at
which it may send packets when the connection
begins
• Polices the average rate

10
 Leaky Bucket:
Doesn’t allow bursty transmissions

• In some cases, we may want to allow short

bursts of packets to enter the network without
smoothing them out
• For this purpose we use a token bucket,
which is a modified leaky bucket

11
 Token Bucket
• The bucket holds tokens instead of packets
• Tokens are generated and placed into the token bucket at a
constant rate
• When a packet arrives at the token bucket, it is transmitted if
there is a token available. Otherwise it is buffered until a token
becomes available.
• The token bucket has a fixed size, so when it becomes full,
subsequently generated tokens are discarded

12
 Token Bucket
Packets from host

Token Generator
(Generates a token
once every T seconds)

Network 13
Token Bucket vs. Leaky Bucket

Case 1: Short burst arrivals

Arrival time at bucket

0 1 2 3 4 5 6

Departure time from a leaky bucket

0 1 2 3 4 5 6 Leaky bucket rate = 1 packet / 2 time units
Leaky bucket size = 4 packets

Departure time from a token bucket

Token bucket rate = 1 tokens / 2 time units
0 1 2 3 4 5 6
Token bucket size = 2 tokens

14
Token Bucket vs. Leaky Bucket

Case 2: Large burst arrivals

Arrival time at bucket

0 1 2 3 4 5 6

Departure time from a leaky bucket

0 1 2 3 4 5 6 Leaky bucket rate = 1 packet / 2 time units
Leaky bucket size = 2 packets

Departure time from a token bucket

Token bucket rate = 1 token / 2 time units
0 1 2 3 4 5 6
Token bucket size = 2 tokens

15
 Flow Specification: Token Bucket

• Characterized by two parameters (r, b)

– r – average rate
– b – token depth
• Assume flow arrival rate <= R bps (e.g., R link capacity)
• A bit is transmitted only when there is an available token
• Arrival curve – maximum amount of bits transmitted by time t

Arrival curve
r bps bits

slope r
b
b bits
slope R

<= R bps
time
regulator 16
 Quality of service issues

• Flow specification
– Flow spec: traffic characteristics, QoS requirements (delay,
jitter,bandwidth)
• Routing
– Routing traffic to best meet demand
• Resource reservation
– End-host signaling to network QoS resource requirements
• Admission control
– Limiting number of reservations
• Packet scheduling
– Packet by packet scheduling (fairness, delay)
• RSVP addresses reservation
17
 Integrated Services Example: Data Path

• Per-flow classification
Receiver
Sender

18
 Integrated Services Example: Data Path

• Per-flow buffer management

Receiver
Sender

19
 Integrated Services Example

• Per-flow scheduling

Receiver
Sender

20
 How Things Fit Together

Routing RSVP
Routing RSVP
Messages messages

Control Plane
Admission
Control

Data Plane
Forwarding Table Per Flow QoS Table

Data In
Route Lookup Classifier Scheduler Data Out

21
 Service Classes

• Multiple service classes

• Service: contract between network and
communication client
– End-to-end service
– Other service scopes possible
• Three common services
– Best-effort (“elastic” applications)
– Hard real-time (“real-time” applications)
– Soft real-time (“tolerant” applications)

22
 Worse-case : Guaranteed Services

• Service contract
– Network to client: guarantee a deterministic upper
bound on delay for each packet in a session
– Client to network: the session does not send more
than it specifies
• Algorithm support
– Admission control based on worst-case analysis
– Per flow classification/scheduling at routers

23
 Average-case: Controlled Load Service

• Service contract:
– Network to client: Average delay, jitter, bandwidth, e.g., makes
network appear as an unloaded, best effort network with
bandwidth and delay
– Client to network: the session does not send more than it
specifies
• Algorithm Support
– Admission control based on measurement of aggregates
– Scheduling for aggregate possible

24
 Role of RSVP in the Architecture

• Signaling protocol for establishing per flow

state
• Carry resource requests from hosts to
routers
• Collect needed information from routers to
hosts
• At each hop
– Consult admission control and policy module
– Set up admission state or informs the requester
of failure

25
 RSVP Design Features

• IP Multicast centric design

• Receiver initiated reservation
• Different reservation styles
• Soft state inside network
• Decouple routing from reservation

26
 RSVP Reservation Model

• Performs signaling to set up reservation state

for a session
• A session is a simplex data flow sent to a
unicast or a multicast address, characterized
by
– <IP dest, protocol number, port number>
• Multiple senders and receivers can be in
session

