<s>
Transmission	B-Protocol
Control	I-Protocol
Protocol	I-Protocol
(	O
TCP	B-Protocol
)	O
uses	O
a	O
congestion	O
control	O
algorithm	O
that	O
includes	O
various	O
aspects	O
of	O
an	O
additive	O
increase/multiplicative	O
decrease	O
(	O
AIMD	O
)	O
scheme	O
,	O
along	O
with	O
other	O
schemes	O
including	O
slow	O
start	O
and	O
congestion	O
window	O
(	O
CWND	O
)	O
,	O
to	O
achieve	O
congestion	O
avoidance	O
.	O
</s>
<s>
The	O
TCP	B-Protocol
congestion-avoidance	O
algorithm	O
is	O
the	O
primary	O
basis	O
for	O
congestion	O
control	O
in	O
the	O
Internet	O
.	O
</s>
<s>
Per	O
the	O
end-to-end	B-Application
principle	I-Application
,	O
congestion	O
control	O
is	O
largely	O
a	O
function	O
of	O
internet	B-Device
hosts	I-Device
,	O
not	O
the	O
network	O
itself	O
.	O
</s>
<s>
There	O
are	O
several	O
variations	O
and	O
versions	O
of	O
the	O
algorithm	O
implemented	O
in	O
protocol	B-Protocol
stacks	I-Protocol
of	O
operating	B-General_Concept
systems	I-General_Concept
of	O
computers	O
that	O
connect	O
to	O
the	O
Internet	O
.	O
</s>
<s>
To	O
avoid	O
congestive	O
collapse	O
,	O
TCP	B-Protocol
uses	O
multi-faceted	O
congestion-control	O
strategy	O
.	O
</s>
<s>
For	O
each	O
connection	O
,	O
TCP	B-Protocol
maintains	O
a	O
CWND	O
,	O
limiting	O
the	O
total	O
number	O
of	O
unacknowledged	O
packets	O
that	O
may	O
be	O
in	O
transit	O
end-to-end	O
.	O
</s>
<s>
This	O
is	O
somewhat	O
analogous	O
to	O
TCP	B-Protocol
's	O
sliding	O
window	O
used	O
for	O
flow	O
control	O
.	O
</s>
<s>
In	O
TCP	B-Protocol
,	O
the	O
congestion	O
window	O
(	O
CWND	O
)	O
is	O
one	O
of	O
the	O
factors	O
that	O
determines	O
the	O
number	O
of	O
bytes	O
that	O
can	O
be	O
sent	O
out	O
at	O
any	O
time	O
.	O
</s>
<s>
When	O
a	O
connection	O
is	O
set	O
up	O
,	O
the	O
congestion	O
window	O
,	O
a	O
value	O
maintained	O
independently	O
at	O
each	O
host	O
,	O
is	O
set	O
to	O
a	O
small	O
multiple	O
of	O
the	O
maximum	B-Protocol
segment	I-Protocol
size	I-Protocol
(	O
MSS	O
)	O
allowed	O
on	O
that	O
connection	O
.	O
</s>
<s>
This	O
means	O
that	O
if	O
all	O
segments	O
are	O
received	O
and	O
the	O
acknowledgments	B-Protocol
reach	O
the	O
sender	O
on	O
time	O
,	O
some	O
constant	O
is	O
added	O
to	O
the	O
window	O
size	O
.	O
</s>
<s>
A	O
system	B-Application
administrator	I-Application
may	O
adjust	O
the	O
maximum	O
window	O
size	O
limit	O
,	O
or	O
adjust	O
the	O
constant	O
added	O
during	O
additive	O
increase	O
,	O
as	O
part	O
of	O
TCP	B-Protocol
tuning	I-Protocol
.	O
</s>
<s>
The	O
flow	O
of	O
data	O
over	O
a	O
TCP	B-Protocol
connection	O
is	O
also	O
controlled	O
by	O
the	O
use	O
of	O
the	O
receive	O
window	O
advertised	O
by	O
the	O
receiver	O
.	O
</s>
<s>
is	O
part	O
of	O
the	O
congestion	O
control	O
strategy	O
used	O
by	O
TCP	B-Protocol
in	O
conjunction	O
with	O
other	O
algorithms	O
to	O
avoid	O
sending	O
more	O
data	O
than	O
the	O
network	O
is	O
capable	O
of	O
forwarding	O
,	O
that	O
is	O
,	O
to	O
avoid	O
causing	O
network	O
congestion	O
.	O
</s>
<s>
The	O
value	O
for	O
the	O
congestion	O
window	O
size	O
can	O
be	O
increased	O
by	O
1	O
MSS	O
with	O
each	O
acknowledgement	B-Protocol
(	O
ACK	B-Protocol
)	O
received	O
,	O
effectively	O
doubling	O
the	O
window	O
size	O
each	O
RTT	O
.	O
</s>
<s>
If	O
the	O
CWND	O
reaches	O
ssthresh	O
,	O
TCP	B-Protocol
changes	O
to	O
congestion	O
avoidance	O
algorithm	O
.	O
</s>
<s>
A	O
common	O
formula	O
is	O
that	O
each	O
new	O
ACK	B-Protocol
