Traffic Management

1
Traffic ManagementTraffic Engineering
2
An example

• Executives participating in a worldwide
  videoconference
• Proceedings are videotaped and stored in an
  archive
• Edited and placed on a Web site
• Accessed later by others
• During conference
• Sends email to an assistant
• Breaks off to answer a voice call

3
What this requires

• For video
• sustained bandwidth of at least 64 kbps
• low loss rate
• For voice
• sustained bandwidth of at least 8 kbps
• low loss rate
• For interactive communication
• low delay (lt 100 ms one-way)
• For playback
• low delay jitter
• For email and archiving
• reliable bulk transport

4
What if

• A million executives were simultaneously
  accessing the network?
• What capacity should each trunk have?
• How should packets be routed? (Can we spread load
  over alternate paths?)
• How can different traffic types get different
  services from the network?
• How should each endpoint regulate its load?
• How should we price the network?
• These types of questions lie at the heart of
  network design and operation, and form the basis
  for traffic management.

5
Traffic management

• Set of policies and mechanisms that allow a
  network to efficiently satisfy a diverse range of
  service requests
• The mechanisms and policies have to be deployed
  at both node level as well as network level
• Tension is between diversity and efficiency
• Traffic management is necessary for providing
  Quality of Service (QoS)
• Subsumes congestion control (congestion loss
  of efficiency)

6
Traffic Engineering

• Engineering of a given network so that the
  underlying network can support the services with
  requested quality
• Encompasses
• Network Design
• Capacity Design (How many nodes, where)
• Link Dimensioning (How many links, what capacity)
• Path Provisioning (How much bandwidth end-to-end)
• Multi-homing (Reliability for customer)
• Protection for Reliability (Reliability in
  Network)
• Resource Allocation
• Congestion Control
• routing around failures
• adding more capacity

7
Why is it important?

• One of the most challenging open problems in
  networking
• Commercially important
• AOL burnout
• Perceived reliability (necessary for
  infrastructure)
• Capacity sizing directly affects the bottom line
• At the heart of the next generation of data
  networks
• Traffic management Connectivity Quality of
  Service

8
Outline

• Economic principles
• Traffic classes
• Time scales
• Mechanisms
• Queueing
• Scheduling
• Congestion Control
• Admission Control
• Some open problems

9
Lets order Pizza for home delivery

• Customer
• calls a closest pizza outlet (what is selection
  based on??)
• orders a pizza
• Requirement specification
• type, toppings (measurable quantities)
• order arrives at home
• Service Quality
• How fast it arrived
• Is the right pizza? Anything missing (quality
  measurements)
• Customer Satisfaction (based on feeling!!, all
  parameters not measurable)
• How was the service?
• Is Pizza cold or hot? Is it fresh?
• Pizza company
• How many customers and how fast to serve
• Customer Satisfaction Only through complaints
  (cannot really measure)
• What they know only what customer ordered
  (Requirement!!)

10
Economics Basics utility function

• Users are assumed to have a utility function that
  maps from a given quality of service to a level
  of satisfaction, or utility
• Utility functions are private information
• Cannot compare utility functions between users
• Rational users take actions that maximize their
  utility
• Can determine utility function by observing
  preferences
• Generally networks do not support signaling of
  utility
• They only support signaling of requirements
  (bandwidth, delay)
• Networks use resource allocation to make sure
  requirements are satisfied
• Measurements and Service Level Agreements (SLAs)
  determine customer satisfaction!!

11
Example File Transfer

• Let u(t) S - ? t
• u(t) utility from file transfer
• S satisfaction when transfer infinitely fast
• t transfer time
• ? rate at which satisfaction decreases with
  time
• As transfer time increases, utility decreases
• If t gt S/ ?, user is worse off! (reflects time
  wasted)
• Assumes linear decrease in utility
• S and ? can be experimentally determined

12
Example Video Conference

• Every packet must receive before a deadline
• Otherwise, the packet is too late and cannot be
  used
• Model
• u(t) if (t lt D) then S
• else (-?)
• t is the end to end delay experienced by a
  packet
• D is the delay deadline
• S is the satisfaction
• -? is the cost (penalty) for missing deadline
• causes performance degradation
• Sophisticated Utility measures not only delay but
  packet loss too
• u(?) S(1- ?) where ? is the packet loss
  probability

13
Social welfare

• Suppose network manager knew the utility function
  of every user
• Social Welfare is maximized when some combination
  of the utility functions (such as sum) is
  maximized while minimizing the infrastructure
  cost
• An economy (network) is efficient when increasing
  the utility of one user must necessarily decrease
  the utility of another
• An economy (network) is envy-free if no user
  would trade places with another (better
  performance also costs more)
• Goal maximize social welfare
• subject to efficiency, envy-freeness, and making
  a profit

14
Example

• Assume
• Single switch, each user imposes load (?0.4)
• As utility 4 - d
• Bs utility 8 - 2d
• Same delay (d) to both users
• Conservation law ?(?idi) Constant
• 0.4d 0.4d C gt d 1.25 C gt Sum of utilities
  12-3.75 C
• If B wants lower delay say to 0.5C, then As
  delay 2C
• Sum of utilities 12 - 3C (Larger than before)
• By giving high priority to users that want lower
  delay, network can increase its utility
• Increase in social welfare need not benefit
  everyone
• A loses utility, but may pay less for service

15
Some economic principles

• A single network that provides heterogeneous QoS
  is better than separate networks for each QoS
• unused capacity is available to others
• Lowering delay of delay-sensitive traffic
  increases welfare
• can increase welfare by matching service menu to
  user requirements
• BUT need to know what users want (signaling)
• For typical utility functions, welfare increases
  more than linearly with increase in capacity
• individual users see smaller overall fluctuations
• can increase welfare by increasing capacity

