Transcript ECE544Lec8DR-QoS_2016x

ECE544: Communication Networks-II,
Spring 2016
D. Raychaudhuri
Lecture 8 (QoS)
Includes teaching materials from D. Raychaudhuri, L. Peterson
Today’s Lecture
• Congestion control in best effort
networks
– Basic principles & mechanisms
– FQ, WFQ, congestion feedback, TCP, RED
• Quality-of-service (QoS)
– Mechanisms (traffic shaping, admission
control, reservation, priority queuing)
– RSVP Intserv and Diffserv, RIO
– Comparison to ATM (CBR, VBR; ABR)
Congestion Control & QoS in
Packet Networks
• Congestion control – reactive methods used
in best effort networks
– Packet scheduling at network nodes
– Feedback congestion control
• End-to-end
• Hop-by-hop
• QoS control – proactive methods used for
premium or guaranteed services:
– Source traffic shaping & policing at entry points
– Priority queuing and packet drop at routers
– End-to-end reservation and admission control
Network Congestion
• All networks have saturating throughput
– Reduction in performance beyond max capacity
– Need to keep input load below G0
– Also must avoid unstable equilibrium point in overload region
Smax
Thru
Capacity Limit
Traffic
margin
Unstable network load
Overload
region
Congestion control policies
Normal operating
Point (G0)
Offered Traffic (G)
Stable network load lines
with congestion control
Queue Scheduling
• A queue scheduler employs 2
strategies:
– Scheduling discipline: Which packet to
serve (transmit) next
– Drop policy: Which packet to drop next
(when required)
FIFO Queuing
• FIFO:first-in-first-out (or FCFS: firstcome-first-serve)
• Arriving packets get dropped when
queue is full regardless of flow or
importance - implies drop-tail
• Important distinction:
– FIFO: scheduling discipline
– Drop-tail: drop policy
Fair Queuing
• Main idea:
– maintain a separate queue for each flow
currently flowing through router
– router services queues in Round-Robin
fashion
FQ illustration
Flow 1
Flow 2
I/P
O/P
Flow n
Variation: Weighted Fair Queuing (WFQ)
Some Complications
• Packets are of different length
• We really need bit-by-bit round-robin
(RR)
• FQ simulates bit-by-bit RR
– Not feasible to interleave bits!
Bit-by-bit RR
• Single flow: suppose clock ticks when a bit is transmitted.
For packet i:
– Pi: length, Ai = arrival time, Si: begin transmit time, Fi: finish
transmit time. Fi = Si+Pi
– Fi = max (Fi-1, Ai) + Pi
• Multiple flows: clock ticks when a bit from all active flows
is transmitted, that is, the clock advances by one tick
when n bits are transmitted (assuming n flows)
– calculate Fi for each packet
– transmit packet with earliest Fi
• RR is only simulated, packet in transmission is not
interrupted.
• If n flows have data to transmit, each gets 1/nth
bandwidth.
Bit-by-bit RR
Start with A(*,*)=0 (all pkts arrive at T=0)
Flow 1
Pkt 1-3=
1 unit
Pkt 1-2=
1 unit
Flow 2
P(1,1) = 2
P(1,2) = 1
P(1,3) = 1
F(1,1) = 1
F(1,2) = 1.5
F(1,3) = 2
Pkt 2-2=2 units
Pkt 1-1=2 units
1-1
F(2,1) = 1.5
F(2,2) = 2.5
Fi = max (Fi-1, Ai) + Pi
Pkt 2-1=3 units
Channel clock -
P(2,1) = 3
P(2,2) = 2
2-1
1-2 1-3 2-2
Weighted Fair Queuing (WFQ)
• Weighted Fair Queuing (WFQ): assign a weight to
each flow
– Assume transmitting wq bits each time the router
serves queue q (simulate in packet scheduling)
– Control the percentage of the link’s bandwidth
that a flow will get
• The bandwidth that flow q gets (n active queues sending
data):
Bq 
wq
n
w
q 1
q
– FQ is a special case of WFQ with a weight of 1 for
each queue
Congestion Control and Congestion Avoidance
•
TCP’s “blind” approach:
– Detect congestion (loss) after it happens and back off on offered
rate
– Increase load trying to maximize utilization until loss occurs
Source
Rate
(bps)
•
Congestion detected
(via packet loss)
Time-out
Alternatively:
– We can try to predict congestion and reduce rate before packets
start being discarded
– This is called congestion avoidance
Congestion Control via
Router Feedback
• Router has unified view of queuing
behavior
• Routers can distinguish between
propagation and persistent queuing
delays
• Routers can decide on transient
congestion, based on workload
Solving the Full Queues
Problem
• Router monitors the load
