Transcript Chapter 2 - William Stallings, Data and Computer

EEE449
Computer Networks
Congestion
En. Mohd Nazri Mahmud
MPhil (Cambridge, UK)
BEng (Essex, UK)
[email protected]
Room 2.14
Semester 1 2009-2010
Copyright USM
What Is Congestion?
• congestion occurs when the no of packets being
transmitted through the network approaches the
packet handling capacity of the network
• congestion control aims to keep no of packets
below a level at which performance falls off
dramatically
• a data network is a network of queues
• generally 80% utilization is critical
• finite queues mean data may be lost
Semester 1 2009-2010
Copyright USM
Queues at a Node
Semester 1 2009-2010
Copyright USM
Interaction of Queues
Semester 1 2009-2010
Copyright USM
Ideal
Network
Utilization
Semester 1 2009-2010
Copyright USM
Effects of
Congestion
No Control
Semester 1 2009-2010
Copyright USM
Mechanisms for
Congestion Control
Semester 1 2009-2010
Copyright USM
Backpressure
• if node becomes congested it can slow down or
halt flow of packets from other nodes
– cf. backpressure in blocked fluid pipe
– may mean that other nodes have to apply control on
incoming packet rates
– propagates back to source
• can restrict to high traffic logical connections
• used in connection oriented nets that allow hop
by hop congestion control
Semester 1 2009-2010
Copyright USM
Choke Packet
• a control packet
– generated at congested node
– sent to source node
– eg. ICMP source quench
• from router or destination
• source cuts back until no more source quench message
• sent for every discarded packet, or anticipated
• is a rather crude mechanism
Semester 1 2009-2010
Copyright USM
Implicit Congestion Signaling
• transmission delay increases with
congestion
• hence a packet may be discarded
• source detects this implicit congestion
indication
• useful on connectionless (datagram)
networks
– eg. IP based
Semester 1 2009-2010
Copyright USM
Explicit Congestion Signaling
• network alerts end systems of increasing
congestion
• end systems take steps to reduce offered load
• Backwards
– congestion avoidance notification in opposite direction
to packet required
• Forwards
– congestion avoidance notification in same direction as
packet required
Semester 1 2009-2010
Copyright USM
Explicit Signaling Categories
• Binary
– a bit set in a packet indicates congestion
• Credit based
– indicates how many packets source may send
– common for end to end flow control
• Rate based
– supply explicit data rate limit
– nodes along path may request rate reduction
Semester 1 2009-2010
Copyright USM
Traffic Management
• fairness
– provide equal treatment of various flows
• quality of service
– different treatment for different connections
• reservations
– traffic contract between user and network
– carry best-effort or discard excess traffic
Semester 1 2009-2010
Copyright USM
Congestion Control in Packet
Switched Networks
• send control packet to some or all source
nodes
– requires additional traffic during congestion
• rely on routing information
– may react too quickly
• end to end probe packets
– adds to overhead
• add congestion info to packets in transit
– either backwards or forwards
Semester 1 2009-2010
Copyright USM
Congestion Control in TCP/IP
networks
• Credit-based flow control also used for
congestion control
– recognize increased transit times & dropped
packets
– react by reducing flow of data
• Three categories
– retransmission timer management
– window management
– Active queue management
Semester 1 2009-2010
Copyright USM
Retransmission Timer Management
• static timer likely too long or too short
• estimate round trip delay by observing pattern of
delay for recent segments
• set time to value a bit greater than estimate
• simple average over a number of segments
• exponential average using time series
• RTT Variance Estimation
Semester 1 2009-2010
Copyright USM
Window Management
• slow start
– larger windows cause problem on connection created
– at start limit TCP to 1 segment
– increase when data ACK, exponential growth
• dynamic windows sizing on congestion
– when a timeout occurs perhaps due to congestion
– set slow start threshold to half current congestion
window
– set window to 1 and slow start until threshold
– beyond threshold, increase window by 1 for each RTT
Semester 1 2009-2010
Copyright USM
Slow start
• TCP transmission is constrained by
awnd = MIN [credit,cwnd]
– awnd = allowed window (the number of segments that TCP is
currently allowed to send without receiving acknowledgement
– Cwnd = congestion window used during startup and to reduce
flow during period of congestion
– Credit = the amount of unused credit granted in the most recent
acknowledgement
• When a new connection is opened, TCP initialises
cwnd=1
• Each time an ACK is received, the value of cwnd is
increased by 1.
• Cwnd increases exponentially; may be too agressive
Semester 1 2009-2010
Copyright USM
Dynamic window sizing on
congestion
• Begins with a slow start, followed by a
linear growth in cwnd
• When time out occurs
– Set a slow-start threshold equal to half the
current congestion window; ssthresh = cwnd/2
– Set cwnd =1 and perform slow start process
until cwnd = ssthreshold.
– For cwnd >= ssthresh, increase cwnd by one
for each round trip time.
Semester 1 2009-2010
Copyright USM
Window Management
Semester 1 2009-2010
Copyright USM
Fast Retransmit
Fast Recovery
• retransmit timer rather longer than RTT
• if segment lost TCP slow to retransmit
• fast retransmit
– if receive 4 ACKs for same segment (ie 3
duplicate ACKs) then immediately retransmit
since likely lost
• fast recovery
– lost segment means some congestion
– halve window then increase linearly
– avoids slow-start
Semester 1 2009-2010
Copyright USM
Active Queue Management
• Dropping packets is inefficient
• Burst are inevitable. Keeping queue size small
and actively managing queues improves a
router's ability to absorb bursts without dropping
excessive packets
• useful to detect impending congestion conditions
and actively manage congestion before it gets
out of hand.
• technique in which routers actively drop packets
from queues as a signal to senders that they
should slow down
Semester 1 2009-2010
Copyright USM
Active Queue Management
• Recovering from many dropped packets is more
difficult than recovering from a single dropped
packet
• Large queue can translate into delay. Active
queue management allows queues to be
smaller, which improves throughput
• Lock-out occurs when a host fills a queue and
prevents other hosts from using the queue.
Active queue management can prevent this
condition.
Semester 1 2009-2010
Copyright USM
Random Early Drop (RED)
• uses statistical methods to drop packets in a probabilistic way before
queues overflow
• slows a source down enough to keep the queue steady and reduces
the number of packets that would be lost when a queue overflows
• makes two important decisions- when to drop packets and what
packets to drop
• keeps track of an average queue size and drops packets when the
average queue size grows beyond a defined threshold
• The average size is recalculated every time a new packet arrives at
the queue
• RED makes packet-drop decisions based on two parameters:
– Minimum threshold (minth) Specifies the average queue size below
which no packets will be dropped.
– Maximum threshold (maxth) Specifies the average queue size
above which all packets will be dropped
Semester 1 2009-2010
Copyright USM
Explicit Congestion Notification in
TCP/IP network
• Explicit Congestion Notification is an extension proposed to RED
which marks a packet instead of dropping it when the average
queue size is between minth and maxth
• Upon receipt of a congestion marked packet, the TCP receiver
informs the sender (in the subsequent ACK) about incipient
congestion which will in turn trigger the congestion avoidance
algorithm at the sender
• the sender triggers its congestion avoidance algorithm by halving its
congestion window, cwnd, and updating its congestion window
threshold value ssthresh.
• Once it has taken these appropriate steps, the sender sets the CWR
bit on the next data outgoing packet to tell the receiver that it has
reacted to the (receiver's) notification of congestion.
• The receiver reacts to the CWR by halting the sending of the
congestion notifications, ECN-Echo (ECE) to the sender if there is
no new congestion in the network
Semester 1 2009-2010
Copyright USM