16
Principles applied

• A single wire that carries both voice and data is
  more efficient than separate wires for voice and
  data
• ADSL
• IP Phone
• Moving from a 20 loaded 10 Mbps Ethernet to a
  20 loaded 100 Mbps Ethernet will still improve
  social welfare
• increase capacity whenever possible
• Better to give 5 of the traffic lower delay than
  all traffic low delay
• should somehow mark and isolate low-delay traffic

17
The two camps

• Can increase welfare either by
• matching services to user requirements or
• increasing capacity blindly
• Which is cheaper?
• no one is really sure!
• small and smart vs. big and dumb
• It seems that smarter ought to be better
• otherwise, to get low delays for some traffic, we
  need to give all traffic low delay, even if it
  doesnt need it
• But, perhaps, we can use the money spent on
  traffic management to increase capacity
• We will study traffic management, assuming that
  it matters!

18
How useful are utility functions and economic
framework?

• Do users really have such functions that can be
  expressed mathematically?
• Practically no or less clear
• Even if users cannot come up with a mathematical
  formula, they can express preference of one set
  of resources over other
• These preferences can be codified as utility
  function
• Best way to think about utility functions is that
  they may allow us to come up with a mathematical
  formulation of the traffic management problem
  that gives some insight
• Practical economic algorithms may never be
  feasible
• But policies and mechanisms based on these are
  still relevant

19
Network Types

• Single-Service Networks
• Provide services for single type of traffic
• e.g., Telephone Networks (Voice), Cable Networks
  (Video), Internet (Best effort Data)
• Multi-Service Networks
• Provide services for multiple traffic types on
  the same network
• e.g., Asynchronous Transfer Mode (CBR, VBR, ABR,
  UBR), Frame Relay, Differentiated Services
  (Diff-Serv), Integrated Services (Int-Serv), MPLS
  with Traffic Engineering
• Application types need to match the service
  provided
• Traffic models are used for the applications in
  order to match services, design, deploy the
  equipment and links.

20
Application Types

• Elastic applications (Adjust bandwidth and take
  what they get)
• Wide range of acceptable rates, although faster
  is better
• E.g., data transfers such as FTP
• Continuous media applications.
• Lower and upper limit on acceptable performance
• Sometimes called tolerant real-time since they
  can adapt to the performance of the network
• E.g., changing frame rate of video stream
• Network-aware applications
• Hard real-time applications.
• Require hard limits on performance intolerant
  real-time
• E.g., control applications

21
Traffic models

• To align services, need to have some idea of how
  applications, users or aggregates of users behave
  traffic model
• e.g. how long a user uses a modem
• e.g. average size of a file transfer
• Models change with network usage
• We can only guess about the future
• Two types of models
• measurements
• educated guesses

22
Telephone traffic models

• How are calls placed?
• call arrival model
• studies show that time between calls is drawn
  from an exponential distribution
• call arrival process is therefore Poisson
• memoryless the fact that a certain amount of
  time has passed since the last call gives no
  information of time to next call
• How long are calls held?
• usually modeled as exponential
• however, measurement studies show it to be heavy
  tailed
• means that a significant number of calls last a
  very long time
• specially after usage of modems!!

23
Traffic Engineering for Voice Networks

• For a switch with N trunks, and with large
  population of users (M??), the probability of
  blocking (i.e., a call is lost) is given by
  Erlang-B formula
• ? is the call arrival rate (calls /sec)
• 1/? is the call holding time (3 minutes)
• Example (For A 12 Erlangs)
• PB 1 for N 20 A/N 0.6
• PB 8 for N 18 A/N 0.8
• PB 30 for N 7 A/N 1.7

24
Distributions

• Long/heavy-tailed distributions
• power law
• PX gt x ? cx?a x??, a,c gt 0
• Pareto
• PX gt x c a x? a, x gt b
• Exponential Distribution
• PX gt x e-ax

25
Pareto distribution

• 1lt?lt2 gt infinite variance

Power law decays more slowly than exponential ?
heavy tail
26
Internet traffic modeling

• A few apps account for most of the traffic
• WWW
• FTP
• telnet
• A common approach is to model apps (this ignores
  distribution of destination!)
• time between app invocations
• connection duration
• bytes transferred
• packet inter-arrival distribution
• Little consensus on models
• But two important features

27
Internet traffic models features

• LAN connections differ from WAN connections
• Higher bandwidth (more bytes/call)
• longer holding times
• Many parameters are heavy-tailed
• examples
• bytes in call
• call duration
• means that a few calls are responsible for most
  of the traffic
• these calls must be well-managed
• also means that even aggregates with many calls
  not be smooth
• can have long bursts
• New models appear all the time, to account for
  rapidly changing traffic mix

28
Outline

• Economic principles
• Traffic classes
• Time scales
• Mechanisms
• Some open problems

29
Traffic classes

• Networks should match offered service to source
  requirements (corresponds to utility functions)
• Example telnet requires low bandwidth and low
  delay
• utility increases with decrease in delay
• network should provide a low-delay service
• or, telnet belongs to the low-delay traffic class
• Traffic classes encompass both user requirements
  and network service offerings
• Applications match the traffic to the service
  offering
• Request resources from the network accordingly

30
Traffic classes - details

• A basic division guaranteed service and best
  effort
• like flying with reservation or standby
• Guaranteed-service
• utility is zero unless app gets a minimum level
  of service quality
• bandwidth, delay, loss
• open-loop flow control with admission control
• e.g. telephony, remote sensing, interactive
  multiplayer games
• Best-effort
• send and pray
• closed-loop flow control
• e.g. email, net news