• Drop (or mark) packets before queue
becomes full (early drop)
• Intuition: notify senders of incipient
congestion
– Simple example:
• If qlen > drop level, drop (or mark) each new
packet with a fixed probability p
• Does not control misbehaving users
Random Early Detection
(RED)
• Motivation:
– High bw-delay flows have large queues to
accommodate transient congestion
– TCP detects congestion from loss - after
queues have built up and increase delay
• Aim:
– Keep throughput high and delay low
– Accommodate bursts
Random Early Detection
(RED)
• Detect incipient congestion, allow bursts
• Keep power (throughput/delay) high
– keep average queue size low
– assume hosts respond to lost packets
• Avoid window synchronization
– randomly mark packets, instead of dropping
• Avoid bias against bursty traffic
• Some protection against ill-behaved users
RED Algorithm
• Maintain running average of queue length
• If avg < minth do nothing
– Low queuing, send packets through
• If avg > maxth, drop all packets
– Protection from misbehaving sources
• Else randomly drop (or mark) some packets
in a manner proportional to queue length
– Notify sources of incipient congestion
RED Operation
• If AvgLen <= MinThreshold
– Queue the packet
Max
threshold
Min
threshold
• If MinThreshold < Avglen < MaxThreshold
– Calculate probability P
– Drop the arriving packets with probability P
• If Avglen >= MaxThreshold
– Drop the arriving packet
• AveLen = (1-W) x AveLen + W x SampleLen
Average queue
length
( AvgLen  MinThresho ld )
( MaxThreshold  MinThresho ld )
P  TemP /(1  count  TempP)
TempP  MaxP 
P(drop)
1.0
• Count:# of newly arriving packets that have
been queued (not dropped) while AvgLen has
Avg length been between the two thresholds
MaxP
minthresh
maxthresh
– Count increases, P increases
– Make drop more evenly distributed (Avoid bias
against bursty traffic)
Explicit Congestion Control
0
4
Version
8
HLen
16
TOS
31
Length
Ident
TTL
19
Flags
Protocol
Offset
Checksum
SourceAddr
DestinationAddr
Options (variable)
Pad
(variable)
Data
•
ToS field => now used for
DiffServ and ECN
– Bits 0-5: Differentiated Services
Code Point (DSCP)
– Bit 6: ECN-capable
– Bit 7: ECN
• Router can signal the
congestion by marking
packets instead of
dropping using RED
– Set the ECN bit (bit 7 of the
IP TOS field) in IP header
• The destination copies the
ECN bit into the ACK sent
back to the source
• The source TCP responds
to the ECN bit set in the
same way as a packet
drop
Quality of Service
Outline
Realtime Applications
Integrated Services
Differentiated Services
Realtime Applications
Microphone
Sampler,
A D
converter
Buffer,
D A
Speaker
• Require “deliver on time” assurances
– must come from inside the network
– Example application (audio)
– sample voice once every 125us
– each sample has a playback time
– packets experience variable delay in network
– add constant factor to playback time: playback point
– Use initial buffering delay to compensate jitter, but result in
longer end-to-end delay
Playback Buffer
Sequence number
Packet
arrival
Packet
generation
Playback
Network
delay
Time
Buffer
Example Distribution of
Delays
90% 97% 98%
Packets (%)
3
99%
2
1
50
100
Delay (milliseconds)
150
200
Application requirements & Services Classes
• Different application requirements
– Elastic: no restrict delay requirements, traditional data
– Real-time: delay bound, jitter, loss
• Loss: intolerant or tolerant to some loss
• Delay: adaptive (e.g. lengthening/shortening the silence between words,
playing back video a little slower, etc) or not adaptive
• Data rate: adaptive (e.g. reduce video quality by compressing video more) or
not adaptive
• Different application requirements=>different service classes (not
only best effort anymore)
• A network that can provide these different levels of service is said to
support QoS
– Integrated service: fine-grained approach, provide QoS to individual
applications or flows
• Allow individual application flows to specify their needs to the routers using an
explicit signaling mechanism (RSVP)
• Scalability is an issue
– Differentiated Service: coarse-grained approach, provide QoS to several
classes of data or aggregated traffic
