Download TCP and UDP: Reliable and Unreliable Data Transfer Protocols and more Study notes Engineering in PDF only on Docsity!
UDP
connection
less
transport
can have
segmentloss
segments arriving
out
of
order
very
unreliable
no
connection
very
low
delay
but not
reliable
so
it's
good for
zoom
calls
DNS
need
asap
so
uses UDP streaming
multimedia
apps
use it
cuz
they
can
tolerate
loss
UDP
segment
format
size
of
entire
segment
app no
reliable transfer
BUT
ucan
check
if
there's
any
bit
errors
in
the
data
part
of
UDP
the
check
is
called
a
UDP
checksum
checksum
it's
goal
is to
detect
errors flippedbits
in
transmitted
segment
sender takes entire segmentout
including
header
and
lay
the
bits
one
row
mail.L.g.eayggggigsmasmmm.mn
l
l
l l
l
r a
o
I
attitbach
flip
the
bits
receiver
when
that segment
arrives at receiver
side
receiver
does
same
computation
and re
calculates
checksum
which is
then comparedto the
checksum
feild
in
packet
header
if YES
they're
equal
then
no
errors if NO
not
equal
then
error
detected
so
discards
it
btw
theres
some other
errors
checksum
can't
detect
if
checksums
aren't
the
same
UDP
doesn't
resend
it
it
just
discards
it so
packets are
lost
outoford
Principles
of reliabledata
transfer
important in
application transport
linklayers
processes
sitting
at
application
layer
it
demants reliable data
transfer
transport
layer
sending
site 8 recieving
side
eachwill
need
to
implement
extramechanism
to
ensure
reliable data
transfer
in
application
layer mags
aregenerated
andpassed
down
to
transport layer
those
mags
are now
called data application layer
calls
fin
rat
senses
to
pass
the
data
down
to
transport
layer
mags
willbe
processed
in to a
number
of
segments
packets
will
begenerated th
calls
fin
udt
send
pushes it to
network
layy
the
nl
calls
fin
ratrevl to
pass
it
up
Tone
P
p
that
mm
propdelay
4R
Proday
waste
of time low utilization
Pipelined
protocols
to
increase
the
utilization
we
send
multiple packets that
br
und in flight
packets not
yet
acked
stops wait pipelines rtt
Egg
EE
E Eir
two
forms
of
pipelined
protocols GoBack
N
and
selective
repeat
Go
Back
N
allows multiple
unacked packets in pipeline
receiversends
comulative ack's
in
orderback to
bae
doesn't
ack
packet
if
theres
a
gap
has a
timer for
the
oldest
unacked
packet when
timer expires
it
goes
back
in and
resents allll
unackled packets
sender has
a
window
of
up
to
N
consecutive
transmitted
but anack
ed packets
window
slides
after ack
maxy.ge
ptsthatcanbesentatareasametome
timeout
makes
all yellow
packets be
resent
retransmitted
men
iiii.it
35
EE Tug
atsmuar filamentgotartment
addthis.tn
mom
Connection
oriented transport
TCP
TCP
is
a
point
to
point
communication
one
sender
one
receiver
reliable
in
order
byte
stream
data
transmission
we
countdata
by
the
bytes
pipelined
based
rat
window
size is
also for
congestion flow
control flow
control
so
it
doesn't
overwhelm
the
receiver
handshaking to setup
the
initial
states before starting
dataexchange
fullduplexdata
data
is going
bi
directional MSS
max segmentpacketsize
comulative ack's
TOP
segment
structure
0 takes
32
bits
hekewursesintryhihiar.is
mansions
daters
sept
1st first.gg
nt
uket
sequence
numbers
byte
stream
ofthefirst
byte
in thedata
segment
Tig
bit
ftp.yf
i
eesaentes
acknowledgements
willbethe
of
thenext
expected
sequence
comulative
ack
I
example
bi
directional
here
btwiexpectputzanextneresiexpect.su
RDT
mechanism implemented in TCP
howdoes TCP
set the perfect
timeout value
be
the
timeout
is thereto
allow
us to
know
wheathe
a
packet islost or
just
delayed
if
timout istoo
short then
it
sends before delayed
ack
comes
doesn't
give
enough
time
for
delayed
ackto
come
if
too
long
then its a
waste
oftime
doesn't recover
on
time
timer
value
should
be settocoverlongerthan
most
average RTTthat
s8r use
to
give
is
enough
time
for
delay
how
dowe
estimate
RTT we use
a sampleRTT a
pair
of
data
goes
outandtheack
comes back
it
calculates
the
time
difbet revofackand
sending of
data
the
sample
will
vary
it'll average
severalrecent
measurements
not
just
the current sampleRTT
it
ignores
retransmissions
to
make
sure it
covers
only
RTT
the average isto
get
the
bes
smoothestestimation
ofRTT
smoothes
RATnext
smoothetRA
now
currentsample
exponential weighted moving average
ELMA
typical
value of 2 0.