increases	O
the	O
CWND	O
by	O
MSS	O
*	O
It	O
increases	O
almost	O
linearly	O
and	O
provides	O
an	O
acceptable	O
approximation	O
.	O
</s>
<s>
If	O
a	O
loss	O
event	O
occurs	O
,	O
TCP	B-Protocol
assumes	O
that	O
it	O
is	O
due	O
to	O
network	O
congestion	O
and	O
takes	O
steps	O
to	O
reduce	O
the	O
offered	O
load	O
on	O
the	O
network	O
.	O
</s>
<s>
These	O
measurements	O
depend	O
on	O
the	O
exact	O
TCP	B-Protocol
congestion	O
avoidance	O
algorithm	O
used	O
.	O
</s>
<s>
When	O
a	O
TCP	B-Protocol
sender	O
detects	O
segment	O
loss	O
using	O
the	O
retransmission	O
timer	O
and	O
the	O
given	O
segment	O
has	O
not	O
yet	O
been	O
resent	O
by	O
way	O
of	O
the	O
retransmission	O
timer	O
,	O
the	O
value	O
of	O
ssthresh	O
must	O
be	O
set	O
to	O
no	O
more	O
than	O
half	O
of	O
the	O
amount	O
of	O
data	O
that	O
has	O
been	O
sent	O
but	O
not	O
yet	O
cumulatively	O
acknowledged	O
.	O
</s>
<s>
A	O
fast	B-Protocol
retransmit	I-Protocol
is	O
sent	O
,	O
half	O
of	O
the	O
current	O
CWND	O
is	O
saved	O
as	O
ssthresh	O
and	O
as	O
new	O
CWND	O
,	O
thus	O
skipping	O
slow	O
start	O
and	O
going	O
directly	O
to	O
the	O
congestion	O
avoidance	O
algorithm	O
.	O
</s>
<s>
While	O
this	O
is	O
an	O
acceptable	O
assumption	O
for	O
many	O
networks	O
,	O
segments	O
may	O
be	O
lost	O
for	O
other	O
reasons	O
,	O
such	O
as	O
poor	O
data	B-Protocol
link	I-Protocol
layer	I-Protocol
transmission	O
quality	O
.	O
</s>
<s>
Thus	O
,	O
slow	O
start	O
can	O
perform	O
poorly	O
in	O
situations	O
with	O
poor	O
reception	O
,	O
such	O
as	O
wireless	B-General_Concept
networks	I-General_Concept
.	O
</s>
<s>
Older	O
web	B-Application
browsers	I-Application
would	O
create	O
many	O
consecutive	O
short-lived	O
connections	O
to	O
the	O
web	O
server	O
,	O
and	O
would	O
open	O
and	O
close	O
the	O
connection	O
for	O
each	O
file	O
requested	O
.	O
</s>
<s>
To	O
avoid	O
this	O
problem	O
,	O
modern	O
browsers	B-Application
either	O
open	O
multiple	O
connections	O
simultaneously	O
or	O
reuse	B-Protocol
one	I-Protocol
connection	I-Protocol
for	O
all	O
files	O
requested	O
from	O
a	O
particular	O
web	O
server	O
.	O
</s>
<s>
Fast	B-Protocol
retransmit	I-Protocol
is	O
an	O
enhancement	O
to	O
TCP	B-Protocol
that	O
reduces	O
the	O
time	O
a	O
sender	O
waits	O
before	O
retransmitting	O
a	O
lost	O
segment	O
.	O
</s>
<s>
A	O
TCP	B-Protocol
sender	O
normally	O
uses	O
a	O
simple	O
timer	O
to	O
recognize	O
lost	O
segments	O
.	O
</s>
<s>
If	O
an	O
acknowledgement	B-Protocol
is	O
not	O
received	O
for	O
a	O
particular	O
segment	O
within	O
a	O
specified	O
time	O
(	O
a	O
function	O
of	O
the	O
estimated	O
round-trip	O
delay	O
time	O
)	O
,	O
the	O
sender	O
will	O
assume	O
the	O
segment	O
was	O
lost	O
in	O
the	O
network	O
,	O
and	O
will	O
retransmit	O
the	O
segment	O
.	O
</s>
<s>
Duplicate	O
acknowledgement	B-Protocol
is	O
the	O
basis	O
for	O
the	O
fast	B-Protocol
retransmit	I-Protocol
mechanism	O
.	O
</s>
<s>
After	O
receiving	O
a	O
packet	O
an	O
acknowledgement	B-Protocol
is	O
sent	O
for	O
the	O
last	O
in-order	O
byte	O
of	O
data	O
received	O
.	O
</s>
<s>
For	O
an	O
in-order	O
packet	O
,	O
this	O
is	O
effectively	O
the	O
last	O
packet	O
's	O
sequence	B-Protocol
number	I-Protocol
plus	O
the	O
current	O
packet	O
's	O
payload	O
length	O
.	O
</s>
<s>
Thus	O
a	O
Duplicate	O
acknowledgement	B-Protocol
occurs	O
.	O
</s>
<s>
Duplicate	O
acknowledgements	B-Protocol
are	O
sent	O
for	O
both	O
of	O
these	O