• Assign packets into a small number of classes that receive differentiated
treatment in the routers
Taxonomy of applications
Applications
Real-Time
Loss, delay
tolerant
adaptive
Delay
adaptive
Non-adaptive
Rate
adaptive
Elastic
Intolerant
Interactive
Rate
adaptive
Asynchronous
Non-adaptive
Interactive-bulk
Components of Integrated Services
architecture
• Flowspec: information of the flow traffic
characteristics and its service
• Reservations (includes reservation signaling protocol)
• Admission control based on flow description and current
load
• Scheduling to meet the reservation
• Traffic shaping at edges to fit reservation
• Traffic policing to mark or drop non-conforming traffic
• Some application adaptation
Types of guarantees
•
•
•
•
Absolute bound on delay and jitter
Absolute bound on delay only
Statistical bound on delay
No quantitative delay bound but
admission control and preferential
treatment
• None
Internet service classes proposed by
IETF
• Guaranteed service
– firm bounds on e2e delays and bandwidth
• Controlled load
– “a QoS closely approximating the QoS that same
flow would receive from an unloaded network
element, but uses capacity (admission) control to
assure that this service is received even when the
network element is overloaded”
• Use a queuing mechanism such as WFQ to isolate the
controlled load traffic from other traffic
• Admission control to limit the total amount of controlled
load traffic
• Best effort
Overview of mechanisms
• Flow specification (flowspec)
– type of service we require
• Admission control
– can the network provide the requested
service?
• Resource reservation protocol
– RSVP
• Packet scheduling
Flowspecs
• Tspec: describes the flow’s traffic
characteristics
• Rspec: describes the service requested
from the network
Traffic Shaping
• Traffic shaping: control traffic in order to conform to
the traffic contract by delaying packets to meet
certain criteria.
– To optimize or guarantee performance (lower latency,
higher usable bandwidth)
– commonly applied at the network edges to control traffic
entering the network, but can also be applied by the traffic
source or in the network
• Token bucket
–
–
–
–
–
tokens are placed in bucket at rate r
if bucket fills, tokens are discarded
sending a packet of size P uses P tokens
if bucket has P tokens, packet sent at max rate,
else must wait for tokens to accumulate
Traffic Policing
• Traffic policing: router monitors the flow traffic for
conformity with a traffic contract
– Drop or tag the packets not conforming to the TSpec that
used to make the reservation
– to enforce compliance with that contract
• Token bucket
– tokens are placed in bucket at rate r
– if bucket fills, tokens are discarded
– When a packet of p bytes arrives, p tokens are removed
from the bucket, and the packet is sent to the network.
– If fewer than p tokens are available, no tokens are removed
from the bucket, and the packet is considered to be nonconformant.
– Drop or mark the non-conformant packets
Token bucket filter
• Described by 2 parameters:
– token rate r: rate of tokens placed in the bucket
– bucket depth B: capacity of the bucket
• Operation:
–
–
–
–
tokens are placed in bucket at rate r
if bucket fills, tokens are discarded
sending a packet of size P uses P tokens
if bucket has P tokens, packet sent at max rate,
else must wait for tokens to accumulate
Token bucket operation
tokens
overflow
tokens
tokens
Packet
Enough tokens
packet goes through,
tokens removed
Packet
Not enough
tokens - wait for
tokens to
accumulate
TB characteristics
• On the long run, rate is limited to r
• On the short run, a burst of size B can
be sent at peak data rate
• Amount of traffic entering at interval T
is bounded by:
– traffic = B + r*T
• Information useful to admission
algorithm
Token bucket specs
Flow A: r = 1 Mbps, B=1 byte
BW
2
Flow B
1
Flow B: r = 1 Mbps, B=1MB
Flow A
1
2
3
Time
Admission control
• When new flow request arrives, look at
Rspec and Tspec and decide whether to
admit or reject
– Can it provide the desired service
requested by the flow, given the currently
available resources without causing any
previously admitted flow to receive worse
service that agreed?
