Transport Layer Flashcards
what are the two transport layer protocols
tcp
udp
which is more reliable tcp or udp
tcp
why are tcp packets sent in order
needs to be assembled in a particular order
what services are not available in the internet transport layer protocols
delay guarantees
bandwidth gaurantee
what type of services use TCP
email, web
because they require that every bit of data which was transferred from the sender to reach the receiver
what types of services use UDP
fast services like video streaming
because it is not essential that all packets reach the receiver as the receiver can interpret the message with some packets missing, can tolerate some loss
what is multiplexing
handling data from multiple sockets
what is demultiplexing
using header info to deliver received segments to the correct socket
how many addresses does the typical host have
3 addresses
how does demultiplexing work
host uses source ip address, destination ip address and port numbers to direct segment to appropriate socket
what is the 4-tuple that a TCP socket can be identified by
source ip address
source port number
destination ip address
destination port number
example of a server supporting many simultaneous TCP sockets
webserver
gets requested from many different users and everyone’s queries need to get to the right processes
the source IP address will allow the server to differentiate the different machines
what does UDP stand for
user datagram protocol
what does it mean when we say that udp is connectionless
no handshaking between udp sender and receiver
each udp segment handles independently of others
why is there no connection establishment in udp
connection establishment adds delay
why is there a udp
- simple (no connections)
- small header size
- no congestion control
what does it mean if the underlying channel is completely reliable
no bit errors
all arrives at the end in exact sequence sent
no loss of packets
what can cause bit errors
noise on the line
how to recover from bit errors
- checksum
- negative acks
what is a negative ack
receiver explicitly tells sender that the packet had erros
with no timers, what happens if an ack/nack is corrupted
sender does not know what happened at the receiver and will wait an infinite amount of time for an ack
it does not want to transmit because there could be a possibility of duplicate packets
in protocol with no timers, how can duplicate packets be handled
sequence numbers
receiver can discard packets with duplicate sequence number
how could a protocol avoid nacks
instead of NAK, receiver sends ACK for last packet received ok, stating explicitly the seq #
if the sender received an ACK with a different seq # than expected, it knows to retransmit the packet that it expected to get the ACK for
approach for channel where packets can get lost
use timers
how are duplicates handled in protocols with timers
with seq #, duplicates are discarded
what is the round trip time
time taken to transmit packet and receive ack for this packet
what is pipelining
multiples packets can be in flight, yet to be acknowledged
two forms of pipelined protocols
go back n
selective repeat
how does the utilization increase as the number of packets allowed in the pipeline increase
exponentially
why can we not have an indefinite amount of packets allowed in the pipeline
buffering problems
overwhelming the network
what end points have a buffer in the go back n protocol
just sender
what is a cumulative ack
wait a little longer after receiving packet, see if more come through, if they do then ack them all together
in go back n, what does the timer represent
the oldest un acked pacl
if the timer expires in go back n, what happens
all the packets in the window (unacked) are retransmitted
which endpoints have buffers in selective repeat
both sender and receiver
is there cumulative acks in selective repeat
no each packet must be acked individually
how many timers are there in selective repeat
one for each unacked packet
why is having one timer per packet in selective repeat more efficient
if a timeout occurs, only the unacked pack needs to be retransmitted
what does the window size depend on
the network situation
can out of order packets be received in go back n
no
what happens with out of order packets in go back n
discarded
can out of order packets be received in selective repeat
yes, because there is a buffer
what occurs on timeout on sender side in selective repeat
resend packet the timer is associated with
reset the timer
what happens in selective repeat when the sender received an ack for a frame within the send base
mark packet as received
if the seq # == send base, advance the window forward
what happens in selective repeat when the receiver gets a packet that is within its receive base
send an ack packet
out of order = buffer
in order = deliver, advance window
what happens in selective repeat when the receiver gets a packet that is not within its receive base
greater than receive base = ignore
less than receive base = duplicate packet = send another ack for this packet to the sender, other one may have gotten lost
what is the dilemma with windows wrapping around in selective repeat
receiver see’s no difference between sequence numbers
some retransmissions then can be treated as new data
what is the dilemma with windows wrapping around in selective repeat
receiver see’s no difference between sequence numbers
some retransmissions then can be treated as new data
how can we resolve the wrap around window problem with selective repeat
about half of the sequence number typically
how can we resolve the wrap around window problem with selective repeat
about half of the sequence number typically
what does it mean to say TCP is point to point
runs only in the end systems, client and server, not the intermediate routers
what does MSS stand for
maximum segment size
what is the MSS
MTU minus 40 bytes (TCP and IP header which are 20 bytes each typically)
