CS640: Introduction to

Computer Networks
Aditya Akella

Lecture 15
TCP – III
Reliability and Implementation Issues
 Reliability
• TCP provides a “reliable byte stream”
– “Loss recovery” key to ensuring this abstraction
– Sender must retransmit lost packets

• Challenges:
– When is a packet lost?
• Congestion related losses
• Reordering of packets
– How to tell the difference between a delayed packet and
lost one?
– Variable packet delays
• What should the timeout be?
– How to recover from losses?

2
 Loss Recovery in a Sliding Window
setup
• Sliding window with cumulative acks
– Receiver can only return a single “ack” sequence number to
the sender.
– Acknowledges all bytes with a lower sequence number
– Starting point for retransmission
– Duplicate acks sent when out-of-order packet received

• Sender only retransmits a single packet.

– Only one that it knows is lost
• Sent after timeout

• Choice of timeout interval  crucial

3
 Round-trip Time Estimation
• Reception success known only after one RTT
– Wait at least one RTT before retransmitting

• Importance of accurate RTT estimators:

– Low RTT estimate
• unneeded retransmissions
– High RTT estimate
• poor throughput

• RTT estimator must adapt to change in RTT

– But not too fast, or too slow!

4
 Jacobson’s Retransmission
Timeout (RTO)
• Original setting:
– Round trip times exponentially averaged:
• New RTT =  (old RTT) + (1 - ) (new sample)
• Recommended value for : 0.8 - 0.9
– Retransmit timer set to (2 * RTT)
– But this can lead to spurious retransmissions

• Key observation:
– At high loads round trip variance is high

• Solution:
– Base RTO on RTT and deviation
• RTO = RTT + 4 * rttvar
– new_rttvar =  * dev + (1- ) old_rttvar
• Dev = linear deviation
• Inappropriately named – actually smoothed linear deviation

5
 AIMD Implementation
• If loss occurs when cwnd = W
– Network can handle < W segments
– Set cwnd to 0.5W (multiplicative decrease)
– Known as “congestion control”

• Upon receiving ACK

– Increase cwnd by (1 packet)/cwnd
• What is 1 packet?  1 MSS worth of bytes
• MSS = maximum segment size
– After cwnd packets have passed by  approximately
increase of 1 MSS
– Known as “congestion avoidance”

• Together these implement AIMD

6
 Control/Avoidance Behavior in
the presence of timeouts
Congestion
Window

Cut Grabbing
Time
Packet loss
+ Timeout Congestion back
Window Bandwidth
and Rate
7
 Improving Loss Recovery:
Fast Retransmit
• Waiting for timeout to retransmit is inefficient

• Are there quicker recovery schemes?

– Use duplicate acknowledgements as an indication
– Fast retransmit

• What are duplicate acks (dupacks)?

– Repeated acks for the same sequence

• When can duplicate acks occur?

– Loss
– Packet re-ordering

• Assume re-ordering is infrequent and not of large magnitude

– Use receipt of 3 or more duplicate acks as indication of loss
– Don’t wait for timeout to retransmit packet

8
 Fast Retransmit

Retransmission
X
Sequence No Duplicate Acks

Note: Timeouts can still

happen (burst losses in
a window)
Packets
Acks
Time
9
 How to Change Window
• When a loss occurs have W packets
outstanding
– A bunch of dupacks arrive
– Rexmit on 3rd dupack
– But dupacks keep arriving
– Must wait for a new ack to send new packets

• New cwnd = 0.5 * cwnd

– Send new cwnd packets in a burst when new ack
arrives
– Risk losing “self clocking” or “packet pacing”

10
 Packet Pacing
• In steady state, a packet is sent when an ack is
received
– Data transmission remains smooth, once it is smooth (steady
state)
– “Self-clocking” behavior
– When self clocking is lost  send packets in a burst  could
momentarily overflow network capacity

11
 Preserving Clocking:
Fast Recovery
• Each duplicate ack notifies sender that single
packet has cleared network

• When < cwnd packets are outstanding

– Allow new packets out with each new duplicate
acknowledgement

• Behavior
– Sender is idle for some time – waiting for ½ cwnd worth
of dupacks
– Transmits at original rate after wait
• Ack clocking rate is same as before loss

12
 Fast Recovery (Reno)
Sent for each dupack after
W/2 dupacks arrive

Sequence No

Packets
Acks
Time 13
 Dupacks may not be enough:
Timeouts can still happen!
X
X
X
X
Now what? - timeout
X
Sequence No

Packets
Acks
Time
14
 Reaching Steady State
• Doing AIMD is fine in steady state…
– But how to get to steady state?

• How does TCP know what is a good initial rate to start with?

• Quick initial phase to help get up to speed

– Called “slow” start (!!)
– Losts of packets sent back to back
– Paced out by the bottleneck link
– Eventually, self clocking is established!

Pb
Pr

Sender Receiver

As Ar
Ab 15
 Slow Start
• Slow start
– Initialize cwnd = 1
– Upon receipt of every
ack, cwnd = cwnd + 1

• Implications
– Window actually
increases to W in RTT
* log2(W)
– Can overshoot window
and cause packet loss
16
 Slow Start Example
One RTT
0R
1
One pkt time

1R 1
2
3
2R 2 3
4 6
5 7
4 5 6 7
3R
8 10 12 14
9 11 13 15
17
 Return to Slow Start
• If too many packets are lost self
clocking is lost as well
– Need to implement slow-start and
congestion avoidance together

• When timeout occurs set ssthresh to

0.5w
– If cwnd < ssthresh, use slow start
– Else use congestion avoidance
18
The Whole TCP “Saw Tooth”
Congestion
Window Timeouts
may still
occur

Slowstart Fast Time

Initial
to pace Retransmit
Slowstart
packets and Recovery

19
 TCP Performance and Role of
Buffering
• Can TCP saturate a link? Depends on buffering in network
• Congestion control
– Increase utilization until… link becomes congested
– React by decreasing window by 50%
– Window is proportional to rate * RTT
• Unbuffered link
– The router can’t fully utilize the link
• If the window is too small, link is not full
• If the link is full, next window increase causes drop
• With no buffer TCP achieves 75% utilization
W Minimum window
for full
utilization

20
t
 TCP Performance
• In the real world, router queues play important role
– Role of Buffers  If window is larger, packets sit in queue on
bottleneck link

• If we have a large router queue  can get 100% utilization

– But, router queues can cause large delays

• How big does the queue need to be?

– Windows vary from W  W/2
• To make sure that link is always full
W/2 > RTT * BW
• W = RTT * BW + Qsize
 Qsize > RTT * BW
– Ensures 100% utilization
– Delay?
• Varies between RTT and 2 * RTT
• Queuing between 0 and RTT

21
 Buffered Link
W

Buffer Minimum window

for full
utilization

• With sufficient buffering we achieve full link

utilization
– The window is always above the “critical” threshold
– Buffer absorbs changes in window size
• Buffer Size = Height of TCP Sawtooth
• Minimum buffer size needed is 2T*C
– This is the origin of the rule-of-thumb
22