packets	O
.	O
</s>
<s>
When	O
a	O
sender	O
receives	O
three	O
duplicate	O
acknowledgements	B-Protocol
,	O
it	O
can	O
be	O
reasonably	O
confident	O
that	O
the	O
segment	O
carrying	O
the	O
data	O
that	O
followed	O
the	O
last	O
in-order	O
byte	O
specified	O
in	O
the	O
acknowledgment	B-Protocol
was	O
lost	O
.	O
</s>
<s>
A	O
sender	O
with	O
fast	B-Protocol
retransmit	I-Protocol
will	O
then	O
retransmit	O
this	O
packet	O
immediately	O
without	O
waiting	O
for	O
its	O
timeout	O
.	O
</s>
<s>
There	O
is	O
no	O
need	O
to	O
acknowledge	O
intermediate	O
packets	O
,	O
since	O
TCP	B-Protocol
uses	O
cumulative	O
acknowledgements	B-Protocol
by	O
default	O
.	O
</s>
<s>
TCP	B-Protocol
Tahoe	O
and	O
Reno	O
algorithms	O
were	O
retrospectively	O
named	O
after	O
the	O
versions	O
or	O
flavours	O
of	O
the	O
4.3BSD	B-Operating_System
operating	B-General_Concept
system	I-General_Concept
in	O
which	O
each	O
first	O
appeared	O
(	O
which	O
were	O
themselves	O
named	O
after	O
Lake	O
Tahoe	O
and	O
the	O
nearby	O
city	O
of	O
Reno	O
,	O
Nevada	O
)	O
.	O
</s>
<s>
The	O
Tahoe	O
algorithm	O
first	O
appeared	O
in	O
4.3BSD-Tahoe	O
(	O
which	O
was	O
made	O
to	O
support	O
the	O
CCI	O
Power	O
6/32	O
"	O
Tahoe	O
"	O
minicomputer	O
)	O
,	O
and	O
was	O
later	O
made	O
available	O
to	O
non-AT	O
&T	O
licensees	O
as	O
part	O
of	O
the	O
4.3BSD	B-Operating_System
Networking	O
Release	O
1	O
;	O
this	O
ensured	O
its	O
wide	O
distribution	O
and	O
implementation	O
.	O
</s>
<s>
Improvements	O
were	O
made	O
in	O
4.3BSD-Reno	O
and	O
subsequently	O
released	O
to	O
the	O
public	O
as	O
Networking	O
Release	O
2	O
and	O
later	O
4.4BSD-Lite	O
.	O
</s>
<s>
While	O
both	O
consider	O
retransmission	O
timeout	O
(	O
RTO	O
)	O
and	O
duplicate	O
ACKs	B-Protocol
as	O
packet	O
loss	O
events	O
,	O
the	O
behavior	O
of	O
Tahoe	O
and	O
Reno	O
differ	O
primarily	O
in	O
how	O
they	O
react	O
to	O
duplicate	O
ACKs	B-Protocol
:	O
</s>
<s>
Tahoe	O
:	O
if	O
three	O
duplicate	O
ACKs	B-Protocol
are	O
received	O
(	O
i.e.	O
</s>
<s>
four	O
ACKs	B-Protocol
acknowledging	O
the	O
same	O
packet	O
,	O
which	O
are	O
not	O
piggybacked	O
on	O
data	O
and	O
do	O
not	O
change	O
the	O
receiver	O
's	O
advertised	O
window	O
)	O
,	O
Tahoe	O
performs	O
a	O
fast	B-Protocol
retransmit	I-Protocol
,	O
sets	O
the	O
slow	O
start	O
threshold	O
to	O
half	O
of	O
the	O
current	O
congestion	O
window	O
,	O
reduces	O
the	O
congestion	O
window	O
to	O
1	O
MSS	O
,	O
and	O
resets	O
to	O
slow	O
start	O
state	O
.	O
</s>
<s>
Reno	O
:	O
if	O
three	O
duplicate	O
ACKs	B-Protocol
are	O
received	O
,	O
Reno	O
will	O
perform	O
a	O
fast	B-Protocol
retransmit	I-Protocol
and	O
skip	O
the	O
slow	O
start	O
phase	O
by	O
instead	O
halving	O
the	O
congestion	O
window	O
(	O
instead	O
of	O
setting	O
it	O
to	O
1	O
MSS	O
like	O
Tahoe	O
)	O
,	O
setting	O
the	O
ssthresh	O
equal	O
to	O
the	O
new	O
congestion	O
window	O
,	O
and	O
enter	O
a	O
phase	O
called	O
fast	O
recovery	O
.	O
</s>
<s>
In	O
both	O
Tahoe	O
and	O
Reno	O
,	O
if	O
an	O
ACK	B-Protocol
times	O
out	O
(	O
RTO	O
timeout	O
)	O
,	O
slow	O
start	O
is	O
used	O
,	O
and	O
both	O
algorithms	O
reduce	O
congestion	O
window	O
to	O
1	O
MSS	O
.	O
</s>
<s>
TCP	B-Protocol
New	O
Reno	O
,	O
defined	O
by	O
(	O
which	O
obsolesces	O
previous	O
definitions	O
in	O
and	O
)	O
,	O
improves	O
retransmission	O
during	O
the	O
fast-recovery	O
phase	O
of	O
TCP	B-Protocol
Reno	O
.	O
</s>
<s>
During	O
fast	O
recovery	O
,	O