27
The Big Picture

Network
Sender

PATH Msg

Receiver

28
The Big Picture (2)

Network
Sender

PATH Msg

Receiver

RESV Msg

29
 RSVP terminology

• Flow descriptor (Flow spec + Filter Spec)

• Flow spec (Rate, max burst)
– Sender can Explicitly specify flow spec or not specify
• Filter Spec (Sender address, TCP/UDP, Port#)
– Aids in combining similar flows
– Filter can be shared (SE-style) or can use wild cards (all senders
on a given port or a given sender on all ports, etc)
– The style may be shared or distinct in a sense that all reservations
may be handled as one single reservation or there may be a single
reservation for each upstream sender respectively.

30
 RSVP Basic Operations

• Sender: sends PATH message via the data delivery path

– Set up the path state each router including the address of
previous hop
• Receiver sends RESV message on the reverse path
– Specifies the reservation style, QoS desired
– Set up the reservation state at each router
• Things to notice
– Receiver initiated reservation
– Decouple routing from reservation
– Two types of state: path and reservation

31
 RSVP messages

• PATH message – sets up state along path

followed by packets
• RESV message – request for reservation
back along setup path path
• PATH_TEAR, RESV_TEAR,
RESV_CONFIRM, RESV_ERROR,
PATH_ERROR

32
RSVP Operation

Sender

Merged reservations

Merged reservations

Receiver3
Receiver1 Receiver2

33
 RSVP PATH MESSAGE

• From sender to receiver (unicast or multicast)

• Intercepted at each RSVP aware hop
• Includes
– Sender TSpec : Traffic characteristics of the sender
• Token bucket rate, depth, max flow rate, max packet size
• forms one side of the ``contract'' between the data flow and the service.
– F-flag: specify whether filtered reservation is allowed
• Routers store:
– Path state, i.e., PHOP address to previous hop (RSVP aware node)
– If F-flag is set, store sender and its flowspec
– Otherwise, just add new link to multicast tree

34
 RSVP RESV MESSAGE

• From receiver to sender(s) to reserve resources

• Sent hop-by-hop using PHOP information
• Reservation style and flow description
– Reservation style (FF,SE, WF)
– Fixed-filter, Shared-explicit, wildcard-filter
– Senders to which the reservation applies
– Rspec, QoS specific requirements
– RSpec is highly specific to the service required, and may include
information like bandwidth allocation, maximum delay, or packet loss
probabilities etc.
• RESV messages processing at each hop
– Merging of RESV messages
– Forwards resv messages using PHOP

35
 Route Pinning

• Problem: asymmetric routes

– You may reserve resources on RS3S5S4S1S, but
data travels on SS1S2S3R !
• Solution: use PATH to remember direct path from S to
R, i.e., perform route pinning
S2 R
S
S1 S3

IP routing
PATH S4 S5
RESV
36
 How Is the Token Bucket Used?

• Can be enforced by
– End-hosts (e.g., cable modems)
– Routers (e.g., ingress routers in a Diffserv domain)
• Can be used to characterize the traffic sent
by an end-host

37
 Source Traffic Characterization

• Arrival curve – maximum amount of bits transmitted

during an interval of time Δt
• Use token bucket to bound the arrival curve

bps bits
Arrival curve

time Δt

38
 QoS Guarantees: Per-hop Reservation

• End-host: specify
– the arrival rate characterized by token-bucket with parameters (b,r,R)
– the maximum maximum admissible delay D
• Router: allocate bandwidth ra and buffer space Ba such that
– no packet is dropped
– no packet experiences a delay larger than D

slope ra

slope r
bits Arrival curve

b*R/(R-r)

D
Ba
39
 End-to-End Reservation

• When R gets PATH message it knows

– Traffic characteristics (tspec): (r,b,R)
– Number of hops
• R sends back this information + worst-case delay in RESV
• Each router along path provide a per-hop delay guarantee
and forward RESV with updated info
– In simplest case routers split the delay
num hops

S2 (b,r,R,3) R
S (b,r,R)
(b,r,R,2,D-d1)
(b,r,R,3,D)
(b,r,R,0,0) S1 (b,r,R,1,D-d1-d2) S3
worst-case delay
PATH
RESV 40
 Reservation Style