31
GS vs. BE (cont.)

• Degree of synchrony
• time scale at which peer endpoints interact
• GS are typically synchronous or interactive
• interact on the timescale of a round trip time
• e.g. telephone conversation or telnet
• BE are typically asynchronous or non-interactive
• interact on longer time scales
• e.g. Email
• Sensitivity to time and delay
• GS apps are real-time
• performance depends on wall clock
• BE apps are typically indifferent to real time
• automatically scale back during overload

32
Best Effort (Flow Control)

• Explicit
• Network Tells at what rate the source should send
  the packets
• Network elements may compute connection fair
  share based on Max-Min allocation (e.g, ABR in
  ATM Networks)
• Or it can be based on 1 bit congestion indicator
  (e.g., EFCI in ABR of ATM Networks)
• Implicit
• Packet drop is detected by the source and adjusts
  the window transmission (e.g., TCP)
• No flow control
• Packets are dropped by the network nodes
• Sources may not react (e.g, UDP, UBR)
• Problems are caused if these two types are mixed!!

33
Traffic subclasses (roadmap)

• ATM Forum
• based on sensitivity to bandwidth
• GS
• CBR, VBR
• BE
• ABR, UBR

• IETF
• based on ToS
• IETF based on RSVP
• based on sensitivity to delay
• GS
• intolerant
• tolerant
• BE
• interactive burst
• interactive bulk
• asynchronous bulk
• IETF based in DiffServ
• PHB
• EF, 4 AFs and BE

34
ATM Basics

• See the ATM Forum Presentation

35
ATM Basics

• Logical or Virtual Connection
• Connection is first established using signaling
  protocol
• Route from the source to the destination is
  chosen
• The same route is used for all cells (fixed size
  packets) of the connection
• No routing decision for every cell (they are
  switched in the same path)

36
Virtual Circuits in ATM

• Virtual Circuit Identifier is represented jointly
  by
• Virtual Channel Identifier (VCI)
• Virtual Path Identifier (VPI)
• Virtual Channel (VC)
• Path for cell associated with a connection
• Supports transportation of a data stream
• Each VC is assigned a unique VCI on a link

37
Virtual Channels in ATM

• Virtual Path (VP)
• Grouping of virtual channels on a physical link
• Switching can be performed on the path basis
• reduced overheads
• Each virtual path is assigned Virtual Path
  Identifier (VPI)

38
VCs In ATM
39
Virtual Path Switch (VP - Switch)
40
VP / VC Switch
41
ATM Network Example
D2
Switch 2
Access
Switch 1
Switch 3
Mux
S2
D1
S1
Core
CAC
CAC

• Each connection has its own traffic descriptors
  such as PCR, SCR, MBS, CDVT, CLR, MCR
• A Connection Admission Control algorithm (CAC)
  will check for the resources at queuing points to
  make a decision on admissibility
• Network efficiency depends upon the CAC

42
ATM Forum GS subclasses

• Constant Bit Rate (CBR)
• constant, cell-smooth traffic
• mean and peak rate are the same
• e.g. telephone call evenly sampled and
  uncompressed
• constant bandwidth, variable quality
• Variable Bit Rate (VBR)
• long term average with occasional bursts
• try to minimize delay
• can tolerate loss and higher delays than CBR
• e.g. compressed video or audio with constant
  quality, variable bandwidth

43
ATM Forum BE subclasses

• Available Bit Rate (ABR)
• users get whatever is available
• zero loss if network signals (in RM cells) are
  obeyed
• no guarantee on delay or bandwidth
• Unspecified Bit Rate (UBR)
• like ABR, but no feedback
• no guarantee on loss
• presumably cheaper
• Guaranteed Frame Rate (GFR)
• like UBR/ABR, expressed in terms of frame rate

44
ATM Attributes

• How do we describe a flow (connection) of ATM
  Service?
• Service Category
• Traffic Parameters or descriptors
• QoS parameters
• Congestion (for ABR)
• Other (for UBR)
• Cell Loss Priority (CLP0 or CLP01)
• Connections are signaled with various parameters
• A Connection Admission Control (CAC) procedure
  checks for resources in the network
• If connection is accepted, a traffic contract
  is awarded to the user (Service Level Agreement)

45
Traffic Descriptors or Parameters

• Connection Traffic Descriptor
• Source Traffic Descriptor PCR, SCR, MBS, MCR,
  MFS
• Cell Delay Variation Tolerance (?) upper bound
  on amount of cell delay that is introduced by the
  network interface and the UNI (due to
  interleaving, physical layer overhead,
  multiplexing, etc.)
• Conformance Definition unambiguous specification
  of conforming cells of a connection at the UNI (
  a policing function is used to check for
  conformance such as Generic Cell Rate Algorithm
  (GCRA))

46
Traffic Parameters (Source Traffic Descriptor)

• Peak Cell Rate (PCR) upper bound on traffic
  submitted by source (PCR 1/T, where T minimum
  cell spacing
• Sustainable Cell Rate (SCR) upper bound on
  average rate of traffic submitted by source
  (over a larger T)
• Maximum Burst Size (MBS) maximum number of cells
  sent continuously at PCR
• Minimum Cell Rate (MCR) used with ABR and GFR,
  minimum cell rate requested, access to unused
  capacity up to PCR (elastic capacity PCR-MCR)
• Maximum Frame Size (MFS) maximum size of a frame
  in cells available for GFR service

47
Cell Rates

• Peak Cell Rate (PCR), Line Cell Rate (LCR)