to	O
keep	O
the	O
transmit	O
window	O
full	O
,	O
for	O
every	O
duplicate	O
ACK	B-Protocol
that	O
is	O
returned	O
,	O
a	O
new	O
unsent	O
packet	O
from	O
the	O
end	O
of	O
the	O
congestion	O
window	O
is	O
sent	O
.	O
</s>
<s>
Partial	O
acknowledgments	B-Protocol
:	O
The	O
ACK	B-Protocol
does	O
not	O
acknowledge	O
all	O
data	O
.	O
</s>
<s>
After	O
a	O
retransmit	O
timeout	O
,	O
it	O
records	O
the	O
highest	O
sequence	B-Protocol
number	I-Protocol
transmitted	O
in	O
the	O
recover	O
variable	O
and	O
exits	O
the	O
fast	O
recovery	O
procedure	O
.	O
</s>
<s>
If	O
this	O
sequence	B-Protocol
number	I-Protocol
is	O
acknowledged	O
,	O
TCP	B-Protocol
returns	O
to	O
the	O
congestion	O
avoidance	O
state	O
.	O
</s>
<s>
A	O
problem	O
occurs	O
with	O
New	O
Reno	O
when	O
there	O
are	O
no	O
packet	O
losses	O
but	O
instead	O
,	O
packets	O
are	O
reordered	O
by	O
more	O
than	O
3	O
packet	O
sequence	B-Protocol
numbers	I-Protocol
.	O
</s>
<s>
Until	O
the	O
mid-1990s	O
,	O
all	O
of	O
TCP	B-Protocol
's	O
set	O
timeouts	O
and	O
measured	O
round-trip	O
delays	O
were	O
based	O
upon	O
only	O
the	O
last	O
transmitted	O
packet	O
in	O
the	O
transmit	O
buffer	O
.	O
</s>
<s>
University	O
of	O
Arizona	O
researchers	O
Larry	O
Peterson	O
and	O
Lawrence	O
Brakmo	O
introduced	O
TCP	B-Protocol
Vegas	I-Protocol
in	O
which	O
timeouts	O
were	O
set	O
and	O
round-trip	O
delays	O
were	O
measured	O
for	O
every	O
packet	O
in	O
the	O
transmit	O
buffer	O
.	O
</s>
<s>
In	O
addition	O
,	O
TCP	B-Protocol
Vegas	I-Protocol
uses	O
additive	O
increases	O
in	O
the	O
congestion	O
window	O
.	O
</s>
<s>
In	O
a	O
comparison	O
study	O
of	O
various	O
TCP	B-Protocol
s	O
,	O
TCP	B-Protocol
Vegas	I-Protocol
appeared	O
to	O
be	O
the	O
smoothest	O
followed	O
by	O
TCP	B-Protocol
CUBIC	I-Protocol
.	O
</s>
<s>
TCP	B-Protocol
Vegas	I-Protocol
was	O
not	O
widely	O
deployed	O
outside	O
Peterson	O
's	O
laboratory	O
but	O
was	O
selected	O
as	O
the	O
default	O
congestion	O
control	O
method	O
for	O
DD-WRT	B-Protocol
firmware	O
v24	O
SP2	O
.	O
</s>
<s>
TCP	B-Protocol
Hybla	O
aims	O
to	O
eliminate	O
penalties	O
to	O
TCP	B-Protocol
connections	O
that	O
incorporate	O
a	O
high-latency	O
terrestrial	O
or	O
satellite	O
radio	O
links	O
.	O
</s>
<s>
Binary	B-Protocol
Increase	I-Protocol
Congestion	I-Protocol
control	I-Protocol
(	O
BIC	B-Protocol
)	O
is	O
a	O
TCP	B-Protocol
implementation	O
with	O
an	O
optimized	O
CCA	O
for	O
high	O
speed	O
networks	O
with	O
high	O
latency	O
,	O
known	O
as	O
long	O
fat	O
networks	O
(	O
LFNs	O
)	O
.	O
</s>
<s>
BIC	B-Protocol
is	O
used	O
by	O
default	O
in	O
Linux	B-Operating_System
kernels	I-Operating_System
2.6.8	O
through	O
2.6.18	O
.	O
</s>
<s>
CUBIC	B-Protocol
is	O
a	O
less	O
aggressive	O
and	O
more	O
systematic	O
derivative	O
of	O
BIC	B-Protocol
,	O
in	O
which	O
the	O
window	O
is	O
a	O
cubic	B-Protocol
function	O
of	O
time	O
since	O
the	O
last	O
congestion	O
event	O
,	O
with	O
the	O
inflection	O
point	O
set	O
to	O
the	O
window	O
prior	O
to	O
the	O
event	O
.	O
</s>
<s>
CUBIC	B-Protocol
is	O
used	O
by	O
default	O
in	O
Linux	B-Operating_System
kernels	I-Operating_System
since	O
version	O
2.6.19	O
.	O
</s>
<s>
Agile-SD	O
is	O
a	O
Linux-based	O
CCA	O
which	O
is	O
designed	O
for	O
the	O
real	O
Linux	B-Operating_System
kernel	I-Operating_System
.	O
</s>
<s>
It	O
has	O
been	O
evaluated	O
by	O
comparing	O
its	O
performance	O
to	O
Compound	O
TCP	B-Protocol
(	O
the	O
default	O