• Not policing
Reservation protocol: RSVP
Upper layer protocols and applications
IP service interface
IP
ICMP IGMP RSVP
Link layer service interface
Link layer modules
RSVP
• Used on connectionless networks
• Relies on soft state: reservations must
be refreshed and do not have to be
explicitly deleted
• Aims to support multicast as effectively
as unicast flows - mcast apps good
candidates for real-time, and are
heterogeneous
• Receiver-oriented approach
Basic message types
• PATH message
• RESV message
• CONFIRMATION message
– generated only upon request
– unicast to receiver when RESV reaches
node with established state
• TEARDOWN message
• ERROR message (if path or RESV fails)
Making a reservation
• Receivers make reservation
• Before making a reservation, receiver
must know:
– type of traffic sender will send (Tspec)
– path the sender’s packets will follow
• Both can be accomplished by sending
PATH messages by the sender
PATH messages
• PATH messages carry sender’s Tspec and
sent from the sender to the receiver
• Record the path from the sender to the
receiver in the PATH message
• Receivers send RESV messages that follow
reverse path and setup reservations
• If reservation cannot be made, user gets an
error
PATH and RESV messages
Sender 1
PATH
R
Sender 2
PATH
RESV (merged)
RESV
R
receiver 1
R
R
RESV
receiver 2
Soft State
• Allow increasing or decreasing the level of
resource reservation
• Adapt to link or router failure and topology
changes
– Routing protocol makes routing changes, RSVP
adjusts reservation state
– In absence of route or membership changes, periodic
PATH and RESV msgs refresh established reservation
state
– When change, new PATH msgs follow new path, new
RESV msgs set reservation
– Non-refreshed state times out automatically
Router handling of RESV
messages
• If new request rejected, send error
message
• If admitted:
– install packet filter into forwarding dbase
– pass flow parameters to scheduler
– activate packet policing if needed
– forward RESV msg upstream
Packet classifying and
scheduling
• Each arriving packet must be:
– classified: associated with the application
reservation
• Examining up to five fields in the packet:
source + destination address, protocol number,
source + destination port
– scheduled: managed in the queue so that
it receives the requested service
• implementation not specified in the service
model
RSVP and multicast
• Reservations from multiple receivers for a
single sender are merged together at
branching points
• Reservations for multiple senders may not be
added up:
– audio conference, not many talk at same time
– Different reservation styles:
• Reserve resources for all speakers
• Reserve resources for any n speakers
• Reserve resources for speakers A and B only
RSVP versus ATM (Q.2931)
• RSVP
–
–
–
–
–
receiver generates reservation
soft state (refresh/timeout)
separate from route establishment
QoS can change dynamically
receiver heterogeneity
• ATM
–
–
–
–
–
sender generates connection request
hard state (explicit delete)
concurrent with route establishment
QoS is static for life of connection
uniform QoS to all receivers
ATM Service Categories
• CBR
– Constant Bit Rate
– Continuous flow of data with tight bounds on delay and delay variation
• rt-VBR
– Real-Time Variable Bit Rate
– Variable bandwidth with tight bounds on delay and delay variation
• nrt-VBR
– Non-Real-Time Variable Bit Rate
– Variable bandwidth with tight bound on cell loss
• UBR
– Unspecified Bit Rate
– No guarantees (i.e., best effort delivery)
• ABR
– Available Bit Rate
– Flow control on source with tight bound on cell loss
Differentiated Services
(DiffServ)
• Analogy:
– airline service, first class, coach, various
restrictions on coach as a function of
payment
• Best-effort expected to make up bulk of
traffic, but revenue from “premium” service
important to economic base (will pay for
more plentiful bandwidth overall)
– Not motivated by real-time, motivated by
economics and assurances
Differentiated Services (cont)
• Divide traffic into a small number of classes
and allocates resource on a per-class basis,
instead of individual flows
– Scalable
– No need for reservations: just mark
packets
• E.g. packet marking can be done at
administrative boundaries before injecting
packets into network
– Significant savings in signaling, much
simpler overall
IP DiffServ
0
4
Version
8
HLen
16
TOS
31
Length
Ident
TTL
19
Flags
Protocol
Offset
Checksum
SourceAddr
DestinationAddr
Options (variable)
Pad
(variable)
Data
•
ToS field => now used for
DiffServ and ECN
– Bits 0-5: Differentiated Services