T1/PCR
t1/LCR

• Sustained Cell Rate (SCR) PCR(Ton/TonToff)

Ton
Toff
48
Quality of Service

• Cell Transfer Delay (CTD)
• Cell Delay Variation (CDV)

time
Cell arrival pattern
Queuing point (e.g. mux, switch)
Cell departure pattern without CDV
Cell departure pattern with CDV
Switch transit delay
Positive CDV
Negative CDV
49
Cell Transfer Delay Probability Density
Variable component of delay, due to buffering
and cell scheduling.
50
QoS Parameters

• Peak-to-peak cell delay variation (CDV)
  acceptable delay variation at destination. The
  peak-to-peak CDV is the (1 - ?) quantile of the
  CTD minus the fixed CTD that could be experienced
  by any delivered cell on a connection during the
  entire connection holding time.
• Maximum Cell Transfer Delay (maxCTD) maximum
  time between transmission of first bit of a cell
  at the source UNI to receipt of its last bit at
  the destination UNI
• Cell Loss Ratio ratio of lost cells to total
  transmitted cells on a connection Lost
  Cells/Total Transmitted Cells

51
Other Attributes

• Congestion Control
• defined only for ABR service category
• uses network feedback controls
• ABR flow control mechanism (more later)
• Other Attributes (introduced July 2000)
• Behavior class selector (BCS)
• for IP differentiated services (DiffServ)
• provides for different levels of service among
  UBR connections
• implementation dependent, no guidance in specs
• Minimum desired cell rate (MDCR)
• UBR application minimum capacity objective

52
Attributes of Each Service Category
53
Service Paradigm

• Quantitative Commitments
• Sets explicit values
• Ensures service quality through resource
  allocation and traffic policing
• Qualitative Commitments
• Relative measure and no explicit guarantees
• Some unspecified level of quality through
  network engineering

54
Quantitative Commitments

• Generally connection oriented transport
• Network nodes maintain per-flow state info
• QoS (or GOS) requirements of each connection is
  explicitly specified and signaled
• Network enforces traffic regulation (policing,
  shaping) if necessary and allocates resources for
  each connection
• Examples Voice networks (POTS), ATM, FR
• Expensive and under-utilized

55
Qualitative Commitments

• Generally connection less transport
• no per-flow state info is maintained due to flow
  aggregation
• QoS requirements are not explicitly specified
• Network may not enforce traffic regulation
• May allocate resources for logical groups (such
  as VPN)
• Examples IP, LANs
• Cheap and over-utilized

56
QoS Building Blocks

• Backbone supporting QoS speed and scale
• Packet / Service classification (sorting)
• Bandwidth management and admission control
• Queue management
• Congestion management
• Granular measurements

57
Functions Needed

• Admission control - some way to limit usage
  relative to resources.
• Packet scheduling - some way to treat different
  packets differently.
• Classifier mechanism - some way to sort packets
  into different treatment groups.
• Policies and rules for allocating resources.

58
IETF

• Internet currently provides only single class of
  best-effort service.
• No admission control and no assurances about
  delivery
• Existing applications are elastic.
• Tolerate delays and losses
• Can adapt to congestion
• Future real-time applications may be inelastic.
• Should we modify these applications to be more
  adaptive or should we modify the Internet to
  support inelastic behavior?

59
IETF ToS (1-byte Type-of-Service)

• Bits 0-2 Precedence.
• Bit 3 0 Normal Delay, 1 Low Delay.
• Bits 4 0 Normal Throughput, 1 High
  Throughput.
• Bits 5 0 Normal Relibility, 1 High
  Relibility.
• Bit 6-7 Reserved for Future Use

60
IETF int-serv (Integrated Services)

• Focus on per-flow QoS.
• Support specific applications such as video
  streaming.
• Based on mathematical guarantees.
• Many concerns
• Complexity
• Scalability
• Business model
• Charging
• Uses RSVP (Resource-Reservation Protocol)
• To signal QoS requirements

61
IETF int-serv (Integrated Services)

• Guaranteed service
• Targets hard real-time applications.
• User specifies traffic characteristics and a
  service requirement.
• Requires admission control at each of the
  routers.
• Can mathematically guarantee bandwidth, delay,
  and jitter.
• Controlled load.
• Targets applications that can adapt to network
  conditions within a certain performance window.
• User specifies traffic characteristics and
  bandwidth.
• Requires admission control at each of the
  routers.
• Guarantee not as strong as with the guaranteed
  service.
• e.g., measurement-based admission control.
• Best effort

62
RSVP

1. Sender sends PATH message to network
2. PATH leads data through the network
3. Routers install per-flow state
4. Receiver responds with RESV
5. RESV follows PATH trail back towards sender
6. Routers accept resource request (commit resources
   to flow) or reject resource request
7. Data is handled in network elements

63
IETF GS subclasses

• Tolerant GS
• nominal mean delay, but can tolerate occasional
  variation
• not specified what this means exactly
• uses controlled-load service
• even at high loads, admission control assures a
  source that its service does not suffer
• it really is this imprecise!
• Intolerant GS
• need a worst case delay bound
• equivalent to CBRVBR in ATM Forum model

64
IETF BE subclasses

• Interactive burst
• bounded asynchronous service, where bound is
  qualitative, but pretty tight
• e.g. paging, messaging, email
• Interactive bulk
• bulk, but a human is waiting for the result
• e.g. FTP
• Asynchronous bulk
• junk traffic
• e.g netnews

65
IETF Diff-Serv (Differentiated Services)

• Intended to address the following difficulties
  with Intserv and RSVP
• Scalability maintaining states by routers in
  high speed networks is difficult due to the very
  large number of flows
• Flexible Service Models Intserv has only two
  classes, want to provide more qualitative service
  classes want to provide relative service
  distinction (Platinum, Gold, Silver, )
• Simpler signaling (than RSVP) many applications
  and users may only want to specify a more
  qualitative notion of service

66
Diffserv PHB (Per-Hop-Behavior)

• Packet is marked in the Type of Service (TOS) in
  IPv4, and Traffic Class in IPv6.
• 6 bits used for Differentiated Service Code Point
  (DSCP) and determine PHB that the packet will
  receive.
• EF, 4 classes of AF, each with 3 drop priorities
  (AF11, AF12, AF13, AF21, AF22, AF23, AF31, AF32,
  AF33, AF41, AF42, AF43)and Best-Effort (BE)
• 2 bits are currently unused.