CCA	O
in	O
MS	O
Windows	O
)	O
and	O
CUBIC	B-Protocol
(	O
the	O
default	O
of	O
Linux	B-Operating_System
)	O
using	O
NS-2	O
simulator	O
.	O
</s>
<s>
Westwood+	O
is	O
a	O
sender-only	O
modification	O
of	O
TCP	B-Protocol
Reno	O
that	O
optimizes	O
the	O
performance	O
of	O
TCP	B-Protocol
congestion	O
control	O
over	O
both	O
wired	O
and	O
wireless	B-General_Concept
networks	I-General_Concept
.	O
</s>
<s>
TCP	O
Westwood+	O
is	O
based	O
on	O
end-to-end	O
bandwidth	O
estimation	O
to	O
set	O
the	O
congestion	O
window	O
and	O
slow-start	O
threshold	O
after	O
a	O
congestion	O
episode	O
,	O
that	O
is	O
,	O
after	O
three	O
duplicate	O
acknowledgments	B-Protocol
or	O
a	O
timeout	O
.	O
</s>
<s>
The	O
bandwidth	O
is	O
estimated	O
by	O
averaging	O
the	O
rate	O
of	O
returning	O
acknowledgment	B-Protocol
packets	O
.	O
</s>
<s>
In	O
contrast	O
with	O
TCP	B-Protocol
Reno	O
,	O
which	O
blindly	O
halves	O
the	O
congestion	O
window	O
after	O
three	O
duplicate	O
ACKs	B-Protocol
,	O
TCP	B-Protocol
Westwood+	O
adaptively	O
sets	O
a	O
slow-start	O
threshold	O
and	O
a	O
congestion	O
window	O
which	O
takes	O
into	O
account	O
an	O
estimate	O
of	O
bandwidth	O
available	O
at	O
the	O
time	O
congestion	O
is	O
experienced	O
.	O
</s>
<s>
Compared	O
to	O
Reno	O
and	O
New	O
Reno	O
,	O
Westwood+	O
significantly	O
increases	O
throughput	O
over	O
wireless	O
links	O
and	O
improves	O
fairness	B-Protocol
in	O
wired	O
networks	O
.	O
</s>
<s>
Compound	O
TCP	B-Protocol
is	O
a	O
Microsoft	O
implementation	O
of	O
TCP	B-Protocol
which	O
maintains	O
two	O
different	O
congestion	O
windows	O
simultaneously	O
,	O
with	O
the	O
goal	O
of	O
achieving	O
good	O
performance	O
on	O
LFNs	O
while	O
not	O
impairing	O
fairness	B-Protocol
.	O
</s>
<s>
It	O
has	O
been	O
widely	O
deployed	O
in	O
Windows	O
versions	O
since	O
Microsoft	B-Application
Windows	I-Application
Vista	I-Application
and	O
Windows	B-Device
Server	I-Device
2008	I-Device
and	O
has	O
been	O
ported	O
to	O
older	O
Microsoft	O
Windows	O
versions	O
as	O
well	O
as	O
Linux	B-Operating_System
.	O
</s>
<s>
TCP	B-Protocol
Proportional	O
Rate	O
Reduction	O
(	O
PRR	O
)	O
is	O
an	O
algorithm	O
designed	O
to	O
improve	O
the	O
accuracy	O
of	O
data	O
sent	O
during	O
recovery	O
.	O
</s>
<s>
In	O
tests	O
performed	O
by	O
Google	B-Application
,	O
PRR	O
resulted	O
in	O
a	O
3	O
–	O
10%	O
reduction	O
in	O
average	O
latency	O
and	O
recovery	O
timeouts	O
were	O
reduced	O
by	O
5%	O
.	O
</s>
<s>
PRR	O
is	O
available	O
in	O
Linux	B-Operating_System
kernels	I-Operating_System
since	O
version	O
3.2	O
.	O
</s>
<s>
Bottleneck	O
Bandwidth	O
and	O
Round-trip	O
propagation	O
time	O
(	O
BBR	O
)	O
is	O
a	O
CCA	O
developed	O
at	O
Google	B-Application
in	O
2016	O
.	O
</s>
<s>
While	O
most	O
CCAs	O
are	O
loss-based	O
,	O
in	O
that	O
they	O
rely	O
on	O
packet	O
loss	O
to	O
detect	O
congestion	O
and	O
lower	O
rates	O
of	O
transmission	O
,	O
BBR	O
,	O
like	O
TCP	B-Protocol
Vegas	I-Protocol
,	O
is	O
model-based	O
.	O
</s>
<s>
Each	O
cumulative	O
or	O
selective	O
acknowledgment	B-Protocol
of	O
packet	O
delivery	O
produces	O
a	O
rate	O
sample	O
which	O
records	O
the	O
amount	O
of	O
data	O
delivered	O
over	O
the	O
time	O
interval	O
between	O
the	O
transmission	O
of	O
a	O
data	O
packet	O
and	O
the	O
acknowledgment	B-Protocol
of	O
that	O
packet	O
.	O
</s>
<s>
As	O
network	O
interface	O
controllers	O
evolve	O
from	O
megabit	O
per	O
second	O
to	O
gigabit	O