• Motivation: achieve more efficient resource

utilization in multicast (M x N)
• Observation: in a video conferencing when
there are M senders, only a few can be active
simultaneously
– Multiple senders can share the same reservation
• Various reservation styles specify different
rules for sharing among senders

41
 Reservation Styles and Filter Spec

• Reservation style
– use filter to specify which sender can use the
reservation
• Three styles
– Wildcard filter: does not specify any sender; all packets
associated to a destination shares same resources
• Group in which there are a small number of
simultaneously active senders
– Fixed filter: no sharing among senders, sender
explicitly identified for the reservation
• Sources cannot be modified over time
– Dynamic filter: resource shared by senders that are
(explicitly) specified
• Sources can be modified over time
42
 Wildcard Filter Example

• Receivers: H1, H2; senders: H3, H4, H5

• Each sender sends B
• H1 reserves B; listen from one server at a time

(B,*) H3
H2
S1 S2 S3
(B,*) (B,*)
(B,*) (B,*) (B,*)
H1 H4
H5

receiver sender

43
 Wildcard Filter Example

• H2 reserves B

(B,*) H3
H2 (B,*)
S1 S2 S3
(B,*) (B,*)
(B,*) (B,*) (B,*)
H1 H4
H5

receiver sender

44
 Wildcard Filter

• Advantages
– Minimal state at routers
• Routers need to maintain only routing state augmented
by reserved bandwidth on outgoing links
• Disadvantages
– May result in inefficient resource utilization

45
 Wildcard Filter: Inefficient Resource
Utilization Example
• H1 reserves 3B; wants to listen from all senders
simultaneously
• Problem: reserve 3B on (S3:S2) although 2B
sufficient!
H3
H2
S1 S2 S3
(3B,*) (3B,*)
(3B,*)
H1 H4
H5

receiver sender

46
 Fixed Filter Example
• Receivers: H2, H3, H4, H5; Senders: H1, H4, H5
• Routers maintain state for each receiver in the
routing table
NextHop Sources
H1 S2(H5, H4)
H2 H1(H1), S2(H5, H4) H3
H2
S1 S2 S3

H1 H4
H5

receiver sender sender+receiver

47
 Fixed Filter Example

• H2 wants to receive B only from H4

H3
H2 (B,H4)

S1 S2 S3
(B,H4) (B,H4)
(B,H4)
H1 H4
H5

receiver sender sender+receiver

48
 Dynamic Filter Example

• H5 wants to receive 2B from any source

H3
H2 (B,H4)
(B,*)
S1 S2 S3
(B,H4) (B,H4)
(B,*) (2B,*) (B,H4)
H1 H4
H5

receiver sender sender+receiver

49
 Soft State

• Per session state has a timer associated with it

– path state, reservation state
• State lost when timer expires
• Sender/Receiver periodically refreshes the state
• Claimed advantages
– no need to clean up dangling state after failure
– can tolerate lost signaling packets
• signaling message need not be reliably transmitted
– easy to adapt to route changes
• State can be explicitly deleted by a Teardown
message

50
 Tear-down Example

• H4 leaves the group

– H4 no longer sends PATH message
– State corresponding to H4 removed

H3
H2 (B,H4)
(B,*)
S1 S2 S3
(B,H4) (B,H4)
(B,*) (2B,*) (B,H4)
H1 H4
H5

receiver sender sender+receiver

51
 Tear-down Example

• H4 leaves the group

– H4 no longer sends PATH message
– State corresponding to H4 removed

H3
H2
(B,*)
S1 S2 S3

(B,*) (2B,*)
H1
H5

receiver sender sender+receiver

52
 RSVP Soft-state

• RSVP control messages need to be sent

periodically
– State will disappear if not refreshed
– Periodic state refresh every t sec (30 sec)
– If no refresh within n*t (n=3) , delete state
• RSVP messages sent as router-alert
message
– Intermediate routers intercept packets and update
state accordingly

53
 Soft State (cont)

• Per session state has a timer associated with it

– Path state, reservation state
• State lost when timer expires
• Sender/Receiver periodically refreshes the state,
resends PATH/RESV messages, resets timer
• Claimed advantages
– No need to clean up dangling state after failure
– Can tolerate lost signaling packets
• Signaling message need not be reliably transmitted
– Easy to adapt to route changes
• State can be explicitly deleted by a Teardown
message