67
PHB Class Selector

• Derived from IP Precedence values
• 6 bit diff-serv code point (DSCP) determines
  per-hop behavior of packet treatment
• Expedited Forwarding (EF) low loss and latency
• Assured Forwarding (AF) 4 classes, 3 drop
  precedence
• Best Effort (BE) classical IP
• No absolute guarantees

68
DiffServ

1. Routers configured for certain PHBs (Per Hop
   Behavior)
2. Resources are allocated to PHBs
3. Edge routers are configured to mark DSCP
   (requests PHB) based on classification
   information
4. Traffic arriving at edge router marked with DSCP
5. Traffic in core routers go to PHB requested by
   DSCP

DSCP marked at edge
SLA defines capacity at each service level (DSCP)
69
Diff-Serv Network Architecture
Backbone
Scalable Solutions Require Cooperative Edge
and Backbone Functions

• Edge Functions
• Packet classification
• Bandwidth management
• L3 metering
• Security filtering
• Access aggregation

• Backbone Functions
• High-speed switching and transport
• QoS enforcement
• QoS interworking

70
Packet Classification

• Up to six traffic classes via ToS precedence bits
• Classification by physical port, IP address,
  application, IP protocol, etc.
• Network or external assignment

71
Multi-field Packet Classification
L3-DA
L3-SA
L4-PROT
Field 1 Field 2 Field k Action
Rule 1 5.3.40.0/21 2.13.8.11/32 UDP A1
Rule 2 5.168.3.0/24 152.133.0.0/16 TCP A2

Rule N 5.168.0.0/16 152.0.0.0/8 ANY AN
Packet Classification Find the action associated
with the highest priority rule matching an
incoming packet header.
Courtesy Nick McKeown_at_Stanford
72
Formal Problem Definition

• Given a classifier C with N rules, Rj, 1 ? j ? N,
  where Rj consists of three entities
• A regular expression Rji, 1 ? i ? d, on each of
  the d header fields,
• A number, pri(Rj), indicating the priority of the
  rule in the classifier, and
• An action, referred to as action(Rj).

For an incoming packet P with the header
considered as a d-tuple of points (P1, P2, ,
Pd), the d-dimensional packet classification
problem is to find the rule Rm with the highest
priority among all the rules Rj matching the
d-tuple i.e., pri(Rm) gt pri(Rj), ? j ? m, 1 ? j
? N, such that Pi matches Rji, 1 ? i ? d. We
call rule Rm the best matching rule for packet P.
Courtesy Nick McKeown_at_Stanford
73
Routing Lookup Instance of 1D Classification

• One-dimension (destination address)
• Forwarding table ? classifier
• Routing table entry ? rule
• Outgoing interface ? action
• Prefix-length ? priority

Courtesy Nick McKeown_at_Stanford
74
Example 4D Classifier
Rule L3-DA L3-SA L4-DP L4-PROT Action
R1 152.163.190.69/255.255.255.255 152.163.80.11/255.255.255.255 Deny
R2 152.168.3/255.255.255 152.163.200.157/255.255.255.255 eq www udp Deny
R3 152.168.3/255.255.255 152.163.200.157/255.255.255.255 range 20-21 udp Permit
R4 152.168.3/255.255.255 152.163.200.157/255.255.255.255 eq www tcp Deny
R5 Deny
Courtesy Nick McKeown_at_Stanford
75
Example Classification Results
Pkt Hdr L3-DA L3-SA L4-DP L4-PROT Rule, Action
P1 152.163.190.69 152.163.80.11 www tcp R1, Deny
P2 152.168.3.21 152.163.200.157 www udp R2, Deny
Courtesy Nick McKeown_at_Stanford
76
Classification algorithms

• Types
• Linear search
• Associative search
• Trie-based techniques
• Crossproducting
• Heuristic algorithms
• Algorithms So far
• Good for two fields, but do not scale to more
  than two fields, OR
• Good for very small classifiers (lt 50 rules)
  only, OR
• Have non-deterministic classification time, OR
• Either too slow or consume too much storage

Another Project Item
77
DiffServ Routers

DiffServ Edge Router
Classifier
Meter
Policer
Marker

DiffServ Core Router
PHB
PHB
Select PHB
Local conditions
PHB
PHB
Extract DSCP
Packet treatment
78
Edge Router/Host Functions

• Classification marks packets according to
  classification rules to be specified.
• Metering checks whether the traffic falls within
  the negotiated profile.
• Marking marks traffic that falls within profile.
• Conditioning delays and then forwards, discards,
  or remarks other traffic.

79
Core Functions

• Forwarding according to Per-Hop-Behavior or
  PHB specified for the particular packet class
  such PHB is strictly based on class marking (no
  other header fields can be used to influence
  PHB).
• BIG ADVANTAGE
• No state info to be maintained by routers!