what does it mean to say that TCP is connection orientated
handshaking is required before any real data can be transmitted
initialises state on both sides
why is there flow control in TCP
so that the sender does not overwhelm the receiver
in TCP the ack sent also contains
the seq # of the next byte expected from the other side
what does rtt stand for
round trip time
if the timer is shorter than the RTT what happens
much more unnecessary resending will be done as timeouts are premature
what happens if the timers are much longer than the RTT
slow reactions to lost segments
how do we estimate RTT
sampleRTT
measured time from segment transmission until ack receipt(ignoring retransmissions as these do not give a true picture of RTT)
what is estimated RTT
average of sampleRTTs
why is estimatedRTT better than sampleRTT
averaging it out as there could be outliers
how is the timeout interval calculated
estimatedRTT + 4*DevRTT
what is the DevRTT
deviations of the sampleRTTs from the estimated RTT
what is the initial value of the timeout interval in TCP
1 second
how does the value of the timeout interval change when a timeout occurs
value doubles
what are retransmissions triggered by in TCP
timeouts
duplicate acks
what additional services does TCP add on top of IP
pipelining
cumulative ACKs
single retransmission timers
how is sequence number deciede
byte stream number of first data byte in segment
how does tcp fast retransmit work
lost segments can be detected via duplicate acks
if sender receives 3 Acks for the same data (Triple duplicate acks) => resend unacked segment with smallest seq #
why is tcp fast retransmit more efficient
saves time rather than waiting for timeout
what is default receive buffer space
4096
what is the free buffer space called in receiver
receive window
how does the receive window help the sender not overwhelm the receiver
sender limits amount of in flight data to receivers rwnd value
if receiver window is full and its advertised rwnd = 0, and no data to send from the sender side, what happens
each side is waiting for the other to do something
sender will send a single byte every so often to see when things clear up and there is more space in the window
how does the sender know the size of the receiver window in TCP
the receiver advertises the free buffer space in the TCP header
what is a handshake
agree to establish connection
agree on connection parameters such as buffer space on each side, ports shared and some other variables
what is a synbit
synchronisation bit
what are the steps involved in a three way handshake
client sends tcp syn message , synbit = 1 and seq # = x, initiating the connecyion
server also chooses random seq #, acking the syn message received, synbit = 1, seq # = y, ackbit = 1 , ackNum = x+1 (next expected)
finally the client acks the servers seq # and say what it expected next, (y+1)
what side must start the handshake process
client
what side must start the connection closing process
either
what is the connection closing process
send finbit = 1, seq # =x
other side immediately sends ack, ackbit =1, acknum = x+1
once the receiver has wrapped things up on it’s end, it also sends finbit =1, seq =y
snder acks this
what is a syn flood attack
attacker sends larger amounts of syn segments and handshakes never get completed
these hanging connections eventually just grind down the server
how are syn attacks resolved
syn cookies
what does a syn cookie consist of
src IP address & port
dest IP address & port
secret seed
what does ISN stand for
initial sequence number
what is an initial sequence number the same as
syn cookies
how does syn cookies work
server sends syn ack
nothing is allocated then on the server side until the client returns the ack segment (third part of handshake)
how does syn cookies stop illegitimate clients
because only legitimate clients will return the ack segment
what is congestion control
making sure not too many sources are sending too much data for the network to handle
how does congestion control differ from flow control
flow control makes sure an individual sender does not overwhelm an individual receiver
where as congestion control is a network problem, making sure the network is not overwhelmed
how do lost packets occur
in routers
buffer overflow
how do long delays occur in a network
because of queueing happening at routers
where are long delays more of an issue
time sensitive data
there is no way to know though if the data being transmitted is time sensitive as all packets are treated the same
what is end to end congestion control
there is no explicit feedback from the network so we must infer what is happening in the network from the packets that are lost and delayed
what protocol uses end to end congestion control
TCP
what is network assisted congestion control
router supply feedback to the end systems
eg the routers use a bit to tell to not send any more data until it says so
what protocol uses network assisted congestion control
ATM, single bit indicating congestion
in the core of the network where fast and high speeds are neede
what does AIMD stand for
additive increase multiplicative decrease
what is aimd
keep increasing congestion window by 1 until a lost is detected
multiplicative decrease = cut congestion window size in half after the loss
what sort of behaviour will result in a aimd (if graphed)
saw tooth behaviour
how is the value of the size of the congestion window kept track of
a variable on each side as it is not in the header
what is tcp slow start
every time an ack is received , the congestion window can be increased exponentially (doubled)
in slow start what is the intial size of the cwnd
1 MSS
what is congestion avoidance in TCP
cwnd initially 1 MSS
exponential increase up to the threshold then linear
upon loss, the cwnd is cut in half + 3 MSS
then begins growing linear again
how is slow start threshold calculated initially
cwnd / 2