54
 RSVP and Routing

• RSVP designed to work with variety of routing

protocols
• Minimal routing service
– RSVP asks routing how to route a PATH message
• Route pinning
– addresses QoS changes due to “avoidable” route
changes while session in progress
• QoS routing
– RSVP route selection based on QoS parameters
– granularity of reservation and routing may differ
• Explicit routing
– Use RSVP to set up routes for reserved traffic

55
 Recap of RSVP

• PATH message
– sender template and traffic spec
– advertisement
– mark route for RESV message
– follow data path
• RESV message
– reservation request, including flow and filter spec
– reservation style and merging rules
– follow reverse data path
• Other messages
– PathTear, ResvTear, PathErr, ResvErr

56
 Why did IntServ fail?

• Economic factors
– Deployment cost vs Benefit
• Is reservation, the right approach?
– Multicast centric view
• Is per-flow state maintenance an issue?
• More about QoS in general …

57
 What is the Problem?

• Goal: provide support for wide variety of

applications:
– Interactive TV, IP telephony, on-line gamming
(distributed simulations), VPNs, etc
• Problem:
– Best-effort cannot do it?
– Intserv can support all these applications, but
• Too complex
• Not scalable

58
 Differentiated Services (Diffserv)

• Build around the concept of domain

• Domain – a contiguous region of network under
the same administrative ownership
• Differentiate between edge and core routers
• Edge routers
– Perform per aggregate shaping or policing
– Mark packets with a small number of bits; each bit
encoding represents a class (subclass)
• Core routers
– Process packets based on packet marking
• Far more scalable than Intserv, but provides
weaker services
59
 Diffserv Architecture
• Ingress routers
– Police/shape traffic
– Set Differentiated Service Code Point (DSCP) in Diffserv (DS) field
• Core routers
– Implement Per Hop Behavior (PHB) for each DSCP
– Process packets based on DSCP

DS-1 DS-2

Ingress Ingress
Egress Egress

Edge router Core router

60
 Differentiated Service (DS) Field
0 5 6 7
DS Filed
0 4 8 16 19 31
Version HLen TOS Length
Identification Flags Fragment offset
TTL Protocol Header checksum IP
Source address header
Destination address
Data

• DS filed reuse the first 6 bits from the former Type of

Service (TOS) byte
• The other two bits are proposed to be used by ECN

61
 Differentiated Services

• Two types of service

– Assured service
– Premium service
• Plus, best-effort service

62
 Assured Service
[Clark & Wroclawski ‘97]

• Defined in terms of user profile, how much assured

traffic is a user allowed to inject into the network
• Network: provides a lower loss rate than best-effort
– In case of congestion best-effort packets are dropped first
• User: sends no more assured traffic than its profile
– If it sends more, the excess traffic is converted to best-
effort

63
 Assured Service

• Large spatial granularity service

• Theoretically, user profile is defined
irrespective of destination
– All other services we learnt are end-to-end,
i.e., we know destination(s) apriori
• This makes service very useful, but hard
to provision (why ?)
Traffic profile

Ingress
64
 Premium Service
[Jacobson ’97]

• Provides the abstraction of a virtual pipe

between an ingress and an egress router
• Network: guarantees that premium packets
(QoS) are not dropped and they experience
low delay
• User: does not send more than the size of
the pipe
– If it sends more, excess traffic is delayed, and
dropped when buffer overflows

65
 Edge Router

Ingress

Traffic conditioner
Class 1
Marked traffic
Traffic conditioner
Class 2
Data traffic
Classifier Scheduler
Best-effort

Per aggregate
Classification
(e.g., user)
66
 Assumptions

• Assume two bits

– P-bit denotes premium traffic
– A-bit denotes assured traffic
• Traffic conditioner (TC) implement
– Metering
– Marking
– Shaping

67
 TC Performing Metering/Marking
• Used to implement Assured Service
• In-profile traffic is marked:
– A-bit is set in every packet
• Out-of-profile (excess) traffic is unmarked
– A-bit is cleared (if it was previously set) in every packet; this
traffic treated as best-effort

r bps
User profile
b bits (token bucket)

assured traffic Set A-bit in-profile traffic

Metering
Clear A-bit out-of-profile traffic

68
 TC Performing Metering/Marking/Shaping
• Used to implement Premium Service
• In-profile traffic marked:
– Set P-bit in each packet
• Out-of-profile traffic is delayed, and when buffer overflows it
is dropped

r bps
User profile
b bits (token bucket)

premium traffic
Metering/
Shaper/ in-profile traffic
Set P-bit

out-of-profile traffic
(delayed and dropped) 69
 Scheduler
• Employed by both edge and core routers
• For premium service – use strict priority, or weighted fair queuing
(WFQ)
• For assured service – use RIO (RED with In and Out)
– Always drop OUT packets first
• For OUT measure entire queue
• For IN measure only in-profile queue
Dropping
probability
1
OUT IN