80
Forwarding (PHB)

• PHB results in a different observable
  (measurable) forwarding performance behavior.
• PHB does not specify what mechanisms to use to
  ensure required PHB performance behavior.
• Examples
• Class A gets x of outgoing link bandwidth over
  time intervals of a specified length.
• Class A packets leave first before packets from
  class B.

81
Forwarding (PHB)

• Expedited Forwarding (EF)
• Guarantees a certain minimum rate for the EF
  traffic.
• Implies isolation guarantee for the EF traffic
  should not be influenced by the other traffic
  classes.
• Admitted based on peak rate.
• Non-conformant traffic is dropped or shaped.
• Possible service providing a virtual wire.

82
Forwarding (PHB)

• Assured Forwarding (AF)
• AF defines 4 classes with some bandwidth and
  buffers allocated to them.
• The intent is that it will be used to implement
  services that differ relative to each other
  (e.g., gold, silver,).
• Within each class, there are three drop
  priorities, which affect which packets will get
  dropped first if there is congestion.
• Lots of studies on how these classes and drop
  priorities interact with TCP flow control.
• Non-conformant traffic is remarked.

83
Example of EF A Virtual Leased Line Service

• Service offers users a dedicated traffic pipe.
• Guaranteed bandwidth between two points.
• Very low latency and jitter since there should be
  no queuing delay (peak rate allocation).
• Admission control makes sure that all links in
  the network core have sufficient EF bandwidth.
• Simple case sum of all virtual link bandwidth is
  less than the capacity of the slowest link.
• Traffic enforcement for EF traffic limits how
  much EF traffic enters the network.

84
Differentiated Services Issues

• The key to making Diffserv work is bandwidth
  management in the network core.
• Simple for simple services such as the virtual
  pipe, but it is much more challenging for complex
  service level agreements.
• Notion of a bandwidth broker that manages the
  core network bandwidth.
• Definition of end-to-end services for paths that
  cross networks with different forwarding
  behaviors
• Some packets will be handled differently in
  different routers.
• Some routers are not DiffServ capable.
• Per-Domain Behavior (PDB)

85
Some points to ponder

• The only thing out there is CBR and asynchronous
  bulk!
• There are application requirements. There are
  also organizational requirements (link sharing)
• Users needs QoS for other things too!
• billing
• privacy and security
• reliability and availability

86
Outline

• Economic principles
• Traffic classes
• Time scales
• Mechanisms
• Some open problems

87
Time scales

• Some actions are taken once per call
• tell network about traffic characterization and
  request resources
• in ATM networks, finding a path from source to
  destination
• Other actions are taken during the call, every
  few round trip times
• feedback flow control
• Still others are taken very rapidly,during the
  data transfer
• scheduling
• policing and regulation
• Traffic management mechanisms must deal with a
  range of traffic classes at a range of time scales

88
Summary of mechanisms at each time scale

• Less than one round-trip-time (cell or packet
  level)
• Scheduling and buffer management
• Regulation and policing
• Policy routing (datagram networks)
• One or more round-trip-times (burst-level)
• Feedback flow control
• Retransmission
• Renegotiation

89
Summary (cont.)

• Session (call-level)
• Signaling
• Admission control
• Service pricing
• Routing (connection-oriented networks)
• Day
• Peak load pricing
• Weeks or months
• Capacity planning

90
Outline

• Economic principles
• Traffic classes
• Mechanisms at each time scale
• Faster than one RTT
• scheduling and buffer management
• regulation and policing
• policy routing
• One RTT
• Session
• Day
• Weeks to months
• Some open problems

91
Faster than RTT

• Scheduling and buffer management
• Policing and Regulation
• In separate set of slides

92
Renegotiation
93
Renegotiation

• An option for guaranteed-service traffic
• Static descriptors dont make sense for many real
  traffic sources
• interactive video
• Multiple-time-scale traffic
• burst size B that lasts for time T
• for zero loss, descriptors (P,0), (A, B)
• P peak rate, A average B Burst Size
• T large gt serving even slightly below P leads to
  large buffering requirements
• one-shot descriptor is inadequate

94
Renegotiation (cont.)

• Renegotiation matches service rate to traffic
• Renegotiating service rate about once every ten
  seconds is sufficient to reduce bandwidth
  requirement nearly to average rate
• works well in conjunction with optimal smoothing
• Fast buffer reservation is similar
• each burst of data preceded by a reservation
• Renegotiation is not free
• signaling overhead
• call admission ?
• perhaps measurement-based admission control

95
RCBR

• Extreme viewpoint
• All traffic sent as CBR
• Renegotiate CBR rate if necessary
• No need for complicated scheduling!
• Buffers at edge of network
• much cheaper
• Easy to price
• Open questions
• when to renegotiate?
• how much to ask for?
• admission control
• what to do on renegotiation failure

96
Outline

• Economic principles
• Traffic classes
• Mechanisms at each time scale
• Faster than one RTT
• One RTT
• Session
• Signaling
• Admission control
• Day
• Weeks to months
• Some open problems

97
Signaling
98
Signaling

• How a source tells the network its utility
  function or resource requirements
• Two parts
• how to carry the message (transport)
• how to interpret it (semantics)
• Useful to separate these mechanisms

99
Signaling semantics

• Classic scheme sender initiated
• SETUP, SETUP_ACK, SETUP_RESPONSE
• Admission control
• Tentative resource reservation and confirmation
• Simplex and duplex setup
• Doesnt work for multicast

100
Resource translation

• Application asks for end-to-end quality
• How to translate to per-hop requirements?
• E.g. end-to-delay bound of 100 ms
• What should be bound at each hop?