per	O
second	O
performance	O
,	O
the	O
latency	O
associated	O
with	O
bufferbloat	O
instead	O
of	O
packet	O
loss	O
becomes	O
a	O
more	O
reliable	O
marker	O
of	O
the	O
maximum	O
throughput	O
,	O
making	O
model-based	O
CCAs	O
which	O
provide	O
higher	O
throughput	O
and	O
lower	O
latency	O
,	O
such	O
as	O
BBR	O
,	O
a	O
more	O
reliable	O
alternative	O
to	O
more	O
popular	O
loss-based	O
algorithms	O
like	O
TCP	B-Protocol
CUBIC	I-Protocol
.	O
</s>
<s>
BBR	O
has	O
been	O
available	O
for	O
Linux	B-Operating_System
TCP	B-Protocol
since	O
Linux	B-Operating_System
4.9	O
.	O
</s>
<s>
It	O
is	O
also	O
available	O
for	O
QUIC	B-Protocol
.	O
</s>
<s>
BBR	O
version	O
1	O
(	O
BBRv1	O
)	O
is	O
efficient	O
and	O
fast	O
,	O
but	O
its	O
fairness	B-Protocol
to	O
non-BBR	O
streams	O
is	O
disputed	O
.	O
</s>
<s>
When	O
implemented	O
at	O
YouTube	B-General_Concept
,	O
BBRv1	O
yielded	O
an	O
average	O
of	O
4%	O
higher	O
network	O
throughput	O
and	O
up	O
to	O
14%	O
in	O
some	O
countries	O
.	O
</s>
<s>
While	O
Google	B-Application
's	I-Application
presentation	O
shows	O
BBRv1	O
co-existing	O
well	O
with	O
CUBIC	B-Protocol
,	O
researchers	O
like	O
Geoff	O
Huston	O
and	O
Hock	O
,	O
Bless	O
and	O
Zitterbart	O
finds	O
it	O
unfair	O
to	O
other	O
streams	O
and	O
not	O
scalable	O
.	O
</s>
<s>
also	O
found	O
"	O
some	O
severe	O
inherent	O
issues	O
such	O
as	O
increased	O
queuing	O
delays	O
,	O
unfairness	O
,	O
and	O
massive	O
packet	O
loss	O
"	O
in	O
the	O
BBR	O
implementation	O
of	O
Linux	B-Operating_System
4.9	O
.	O
</s>
<s>
For	O
instance	O
,	O
when	O
a	O
CUBIC	B-Protocol
flow	O
(	O
which	O
is	O
the	O
default	O
TCP	B-Protocol
implementation	O
in	O
Linux	B-Operating_System
,	O
Android	O
,	O
and	O
MacOS	O
)	O
coexists	O
with	O
a	O
BBR	O
flow	O
in	O
the	O
network	O
,	O
the	O
BBR	O
flow	O
can	O
dominate	O
the	O
CUBIC	B-Protocol
flow	O
and	O
get	O
the	O
whole	O
link	O
bandwidth	O
from	O
it	O
(	O
see	O
figure	O
16	O
in	O
)	O
.	O
</s>
<s>
Version	O
2	O
attempts	O
to	O
deal	O
with	O
the	O
issue	O
of	O
unfairness	O
when	O
operating	O
alongside	O
loss-based	O
congestion	O
management	O
such	O
as	O
CUBIC	B-Protocol
.	O
</s>
<s>
Cellular	O
Controlled	O
Delay	O
TCP	B-Protocol
(	O
C2TCP	O
)	O
was	O
motivated	O
by	O
the	O
lack	O
of	O
a	O
flexible	O
end-to-end	O
TCP	B-Protocol
approach	O
that	O
can	O
satisfy	O
various	O
QoS	O
requirements	O
for	O
different	O
applications	O
without	O
requiring	O
any	O
changes	O
in	O
the	O
network	O
devices	O
.	O
</s>
<s>
C2TCP	O
aims	O
to	O
satisfy	O
ultra-low	O
latency	O
and	O
high-bandwidth	O
requirements	O
of	O
applications	O
such	O
as	O
virtual	B-Application
reality	I-Application
,	O
video	B-Application
conferencing	I-Application
,	O
online	B-Application
gaming	I-Application
,	O
vehicular	O
communication	O
systems	O
,	O
etc	O
.	O
</s>
<s>
C2TCP	O
works	O
as	O
an	O
add-on	B-Application
on	O
top	O
of	O
loss-based	O
TCP	B-Protocol
(	O
e.g.	O
</s>
<s>
Reno	O
,	O
NewReno	O
,	O
CUBIC	B-Protocol
,	O
BIC	B-Protocol
,	O
...	O
)	O
,	O
it	O
is	O
only	O
required	O
to	O
be	O
installed	O
on	O
the	O
server-side	O
and	O
makes	O
the	O
average	O
delay	O
of	O
packets	O
bounded	O
to	O
the	O
desired	O
delays	O
set	O
by	O
the	O
applications	O
.	O
</s>
<s>
Researchers	O
at	O
NYU	O
showed	O
that	O
C2TCP	O
outperforms	O
the	O
delay	O
and	O
delay-variation	O
performance	O
of	O
various	O
state-of-the-art	O
TCP	B-Protocol
schemes	O
.	O
</s>
<s>
