CONGESTION CONTROL - PowerPoint PPT Presentation

1 / 30
About This Presentation
Title:

CONGESTION CONTROL

Description:

a heterogeneous network that uses different lower layer ... users with high demand should be asked to throttle. 8/17/09. 21. Congestion control. Iskra Dj. ... – PowerPoint PPT presentation

Number of Views:35
Avg rating:3.0/5.0
Slides: 31
Provided by: Oliver136
Category:

less

Transcript and Presenter's Notes

Title: CONGESTION CONTROL


1
CONGESTION CONTROL
2
Flow Control, Congestion control, Quality of
Service
  • Flow control
  • adjusting the speed between sender and receiver
  • Congestion control
  • penalty for dynamically shared resources
  • Similarities with the highway system
  • The influence of routing on the congestion
  • Quality of service

3
Quality of Service
  • Internet
  • a heterogeneous network that uses different lower
    layer technologies
  • the best effort service
  • Service quality - not precisely defined
  • Parameters used
  • delay, jitter, bandwidth, reliability

4
Consequences in the Case of Overload
DELAY
OFFERED LOAD
5
Controlled Environment
Ideal
Controlled
EFFECTIVE THROUGHPUT
Uncontrolled
OFFERED LOAD
6
PROBLEMS CAUSED BY OVERLOAD
  • Loss of efficiency
  • Unfair allocation of resources among competing
    users
  • Various types of loops and deadlocks

7
LOSS OF EFFICIENCY
  • The main cause - wastage of resources
  • The most commonly wasted are
  • communications capacity
  • storage capacity
  • Some users unnecessary monopolize the buffers at
    some congested node

8
EXAMPLE FOR WASTAGE OF BUFFERS
9
UNFAIRNESS
  • A natural byproduct of uncontrolled competition
  • Some users can obtain preferential treatment due
    to
  • relative position in the network
  • particular selection of network and traffic
    parameters

10
EXAMPLE
11
Another example
r14 1
1
1
1
1
2
3
4
r34 1
r23 1
r12 1
  • When 1, 2, 3, and 4 are fully using the capacity
    of the links 1 to 4 traffic is stopped
  • If 1/2 of traffic is allowed to everyone, the
    total traffic is smaller

12
Loops and Deadlocks
  • Routing loops
  • caused by inconsistency in routing tables
  • Deadlocks
  • Caused by lack of buffers in the nodes
  • reassemble deadlocks
  • store-and-forward deadlocks

13
Performance Goals
  • Minimum delay and maximum throughput
  • conflict between them
  • Power - often used to reflect tradeoff between
    the delay and throughput
  • Power throughput /delay
  • a gt 1 if throughput is more important (bulk
    traffic)
  • a lt1 if delay is more important (interactive
    traffic)
  • a 1 if they are equally important

a
14
Maximum Power
  • Obtained at the knee of the curve
  • Below the knee - underutilized
  • Over the knee - overutilized

DELAY
Max. power
OFFERED LOAD
15
Handling Congestion
  • Two categories of handling congestion
  • congestion recovery
  • reactive in nature
  • without congestion recovery, the network may come
    to a zero throughput
  • congestion avoidance
  • preventive in nature
  • even when it is implemented, congestion recovery
    is still required

16
Quality of Service in Internet
  • Most exploited expression in the last years
  • Hard to implement in the best effort
    environment
  • Usual response is
  • using technology that implements QoS
  • ATM, Frame relay
  • increase of bandwidth
  • alternate physical paths
  • Congestion avoidance mechanisms

17
Methods proposed
  • Source quench
  • Drop of packets using various policies
  • Queuing discipline
  • Slow start
  • Slow Start and Search
  • DUAL
  • TCP Vegas

18
Source quench
  • The oldest method in Internet used for congestion
    recovery
  • ICMP message
  • when a router drops the packets from the queue,
    it sends the ICMP message to the source
  • the message is sent with a limited frequency to
    conserve CPU resources during the time of heavy
    load

19
Source Quench Drawbacks
  • End system is not informed about the reasons of
    ICMP message
  • the last packet in the queue is drooped
  • not always indication that that particular system
    is cause of congestion

20
Random Drop
  • Proposed as an improvement to packet dropping
  • Packet to be dropped is selected randomly
  • users packet with demand is with low
    probability of drop
  • users with high demand should be asked to throttle

21
Other Proposals for Packet Drop
  • Type of service bit should be considered
  • the higher the precedence, the lower the
    probability of drop
  • has to be implemented on every router in the path
    - sometimes impossible, having in mind different
    paths packet can take

22
Queue Discipline
  • FIFO - usual discipline
  • Priority queues
  • form several queues with different priorities,
    the one with highest priority is first
  • Weighted round-robin
  • form different queues, but serve all of them with
    weighted preference
  • Weighted fair queuing

23
Weighted Fair Queuing
  • Calculates time of each queued packet
  • Packets with smallest finish time is transmitted
    first
  • Pretty fair to everyone
  • Complicated implementation

24
TCPs Slow Start Mechanism
  • A window flow-control mechanism
  • At initial startup phase - one segment is
    transmitted
  • Upon receiving an ACK two segments are
    transmitted for every ACK received
  • Upon packet loss, the window is drooped to one
    half and after that it is increased one segment
    after every ACK

25
TCP Flow Behavior
Packet drop level
Network buffer capacity
Available bandwidth
TCP flow rate
26
Effectiveness of Slow Start
  • For four TCP conversations going between eight
    computers on 10Mbps LAN and 230Kbps link over
    Internet 200 increase increase in effective
    bandwidth was noticed when Slow start was
    implemented

27
Slow Start and Search (Three-S)
  • Uses the change in throughput as an indication of
    congestion
  • At first the window size is set to one basic
    adjustment unit (BAU) and increased by one packet
    after after each ACK
  • When a packet times out window is set to one BAU
    and increased by one BAU

28
DUAL
  • Similar to Slow start
  • It uses the round trip time (RTT) in addition to
    packet loss to detect detection
  • Every two RTT DUAL checks if the current RTT is
    greater than the previous and reduces the
    congestion window by 7/8

29
TCP Vegas
  • Proposes a new implementation of TCP
  • Uses a time stamp on each packet sent to
    calculate the RTT for each ACK received
  • Can predict with certain probability if packet
    will be lost
  • Retransmit the packets before they time out

30
Conclusion
  • Congestion avoidance and congestion recovery
    should receive attention
  • For the quality of service in the Internet there
    are no perfect solutions, at least not for the
    foreseeable future
  • It is possible to deliver differentiated levels
    of best effort traffic
Write a Comment
User Comments (0)
About PowerShow.com