offers
TCP
delayed
ack
Tsents
Lack
rm
TCP
retransmission scenarios
TCP fast
retransmit
TCP
flow
control
Whathappens
if
network
layer
delivers
data
faster
than
application
layer
removes datafrom
socketbuffers
when outgoing is
slow incoming is fastwe
need
flow
control
or
else
the
socket
spacewillbe
exhausted
8
data
willbe
dropped
flow
control receiver
controls sender
usually
to
slow
down
abitso that
sender
won't
overflowreceiver'sbuffer
by
transmitting
toomuch
or
too
fast
the
way
flow
control
works is
by
trackingtherund
in
OS
there's
the
Reubuffer which
is
the size
of
the receive
buffer
space initially
rund
Reubuffer whur TCP
sends
data
ihY
p
pi
Finger
therund is
going
tobe
subtracted
by
the
amount
of
bytesbeing
passed to
it and whur apple
takes in
data
out
ofthe
buffer
rundis
increased
by
these data bytes so
rund
is basically
tracking
thefree buffer
space
TCP
reur will
include rundvalue in
TCPheader and
sends
to
sender
to
know
it
needs
to limit
its
sending
windo
Connection
management
in
TCP
initially
done
before
exchangingdata
using
aprocesses called
a 3
way
handshake
whereafter
a
connection
is
established
a
new
socket
whichthen also
exchanges a
number
of
parameters ex
starting
seq
s sothat
the
server can
decide
weather to
extract data
and
send
up
to
apple
or no
only
tractates
TCP
congestion
control
congestion having
too
many
sources sending
too
much data too
fast
into
the
network
coreof
internet
occurring
in routers
router
buffer
Congestion
window
TCP
sender
limits
transmission
by
assuming rund isso
large
so thelimit
is
just
cund
curd
is
adjusted
in
response to
observed
network congestion
it
dec
if
lost
inc
if
good
TCPsending rate
TCP
initialization
slow
start
when
TCP
connection
is
just
starting
the
initial
cund 1 MSS
max
seg
size
AIMD
increases
the
cund
by
only
1 for
every
ACK
it
receives
soit's
technically doubling cundeach
RTT
is
Congestion
avoidance
the
sender
detects
a packetloss
by
two
ways
1 timout 2
doup ack
if
loss
is
detected
by
timout
curd
is
set
to
MSS
window
then
grows
exponentially
until
cund
slow
start
getsto the
threshold
value
ss
thresh
on
loss
event ss
thresh
is
cut
to
halfofprev
cund
if
loss
is
detected
by
douplicate
ack curdis
cutin half
moreaccuratelyhalf
suss
then
grows
linearly
soup
ack
detected
Orhalftsuss
cuttohalf
agoubling
IMSS
AIMD
Additive Increase
Multiplicative
Decrease
the
sender
will
increase the
sending
rate
gradually
as
long
as
no
loss
defected
but
as
soon
as
there's
packet
loss we
drop sending
rate