101
Signaling transport

• Telephone network uses Signaling System 7 (SS7)
• Carried on Common Channel Interoffice Signaling
  (CCIS) network
• CCIS is a datagram network
• SS7 protocol stack is loosely modeled on ISO (but
  predates it)
• Signaling in ATM networks uses Q.2931 standard
• part of User Network Interface (UNI)
• complex
• layered over Service Specific Connection Oriented
  Protocol SSCOP (a reliable transport protocol)
  and AAL5

102
Internet signaling transport RSVP

• Main motivation is to efficiently support
  multipoint multicast with resource reservations
• In unicast, a source communicates with only one
  destination
• In multicast, a source communicates with more
  than one destination
• Signalling Progression
• Unicast
• Naive multicast
• Intelligent multicast
• Naive multipoint multicast
• RSVP

103
RSVP motivation
104
Multicast reservation styles

• Naive multicast (source initiated)
• source contacts each receiver in turn
• wasted signaling messages
• Intelligent multicast (merge replies)
• two messages per link of spanning tree
• source needs to know all receivers
• and the rate they can absorb
• doesnt scale
• Naive multipoint multicast
• two messages per source per link
• cant share resources among multicast groups

105
RSVP

• Receiver initiated
• Reservation state per group, instead of per
  connection
• PATH and RESV messages
• PATH sets up next hop towards source(s)
• RESV makes reservation
• Travel as far back up as necessary
• how does receiver know of success?

106
Reservation Styles

• How resource reservations are aggregated/merged
  for multiple receivers in the same multicast
  group
• Two options, specified in the receivers
  reservation requests
• Reservation attribute reservation is shared over
  flows from multiple senders, or distinct for each
  sender
• Sender selection explicit list or wildcard
• Three reservation styles are defined

107
Filters

• Allow receivers to separate reservations
• Fixed filter
• receive from exactly one source
• Dynamic filter
• dynamically choose which source is allowed to use
  reservation

• Fixed-Filter
• Specifies a distinct reservation for each sender
  and an explicit list of senders
• Symbolic representation FF(S1Q1, S2Q2, )

• Shared-Explicit
• Specifies that a single resource reservation is
  to be shared by an explicit list of senders
• Symbolic representation SE(S1, S2, Q)

• Wildcard-Filter
• Specifies that a single resource reservation is
  to be shared by all senders to this address
• Symbolic representation WF(Q)

108
Soft state

• State in switch controllers (routers) is
  periodically refreshed
• On a link failure, automatically find another
  route
• Transient!
• But, probably better than with ATM

109
Why is signaling hard ?

• Complex services
• Feature interaction
• call screening call forwarding
• Tradeoff between performance and reliability
• Extensibility and maintainability

110
Outline

• Economic principles
• Traffic classes
• Mechanisms at each time scale
• Faster than one RTT
• One RTT
• Session
• Signaling
• Admission control
• Day
• Weeks to months
• Some open problems

111
Admission control
112
Connection Admission Control (CAC)

• Can a call be admitted?
• ? (bandwidth allocated for all connections) ?
  Link Rate
• Otherwise call is inadmissible
• What bandwidth to allocate to connections??
• Depends upon the traffic, traffic model assumed
  and the Queueing methodology deployed and model
  used to estimate the required bandwidth
• Procedure
• Map the traffic descriptors associated with a
  connection onto a traffic model
• Use this traffic model with an appropriate
  queuing model for each congestion point, to
  estimate whether there are enough system
  resources to admit the connection in order to
  guarantee the QoS at every congestion (or
  queuing) point.
• Allocate resources if the connection is accepted.

113
CAC (continued ..)

• Depending on the traffic models used, the CAC
  procedures can be too conservative by over
  allocating the resources.
• This reduces the statistical gains
• An efficient CAC is the one which produces
  maximum amount of statistical gain at a given
  congestion point without violating the QoS.
• The efficiency of the CAC thus depends on how
  closely the two steps (traffic model and queuing
  model) above model reality.
• Both the traffic and queuing models are well
  researched and widely published in the
  literature.

114
CBR and UBR Admission Control

• CBR admission control (Peak Rate Allocation)
• simple
• on failure try again, reroute, or hold
• Best-effort admission control
• trivial
• if minimum bandwidth needed, use CBR test

115
CAC for CBR (with small jitter)

• Given the buffer size B, the link capacity C and
  the peak cell rate of the connection PCRi,
  determine a load ? such that the probability of
  queue length exceeding B is less than ?, where ?
  is a small number such as 10-10
• Using M/D/1 model
• Using nD/D/1 model

116
Cell Loss Probability versus Buffer Size

• ?0.9
• M/D/1 is conservative
• For large N, both give similar performance

117
VBR admission control

• VBR
• peak rate differs from average rate burstiness
• if we reserve bandwidth at the peak rate, wastes
  bandwidth
• if we reserve at the average rate, may drop
  packets during peak
• key decision how much to overbook
• Four known approaches
• peak rate admission control
• worst-case admission control
• admission control with statistical guarantees
• measurement-based admission control

118
1. Peak-rate admission control

• Reserve at a connections peak rate
• Pros
• simple (can use FIFO scheduling)
• connections get negligible delay and loss
• works well for a small number of sources
• Cons
• wastes bandwidth
• peak rate may increase because of scheduling
  jitter