For	O
instance	O
,	O
they	O
showed	O
that	O
compared	O
to	O
BBR	O
,	O
CUBIC	B-Protocol
,	O
and	O
Westwood	B-Protocol
on	O
average	O
,	O
C2TCP	O
decreases	O
the	O
average	O
delay	O
of	O
packets	O
by	O
about	O
250%	O
,	O
900%	O
,	O
and	O
700%	O
respectively	O
on	O
various	O
cellular	O
network	O
environments	O
.	O
</s>
<s>
Elastic-TCP	O
was	O
proposed	O
in	O
February	O
2019	O
to	O
increase	O
the	O
bandwidth	O
utilization	O
over	O
high-BDP	O
networks	O
in	O
support	O
of	O
cloud	O
computing	O
.	O
</s>
<s>
It	O
is	O
a	O
Linux-based	O
CCA	O
that	O
is	O
designed	O
for	O
the	O
Linux	B-Operating_System
kernel	I-Operating_System
.	O
</s>
<s>
It	O
has	O
been	O
evaluated	O
by	O
comparing	O
its	O
performance	O
to	O
Compound	O
TCP	B-Protocol
(	O
the	O
default	O
CCA	O
in	O
MS	O
Windows	O
)	O
,	O
CUBIC	B-Protocol
(	O
the	O
default	O
for	O
Linux	B-Operating_System
)	O
and	O
TCP-BBR	O
(	O
the	O
default	O
of	O
Linux	B-Operating_System
4.9	O
used	O
by	O
Google	B-Application
)	O
using	O
the	O
NS-2	O
simulator	O
and	O
testbed	O
.	O
</s>
<s>
Elastic-TCP	O
significantly	O
improves	O
the	O
total	O
performance	O
in	O
terms	O
of	O
average	O
throughput	O
,	O
loss	O
ratio	O
,	O
and	O
delay	O
.	O
</s>
<s>
proposed	O
NATCP	O
(	O
Network-Assisted	O
TCP	B-Protocol
)	O
a	O
TCP	B-Protocol
design	O
targeting	O
multi-access	O
edge	O
computing	O
(	O
MEC	O
)	O
.	O
</s>
<s>
The	O
key	O
idea	O
of	O
NATCP	O
is	O
that	O
if	O
the	O
characteristics	O
of	O
the	O
network	O
were	O
known	O
beforehand	O
,	O
TCP	B-Protocol
would	O
have	O
been	O
designed	O
differently	O
.	O
</s>
<s>
Therefore	O
,	O
NATCP	O
employs	O
the	O
available	O
features	O
and	O
properties	O
in	O
the	O
current	O
MEC-based	O
cellular	O
architectures	O
to	O
push	O
the	O
performance	O
of	O
TCP	B-Protocol
close	O
to	O
the	O
optimal	O
performance	O
.	O
</s>
<s>
As	O
preliminary	O
results	O
show	O
,	O
NATCP	O
outperforms	O
the	O
state-of-the-art	O
TCP	B-Protocol
schemes	O
.	O
</s>
<s>
TCP	B-Protocol
New	O
Reno	O
was	O
the	O
most	O
commonly	O
implemented	O
algorithm	O
,	O
SACK	O
support	O
is	O
very	O
common	O
and	O
is	O
an	O
extension	O
to	O
Reno/New	O
Reno	O
.	O
</s>
<s>
Starting	O
with	O
2.6.8	O
the	O
Linux	B-Operating_System
kernel	I-Operating_System
switched	O
the	O
default	O
implementation	O
from	O
New	O
Reno	O
to	O
BIC	B-Protocol
.	O
</s>
<s>
The	O
default	O
implementation	O
was	O
again	O
changed	O
to	O
CUBIC	B-Protocol
in	O
the	O
2.6.19	O
version	O
.	O
</s>
<s>
FreeBSD	B-Operating_System
uses	O
New	O
Reno	O
as	O
the	O
default	O
algorithm	O
.	O
</s>
<s>
When	O
the	O
per-flow	O
product	O
of	O
bandwidth	O
and	O
latency	O
increases	O
,	O
regardless	O
of	O
the	O
queuing	O
scheme	O
,	O
TCP	B-Protocol
becomes	O
inefficient	O
and	O
prone	O
to	O
instability	O
.	O
</s>
<s>
TCP	B-Protocol
Interactive	O
(	O
iTCP	O
)	O
allows	O
applications	O
to	O
subscribe	O
to	O
TCP	B-Protocol
events	O
and	O
respond	O
accordingly	O
enabling	O
various	O
functional	O
extensions	O
to	O
TCP	B-Protocol
from	O
outside	O
TCP	B-Protocol
layer	O
.	O
</s>
<s>
Most	O
TCP	B-Protocol
congestion	O
schemes	O
work	O
internally	O
.	O
</s>
<s>
Zeta-TCP	B-Protocol
detects	O
congestion	O
from	O
both	O
latency	O
and	O
loss	O
rate	O
measures	O
.	O
</s>
<s>
To	O
maximize	O
the	O
goodput	O
Zeta-TCP	B-Protocol
and	O
applies	O
different	O
congestion	O
window	O
backoff	O
strategies	O
based	O
on	O
the	O
likelihood	O
of	O
congestion	O
.	O
</s>
