1. Congestion Control Definitions
and  Flow control:: keep a fast sender from
overrunning a slow receiver.
Resource Allocation  Congestion control:: the efforts made by network
nodes to prevent or respond to overload
conditions.
Reference:- “Computer Networks A Systems Congestion control is intended to keep a fast
Approach”, Third Edition, Peterson and sender from sending data into the network
Davie, due to a lack of resources in the network
{e.g., available link capacity, router buffers}.
Morgan Kaufmann, 2003.
1 2
Congestion Control
Congestion
3
 Congestion control is concerned with the 6
bottleneck routers in a packet switched
network. 1
4
 Congestion control can be distinguished 8
from routing in that sometimes there is no 2
7
5
way to ‘route around’ a congested router.
Copyright ©2000 The McGraw Hill
Companies
3 4
Created with Print2PDF. To remove this line, buy a license at: http://www.binarynow.com/

2. Flows
Source
1 10-M  flow :: a sequence of packets sent between a
bps
Ethe
rnet Router source/destination pair and following the same
Destination
1.5-Mbps T1 link route through the network.
DDI
Source 100
-Mb
ps F  Connectionless flows within the TCP/IP model::
2
The connection-oriented abstraction, TCP, is
implemented at the transport layer while IP
provides a connectionless datagram delivery
Congestion in a packet-switched
packet- service.
network  With connectionless flows, there exists no state at
the routers.
5 6
Flows Source
1
Router
Destination
 Connection-oriented flows (e.g., X.25) – 1
connection-oriented networks maintain hard state Router
at the routers. Source
2
 Soft state :: represents a middle ground where Router
soft state is not always explicitly created and Destination
2
removed by signaling. Source
 Correct operation of the network does not depend 3
on the presence of soft state, but soft state can
permit the router to better handle packets. Multiple Flows passing through a set
of routers
7 8
Created with Print2PDF. To remove this line, buy a license at: http://www.binarynow.com/

3. Service Lack of Congestion Control
 Best-effort service :: The hosts are given no
opportunity to ask for guarantees on a Controlled
Throughput
flow’s service.
 QoS (Quality of Service) :: is a service Uncontrolled
model that supports some type of
guarantee for a flow’s service.
Offered load
9 10
Congestion Control Taxonomy Congestion Control Taxonomy
 Router-Centric  Reservation-Based – the hosts attempt to
The internal network routers take responsibility for:

 Which packets to forward
reserve network capacity when the flow is
 Which packets to drop or mark established.
 The nature of congestion notification to the hosts.  The routers allocate resources to satisfy
 This includes the Queuing Algorithm to manage the reservations or the flow is rejected.
buffers at the router.
 Host-Centric  The reservation can be receiver-based (e.g.,
 The end hosts adjust their behavior based on RSVP) or sender-based.
observations of network conditions.
 (e.g., TCP Congestion Control Mechanisms)
11 12
Created with Print2PDF. To remove this line, buy a license at: http://www.binarynow.com/

4. Congestion Control Taxonomy
Evaluation Criteria
 Feedback-Based - The transmission rate is adjusted
(via window size) according to feedback received from  Evaluation criteria are needed to decide how well a
the sub network. network effectively and fairly allocates resources.
 Explicit feedback – FECN
 Implicit feedback – router packet drops.  Effective measures – throughput, utilization,
 Window-Based - The receiver sends an advertised efficiency, delay, queue length, goodput and power.
window to the sender or a window advertisement can throughput α
be used to reserve buffer space in routers. Power = --------------
 Rate-Based – The sender’s rate is controlled by the delay
receiver indicating the bits per second it can absorb.
Where 0 < α < 1
13 14
Fairness Congestion Control
 Jain’s fairness index (at the router)
 Queuing algorithms determine:
For any given set of user throughputs (x1, x2,…xn ), the
(x x2
fairness index to the set is defined:  How packets are buffered.
 Which packets get transmitted.
 n 
2
 Which packets get marked or dropped.
  xi 
f(x1, x2, …, xn) =  i 1   Indirectly determine the delay at the router.
n
n  x i2  Queues at outgoing links drop/mark packets to
 Max-min fairness i 1 implicitly signal congestion to TCP sources.
Essentially ‘borrow’ from the rich-in-performance to help the poor-in-
performance
 Remember to separate queuing policy from queuing
For example, CSFQ
mechanism.
15 16
Created with Print2PDF. To remove this line, buy a license at: http://www.binarynow.com/

5. Congestion Control Drop Tail Router [FIFO]
(at the router)
 Some of the possible choices in queuing
algorithms: • First packet to arrive is first to be transmitted.
 FIFO (FCFS) also called Drop-Tail • FIFO queuing mechanism that drops packets from
 Fair Queuing (FQ) the tail of the queue when the queue overflows.
 Weighted Fair Queuing (WFQ) • Introduces global synchronization when packets are
 Random Early Detection (RED) dropped from several connections.
 Explicit Congestion Notification (ECN). • FIFO is the scheduling mechanism, Drop Tail is the
policy
17 18
Priority Queuing Priority Queuing
 Mark each packet with a priority (e.g., in  Problem:: high priority packets can
TOS (Type of Service field in IP) ‘starve’ lower priority class packets.
 Implement multiple FIFO queues, one for  Priority queuing is a simple case of
each priority class. “differentiated services” [DiffServ].
 Always transmit out of the highest priority  One practical use in the Internet is to
non-empty queue. protect routing update packets by giving
 Still no guarantees for a given priority them a higher priority and a special queue
class. at the router.
19 20
Created with Print2PDF. To remove this line, buy a license at: http://www.binarynow.com/

6. Fair Queuing [FQ] Flow 1
 The basic problem with FIFO is that it does Flow 2
Round-robin
not separate packets by flow. service
Flow 3
 Another problem with FIFO :: an “ill-
behaved” flow can capture an arbitrarily
large share of the network’s capacity. Flow 4
Idea:: maintain a separate queue for each
flow, and Fair Queuing (FQ) services
these queues in a round-robin fashion.
Fair Queuing
21 22
Fair Queuing [FQ] Fair Queuing [FQ]
 FQ simulates bit-by-bit behavior by using
timestamps (too many details for here!).
 “Ill-behaved” flows are segregated into
 One can think of FQ as providing a guaranteed
their own queue. minimum share of bandwidth to each flow.
 There are many implementation details for  FQ is work-conserving in that the server is never
idle as long as there is a customer in the queue.
FQ, but the main problem is that packets * Note: The per-flow state information kept at
are of different lengths  simple FQ is not the router is expensive (it does not scale).
fair!!
 Ideal FQ:: do bit-by-bit round-robin.
23 24
Created with Print2PDF. To remove this line, buy a license at: http://www.binarynow.com/

7. Weighted Fair Queuing [WFQ]
WFQ idea:: Assign a weight to each flow (queue) such that the
weight logically specifies the number of bits to transmit each
time the router services that queue.
 This controls the percentage of the link capacity that the flow
will receive.
 The queues can represent “classes” of service and this
becomes DiffServ.
 An issue – how does the router learn of the weight
assignments?
 Manual configuration
 Signaling from sources or receivers.
25
Created with Print2PDF. To remove this line, buy a license at: http://www.binarynow.com/