119
2. Worst-case admission control

• Characterize source by average rate and burst
  size (LBAP)
• Use WFQ or rate-controlled discipline to reserve
  bandwidth at average rate
• Pros
• may use less bandwidth than with peak rate
• can get an end-to-end delay guarantee
• Cons
• for low delay bound, need to reserve at more than
  peak rate!
• implementation complexity

rate
time
120
3. Admission with statistical guarantees

• Key insight is that as number of calls increases,
  probability that multiple sources send a burst
  decreases
• sum of connection rates is increasingly smooth
• With enough sources, traffic from each source can
  be assumed to arrive at its average rate
• Put in enough buffers to make probability of loss
  low
• Theory of large deviations quantitatively bounds
  the overflow probability
• By allowing a small loss, we can reduce the
  resources considerably

121
Example

• Consider an ensemble of 10 identical and
  independent sources, each of which is on with a
  probability 0.1. When on has a transmission
  rate of 1.0. What is the probability that they
  overflow a shared link of capacity 8?
• The probability that n sources are on out of 10
  is given by

The probability of loss is less than 10-6 For
peak allocation we need a capacity of 10 By
allowing loss, we reduced resources by 20!!
122
3. Admission with statistical guarantees (contd.)

• Assume that traffic from a source is sent to a
  buffer of size B which is drained at a constant
  rate R
• If source sends a burst, its delay goes up
• If the burst is too large, bits are lost
• Equivalent bandwidth (EBW) of the source is the
  rate at which we need to drain this buffer so
  that the probability of loss is less than L (and
  the delay in leaving the buffer is less than d)
• If many sources share a buffer, the equivalent
  bandwidth of each source decreases (why?)
• Equivalent bandwidth of an ensemble of
  connections is the sum of their equivalent
  bandwidths

123
3. Admission with statistical guarantees (contd.)

• When a source arrives, use its performance
  requirements and current network state to assign
  it an equivalent bandwidth
• Admission control sum of equivalent bandwidths
  at the link should be less than link capacity
• Pros
• can trade off a small loss probability for a
  large decrease in bandwidth reservation
• mathematical treatment possible
• can obtain delay bounds
• Cons
• assumes uncorrelated sources
• hairy mathematics

124
Effective Bandwidth

• This model maps each connections traffic
  parameters into a real number EBWi, called the
  Equivalent Bandwidth or Effective Bandwidth of
  the connection such that the QoS constraints are
  satisfied.
• Thus, the effective bandwidth is derived as a
  source property and with this mapping, the CAC
  rule becomes very simple
• For a connection with an average rate SCRi and
  peak rate as PCRi, the effective bandwidth is a
  number between the SCRi and PCRi. That is,
• There are many methods and models published in
  the literature

125
Properties of EBW

• Additive Property Effective bandwidths are
  additive, i.e., the total effective bandwidth
  needed for N connections equals to the sum of
  effective bandwidth of each connection
• Independence Property Effective bandwidth for a
  given connection is only a function of that
  connections parameters.
• due to the independence property, the effective
  bandwidth method could be far more conservative
  than a method which considers the true
  statistical multiplexing (i.e., the method which
  considers the presence of other connections)
• With the effective bandwidths method, the CAC
  function can add (or subtract) the effective
  bandwidth of the connection which is being set-up
  (or torn down) from the total effective
  bandwidth. This is not easily possible with any
  method which does not have the independence
  property

126
EBW (First Approach by Roberts)

• Assumes fluid sources and zero buffering (so that
  two simultaneously active sources would cause
  data loss)
• Let each source has a peak rate P, mean rate m
  and link capacity is C and required cell loss is
  smaller than 10-9
• The heuristic to estimate the EBW of a source is
• EBW 1.2m 60m(P-m) / C
• First term says EBW is 1.2 times of mean rate
• Second term increases EBW in proportion to the
  gap between peak and mean (an indicator of source
  burstiness). This is mitigated by the large link
  capacity.
• Expression is independent of cell loss!!

127
EBW (Second approach by Gibbens and Hunt)

• on-off sources with exponentially distributed
  on and off periods
• Let a source mean on period be and mean
  off period be . When the source is on,
  it is assumed to produce information at a
  constant rate
• Let B be the buffer size CLR is the cell loss
  ratio required and
• The Effective Bandwidth is given by

128
Example

• Let traffic descriptors are SCR, PCR100Mb/s,
  CLR10-7 and ABS (Average Burst Size)50 cells

129
EBW Observations

• Equation implies that for large B, ??0 and EBW
  (ci ) equals to the mean rate of the source
• For a small buffer B, ??-? and the effective
  bandwidth of the source will be , the peak
  information rate
• The queue length distribution is assumed to be
  asymptotically exponential of form

130
EBW for Self-similar traffic (By Norros)

• Let m is the mean bit rate of the traffic stream,
  a is the coefficient of variation, B is the
  buffer size, H is the Hurst parameter of the
  stream (0.5?H?1), CLR is the target cell loss
  ratio.
• The EBW is given by
• Note that this equation does not follow the
  asymptotic exponential queue length distribution

131
Multi-class CAC

• In the real world, the traffic flow consists of
  multiple QoS classes, where, the services may be
  partitioned and queued separately
• To guarantee QoS, a certain amount of bandwidth
  (or capacity) is reserved for each of the service
  categories.
• With effective bandwidth approach, this
  assignment becomes very simple.
• Let Nj be the number of sources for class j and
  let ?j be the effective bandwidth of a source
  belonging to class j. Let there be K such
  classes. Then, the CAC for multi-class traffic
  should check that the total estimated capacity is
  less than the service rate. That is,

132
4. Measurement-based admission

• For traffic that cannot describe itself
• also renegotiated traffic
• Measure real average load due to ensemble of
  connections
• Users tell peak
• If peak measured average load lt capacity, admit
• Over time, new call becomes part of average
• Problems
• assumes that past behavior is indicative of the
  future
• how long to measure?
• when to forget about the past?

133
Outline

• Economic principles