Average queue length

70
 Scheduler Example

• Premium traffic sent at high priority

• Assured and best-effort traffic pass
through RIO and then sent at low priority

yes
P-bit set? high priority

no
yes
A-bit set? no RIO low priority

71
 Control Path

• Each domain is assigned a Bandwidth Broker (BB)

– Usually, used to perform ingress-egress bandwidth
allocation
• BB is responsible to perform admission control in
the entire domain
• BB not easy to implement
– Require complete knowledge about domain
– Single point of failure, may be performance bottleneck
– Designing BB still a research problem

72
 Example

• Achieve end-to-end bandwidth guarantee

2 3

BB BB BB
7 5
1 9
8 profile 6 4 profile
profile
receiver
sender

73
 Comparison: Best-Effort, Diffserv, Intserv

Best-Effort Diffserv Intserv

Service Connectivity Per aggregate Per flow isolation

No isolation isolation Per flow
No guarantees Per aggregate guarantee
guarantee
Service End-to-end Domain End-to-end
scope
Complexity No setup Long term setup Per flow setup
Scalability Highly scalable Scalable Not scalable
(nodes (edge routers (each router
maintain only maintains per maintains per
routing state) aggregate state; core flow state)
routers per class 74
state)
 Summary

• Diffserv more scalable than Intserv

– Edge routers maintain per aggregate state
– Core routers maintain state only for a few traffic classes
• But, provides weaker services than Intserv, e.g.,
– Per aggregate bandwidth guarantees (premium service) vs.
per flow bandwidth and delay guarantees
• BB is not an entirely solved problem
– Single point of failure
– Handle only long term reservations (hours, days)

75
 Non-technical Factors Impacting QoS

• Existing Networks
– What is available today to solve our needs? Why
switch?
• Business Models
– How QoS make doing business harders
• Deployment Issues
– How QoS makes running the network harder.

76
 Existing Networks

• Motivating applications?
– Tele/Video conferencing, video distribution, VPN,
games.
• IP+QoS must be better AND cheaper than:
– PSTN with N-way calling
– Cable TV with digital recorders (Tivo)
– Telecom leased lines (ISDN, ATM, SONET)
– Peer to Peer networks

77
 Business Issues

• Service provider offers Qos(premium service)

• Must be something customer can:
– Understand
• Counterexample: Complex statistical reasoning
– Verify
• 3rd party?
• How do you know it works? Simulate a DoS attack?
– Reclaim loss if service is not delivered
• If you buy a lock and it doesn’t work, do you try to get
your $ back? What if no one tried to break in?

78
 Deployment Issues

• Today’s IP operators use simple models to

reason about what is a “good network”
• Things you worry about:
• IP packets
• BGP routing
• Simple Service Level Agreements (SLA)

79
 Deployment Issues

• QoS introduces extra effort for operators:

– shaping, policing, reservation signaling, per-reservation
billing and settlement.
• QoS deployment changes:
– Interface between an ISP and its neighbors
– adds whole new complexities for customer and support
personnel,
– creates the need for accurate service auditing,
• Increases the risk of litigation
• Tradeoff:
– Use QoS vs. make sure utilization is low most of the time?
Which is easier?

80
 Non-technical Issues summary

• Working on QoS for IP for 20 years?

– Why little/no progress?
• QoS benefits must significant to overcome all
non-technical obstacles.
– Value to users must exceed all costs
– A typical technology adoption problem?
-> Technically better isn’t always good enough
– QWERTY 10x backward compatibility rule?
– QoS not cheaper, so 1000x?

81
 Over Provisioning Vs. QoS

• QoS:
– Insure resources per flow
• Over Provisioning
– Domain of traffic engineering
– Insure sufficient resources aggregate demand
– Formalism for this approach?

82
 Over Provisioning Formalism

• Start with Traffic matrix

– Cell is the traffic from source I to dest. J
• Insure network delivers adequate
performance for all “expected” matrixes

83