High-speed TCP connections

1
Lecture 14

• High-speed TCP connections
• Wraparound
• Keeping the pipeline full
• Estimating RTT
• Fairness of TCP congestion control
• Internet resource allocation and QoS

2
Protection against wraparound

• What is wraparound A byte with a sequence number
  x may be sent at one time and then on the same
  connection a byte with the same sequence number x
  may be sent again.
• Wrap Around controlled by the 32-bit SequenceNum
• The maximum lifetime of an IP datagram is 120 sec
  thus we need to have a wraparound time at least
  120 sec.
• For slow links OK but no longer sufficient for
  optical networks.
• Bandwidth Time Until Wrap Around

Bandwidth T1 (1.5Mbps) Ethernet (10Mbps) T3
(45Mbps) FDDI (100Mbps) STS-3 (155Mbps) STS-12
(622Mbps) STS-24 (1.2Gbps)
Time Until Wrap Around 6.4 hours 57 minutes 13
minutes 6 minutes 4 minutes 55 seconds 28 seconds
3
Keeping the pipe full

• The SequenceNum, the sequence number space (32
  bits) should be twice as large as the window size
  (16 bits). It is.
• The window size (the number of bytes in transit)
  is given by the the AdvertisedWindow field (16
  bits).
• The higher the bandwidth the larger the window
  size to keep the pipe full.
• Essentially we regard the network as a storage
  system and the amount of data is equal to (
  bandwidth x delay )

4

• Required window size for a 100 msec RTT.

Bandwidth T1 (1.5Mbps) Ethernet (10Mbps) T3
(45Mbps) FDDI (100Mbps) STS-3 (155Mbps) STS-12
(622Mbps) STS-24 (1.2Gbps)
Delay x Bandwidth Product 18KB 122KB 549KB 1.2MB 1
.8MB 7.4MB 14.8MB
5
Original Algorithm for Adaptive Retransmission

• Measure SampleRTT for each segment/ACK pair
• Compute weighted average of RTT
• EstimatedRTT a x EstimatedRTT (1- a) x
  SampleRTT
• where 0.8 lt a lt 0.9
• Set timeout based on EstimatedRTT
• TimeOut 2 x EstimatedRTT

6
Karn/Partridge Algorithm

• Do not sample RTT when re-transmitting
• Double timeout after each retransmission

7
Karn/Partridge Algorithm
8
Jacobson/Karels Algorithm

• New calculation for average RTT
• Diff SampleRTT - EstimatedRTT
• EstimatedRTT EstimatedRTT (d x
• Deviation Deviation d(Diff- Deviation)
• where d is a fraction between 0 and 1
• Consider variance when setting timeout value
• TimeOut m x EstimatedRTT f x Deviation
• where m 1 and f 4
• Notes
• algorithm only as good as granularity of clock
  (500 microseconds on Unix)
• accurate timeout mechanism important to
  congestion control (later)

9
Congestion Control Mechanisms

• The sender must perform retransmissions to
  compensate for lost packets due to buffer
  overflow.
• Unneeded retransmissions by the sender due to
  large delays causes a router to use link
  bandwidth to forward unneeded copies of a packet.
• When a packet is dropped along a path the
  capacity used used at each upstream routers to
  forward packets to the point where it was dropped
  was wasted.

10
Delay/Throughput Tradeoffs
11
(No Transcript)
12
Router with infinite buffer capacity
13
Fairness of TCP congestion mechanism
14
Flows and resource allocation

• Flow sequence of packets with a common
  characteristics
• A layer-N flow ? the common attribute a layer-N
  attribute
• All packets exchanged between two hosts ? network
  layer flow
• All packets exchanged between two processes ?
  transport layer flow

15
(No Transcript)
16
Min-max fair bandwidth allocation

• Goal fairness in a best-effort network.
• Consider
• Unidirectional flows
• Routers with infinite buffer space
• Link capacity is the only limiting factor.

17
Algorithm

• Start with an allocation of zero Mbps for each
  flow.
• Increment equally the allocation for each flow
  until one of the links of the network becomes
  saturated. Now all the flows passing through the
  saturated link get an equal fraction of the link
  capacity.
• Increment equally the allocation for each flow
  that does not pass through the first saturated
  link until a second link becomes saturated. Now
  all the flows passing through the saturated link
  get an equal fraction of the link capacity.
• Continue by incrementing equally the allocations
  of all flows that do not use a saturated link
  until all flows use at least one saturated link.

18
QoS in a datagram network?

• Buffer acceptance algorithms.
• Explicit Congestion Notification.
• Packet Classification.
• Flow measurements

19
Buffer acceptance algorithms

• Tail Drop.
• RED Random Early Detection
• RIO Random Early Detection with In and Out
  packet dropping strategies.

20
(No Transcript)
21
Explicit Congestion Notification (ECN)

• The TCP congestion control mechanism discussed
  earlier has a major flow it detects congestion
  after the routers have already started dropping
  packets. Network resources are wasted because
  packets are dropped at some point along their
  path, after using link bandwidth as well as
  router buffers and CPU cycles up to the point
  where they are discharged.
• The question that comes to mind is Could
  routers prevent congestion by informing the
  source of the packets when they become lightly
  congested, but before they start dropping
  packets? This strategy is called source quench.

22
Source quench

• Send explicit notifications to the source, e.g.,
  use the ICMP. Yet, sending more packets in a
  network that shows signs of congestion may not be
  the best idea.
• Modify a congestion notification flag in the IP
  header to inform the destination then have the
  destination inform the source by setting a flag
  in the TCP header of segments carrying
  acknowledgments.

23
Problems with ECN

• (1) TCP must be modified to support the new flag.
• (2) Routers must be modified to distinguish
  between ECN-capable flows and those who do not
  support ECN.
• (3) IP must be modified to support the congestion
  notification flag.
• (4) TCP should allow the sender to confirm the
  congestion notification to the receiver, because
  acknowledgments could be lost.

24
Maximum and minimum bandwidth guarantees

• A. Packet classification.
• Identify the flow the packet belongs to.
• At what layer should be done? Network layer?
• At each router ? too expensive.
• The edge routers may be able to do that.
• At application layer? Difficult.
• MPLS multi protocol label switch. Add an extra
  header in front of the IP header. Now a router
  decides the output link based upon the input link
  and the MPLS header.

25
Maximum and minimum bandwidth guarantees

• B. Flow measurements
• How to choose the measurement interval to
  accommodate bursty traffic?
• Token bucket

26
The token bucket filter

• Characterized by (1) A token rate R, and (2)
  The depth of the bucket, B
• Basic idea the sender is allocated tokens at a
  given rate and can accumulate tokens in the
  bucket until the bucket is filled. To send a byte
  the sender must have a token. The maximum burst
  can be of size B because at most B token can be
  accumulated.

27
Example

• Flow A generates data at a constant rate of 1
  Mbps. Its filter will support a rate of 1 Mbps
  and a bucket depth of 1 byte,
• Flow B alternates between 0.5 and 2.0 Mbps. Its
  filter will support a rate of 1 Mbps and a bucket
  depth of 1 Mbps
• Note a single flow can be described by many
  token buckets.

28
Example
29
(No Transcript)
30
Token bucket

• L packet length
• C of tokens in the bucket
• --------------------------------------------------
  -
• if ( L lt C )
• accept the packet
• C C - L
• else
• drop the packet

31
A shaping buffer delays packets that do not
confirm to the traffic shape

• if ( L lt C )
• accept the packet
• C C - L
• else / the packet arrived early, delay it /
• while ( C lt L )
• wait
• transmit the packet
• C C - L

32
A QoS Capable Router