<s>
BIC	B-Protocol
TCP	I-Protocol
(	O
Binary	B-Protocol
Increase	I-Protocol
Congestion	I-Protocol
Control	I-Protocol
Protocol	O
)	O
uses	O
a	O
concave	O
increase	O
of	O
the	O
sources	O
rate	O
after	O
each	O
congestion	O
event	O
until	O
the	O
window	O
is	O
equal	O
to	O
that	O
before	O
the	O
event	O
,	O
in	O
order	O
to	O
maximise	O
the	O
time	O
that	O
the	O
network	O
is	O
fully	O
utilised	O
.	O
</s>
<s>
CUBIC	B-Protocol
TCP	I-Protocol
–	O
a	O
less	O
aggressive	O
and	O
more	O
systematic	O
derivative	O
of	O
BIC	B-Protocol
,	O
in	O
which	O
the	O
window	O
is	O
a	O
cubic	B-Protocol
function	O
of	O
time	O
since	O
the	O
last	O
congestion	O
event	O
,	O
with	O
the	O
inflection	O
point	O
set	O
to	O
the	O
window	O
prior	O
to	O
the	O
event	O
.	O
</s>
<s>
TCP	B-Protocol
Vegas	I-Protocol
–	O
estimates	O
the	O
queuing	O
delay	O
,	O
and	O
linearly	O
increases	O
or	O
decreases	O
the	O
window	O
so	O
that	O
a	O
constant	O
number	O
of	O
packets	O
per	O
flow	O
are	O
queued	O
in	O
the	O
network	O
.	O
</s>
<s>
Vegas	O
implements	O
proportional	O
fairness	B-Protocol
.	O
</s>
<s>
FAST	B-Protocol
TCP	I-Protocol
–	O
achieves	O
the	O
same	O
equilibrium	O
as	O
Vegas	O
,	O
but	O
uses	O
proportional	O
control	O
instead	O
of	O
linear	O
increase	O
,	O
and	O
intentionally	O
scales	O
the	O
gain	O
down	O
as	O
the	O
bandwidth	O
increases	O
with	O
the	O
aim	O
of	O
ensuring	O
stability	O
.	O
</s>
<s>
TCP	B-Protocol
BBR	O
–	O
estimates	O
the	O
queuing	O
delay	O
but	O
uses	O
exponential	O
increase	O
.	O
</s>
<s>
Intentionally	O
slows	O
down	O
periodically	O
for	O
fairness	B-Protocol
and	O
decreased	O
delay	O
.	O
</s>
<s>
TCP-Westwood	B-Protocol
(	O
TCPW	B-Protocol
)	O
–	O
a	O
loss	O
causes	O
the	O
window	O
to	O
be	O
reset	O
to	O
the	O
sender	O
's	O
estimate	O
of	O
the	O
bandwidth-delay	O
product	O
(	O
the	O
smallest	O
measured	O
RTT	O
multiplied	O
by	O
the	O
observed	O
rate	O
of	O
receiving	O
ACKs	B-Protocol
)	O
.	O
</s>
<s>
Random	B-Algorithm
Early	I-Algorithm
Detection	I-Algorithm
(	O
RED	O
)	O
randomly	O
drops	O
packets	O
in	O
proportion	O
to	O
the	O
router	O
's	O
queue	O
size	O
,	O
triggering	O
multiplicative	O
decrease	O
in	O
some	O
flows	O
.	O
</s>
<s>
NATCP	O
-	O
Network-Assisted	O
TCP	B-Protocol
uses	O
out-of-band	O
explicit	O
feedback	O
indicating	O
minimum	O
RTT	O
of	O
the	O
network	O
and	O
capacity	O
of	O
the	O
cellular	O
access	O
link	O
.	O
</s>
<s>
The	O
following	O
algorithms	O
require	O
custom	O
fields	O
to	O
be	O
added	O
to	O
the	O
TCP	B-Protocol
packet	O
structure	O
:	O
</s>
<s>
XCP	O
routers	O
set	O
the	O
feedback	O
value	O
explicitly	O
for	O
efficiency	O
and	O
fairness	B-Protocol
.	O
</s>
<s>
The	O
rate	O
is	O
set	O
as	O
a	O
function	O
of	O
this	O
maximum	O
congestion	O
,	O
resulting	O
in	O
max-min	B-Algorithm
fairness	I-Algorithm
.	O
</s>
<s>
BIC	B-Protocol
is	O
used	O
by	O
default	O
in	O
Linux	B-Operating_System
kernels	I-Operating_System
2.6.8	O
through	O
2.6.18	O
.	O
</s>
<s>
CUBIC	B-Protocol
is	O
used	O
by	O
default	O
in	O
Linux	B-Operating_System
kernels	I-Operating_System
since	O
version	O
2.6.19	O
.	O
</s>
<s>
PRR	O
is	O
incorporated	O
in	O
Linux	B-Operating_System
kernels	I-Operating_System
to	O
improve	O
loss	O
recovery	O
since	O
version	O
3.2	O
.	O
</s>
<s>
BBR	O
is	O
incorporated	O
in	O
Linux	B-Operating_System
kernels	I-Operating_System
to	O
enable	O
model-based	O
congestion	O
control	O
since	O
version	O
4.9	O
.	O
</s>