Code Point (DSCP)
– Bit 6: ECN-capable
– Bit 7: ECN
• IP packets carry 6-bit
service code points
(DSCP)
– Potentially support 64different classes of
services
– In implementation, the
number of DSCPs used in
a network is much
smaller, e.g. simple twoclass network
DiffServ
• Where and how to set the DSCP value
– Host set DSCP based on knowledge of application
requirements
– Router at an administrative boundary set DSCP according to
a certain policy
• What does a router do when it receives a packet
marked with a certain DSCP
– Routers map DSCP to per-hop-behavior (PHB)
– PHBs can be standard or local
– Standard PHBs include
• Default: No special treatment or best effort
• Expedited forwarding (EF): should be forwarded with low delay
and loss
• Assured forwarding (AF): Multiple classes, each class with
multiple drop preference
Expedited Forwarding (EF)
• A router should forward the EF packets with
minimal delay and low loss
• To implement, put all EF packets in a
dedicated EF queue and ensure that the
arrival rate of packets to the queue is less
than the service rate
– Configure the router at the edge of an
administrative domain to allow a certain max rate
of EF packets arrivals into the domain
• Limit the EF packet rate < the bandwidth of the slowest
link in the domain
– Give EF packet queue high priority, e.g. WFQ with
a high weight to EF queue
Assured Forwarding (AF)
• A set of AF PHBs is defined
• AFxy:
– x: class, typically determine a queue for the packet
– y:drop preference
– For example: AF11, AF12 and AF13 packets are
put into the same queue, but AF13 would have a
greater chance to be dropped due to congestion.
AF11 and AF2y would go to different queues
• 12 AF PHBs are recommended by IETF, 4 AF
classes with three drop preference levels each
An Example DiffServ Network
Implementation
• Support Two-type DiffServ
–Premium (expedited) service: (type P)
• User sends within profile, network commits to delivery with
the agreed profile
– Classify packet in one of two queues, use priority
– Shaping at trust boundaries only, using token bucket
• Allocate resource based on the agreed peak rate
– conservative, virtual wire services
– unused premium goes to best effort (subsidy!)
–Assured service: (type A)
• based on expected capacity usage profiles
• traffic unlikely to be dropped if user maintains profile. Outof-profile traffic marked
Premium traffic flow
Company A
Packets in premium
flows have bit set
Premium packet flow
restricted to R bytes/sec
internal
router
host
first hop
router
ISP
border
router
border
router
Unmarked
packet flow
2-bit differentiated service
• Use precedence field to encode P & A type
packets
• P packets are queued at higher priority than
ordinary best effort
• A packets treated preferentially wrt dropping
probability in the normal queue
• Leaf and border routers have input and
output tasks - other routers just output
Leaf router input functionality
Marker 1
Marker N
Arriving
packet
•
Clear
A&P
bits
Packet
classifier
Best effort
Forwarding
engine
Marker: Leaf routers have traffic profiles - they classify packets based
on packet header (destination addr & port, source addr & port, protocol
ID, etc)
– If no profile present, pass as best effort
– If profile is for A:
• mark in-profile packets with A, forward others unmarked
– If profile is for P:
• delay out-of -profile packets to shape into profile
Markers to implement two different
services
P service
Drop on overflow
Packet
input
Wait for
token
Set P bit
Packet
output
A service
No token
Packet
input
Test if
token
token
Set A bit
Packet
output
Router output interface for two-bit
architecture
• 2 queues: P packets on higher priority queue
• Lower priority queue implements RED “In or Out”
scheme (RIO)
P-bit set?
yes
High-priority Q
Packets out
no
If A-bit set
incr A_cnt
Low-priority Q
RIO queue
management
If A-bit set
decr A_cnt
Border router input interface Profile Meters
• At border routers, profile meters test marked flows:
– drop P packets out of profile
– unmark A packets
A set
Arriving
packet
Is packet
marked?
Token
available?
no
Clear A-bit
token
Forwarding
engine
Not marked
token
P set
Token
available?
no
Drop packet
Red with In or Out (RIO)
•
•
•
•
Similar to RED, but with two separate probability curves
Has two classes, “In” and “Out” (of profile)
“Out” class has lower Minthresh, so packets are dropped from this class
first
As avg queue length increases, “in” packets are dropped
P(drop)
1.0
MaxP
AvgLen
Minout
Minin
Maxout
Maxin
Today’s Homework
• Chapter 6
-6.13
-6.32
-6.43
-6.44
4th ed
Due 4/3/15
65