Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
9
, No
.
5
,
Octo
ber
201
9
, pp.
4163
~4
175
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v9
i
5
.
pp4163
-
41
75
4163
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
A n
ew
f
rame
wor
k to
alleviate
DDoS v
uln
erabiliti
es in
cloud
c
om
pu
ting
A.
S
ar
avanan
1
, S
.
Sath
ya
B
am
a
2
, Sei
fedine
Kadry
3
, Laks
hma
n
a Kum
ar
Ram
as
am
y
4
1
Depa
rtment of
Com
pute
r
Scie
n
ce
and
Appl
ic
a
tions
,
Sree
Sara
sw
at
hi
Th
y
ag
araja
Coll
ege,
Pol
lach
i,
Ind
ia
2
Inde
pende
n
t
R
e
sea
rch
er
,
Co
imbatore,
Ind
ia
3
Depa
rtment of
Mathe
m
at
i
cs
an
d
Com
pute
r
Sci
e
nce
,
Facu
lty
o
f
S
ci
en
ce,
Be
irut Arab
Univer
si
t
y
,
Le
banon
4
Depa
rtment of
MCA
,
Hindusthan
Col
le
ge
of En
gine
er
ing
and
T
e
chnol
og
y
,
Indi
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
a
n
14
, 2
01
9
Re
vised
Ma
r
16
, 2
01
9
Accepte
d
Apr
2
7
, 201
9
In
the
comm
unic
a
ti
on
ag
e,
the
Inte
rne
t
has
gr
owing
ver
y
fast
and
m
ost
industri
es
re
l
y
on
it
.
An
essen
ti
al
p
art
of
In
ternet
,
W
eb
appl
i
ca
t
ions
li
k
e
onli
ne
booking
,
e
-
banki
ng
,
onli
n
e
shopping,
and
e
-
learni
ng
pl
a
y
s
a
vit
a
l
role
in
eve
r
y
da
y
li
fe
.
Enha
nce
m
en
ts
h
ave
be
en
m
ade
in
thi
s
dom
ai
n,
i
n
which
the
web
serve
rs
depe
nd
on
cl
oud
loc
a
ti
on
for
resourc
es.
Man
y
o
rga
nizati
ons
aro
und
the
worl
d
cha
nge the
ir
o
per
ations a
nd
da
ta
storag
e
from
loc
al
to cl
ou
d
pla
tforms
for
ma
n
y
r
ea
sons
espe
cially
th
e
avail
abi
lit
y
factor.
E
ven
though
cl
oud
computin
g
is
conside
r
e
d
a
r
enowne
d
technolog
y
,
it
has
m
an
y
cha
l
le
nges,
the
m
ost
important
one
is
sec
uri
t
y
.
One
of
the
m
aj
o
r
issue
in
th
e
cl
oud
se
cur
ity
is
Distribut
ed
Den
ia
l
of
Servi
ce
at
t
ac
k
(DD
oS),
wh
ic
h
r
esult
s
in
serious
loss
if
the
at
t
ac
k
is
succ
essful
and
le
ft
unnoti
c
ed.
Thi
s
pape
r
foc
uses
on
pre
v
ent
ing
and
d
et
e
c
ti
ng
DD
oS
at
t
acks
in
distri
bu
te
d
and
cl
ou
d
envi
ronm
ent
.
A
new
fra
m
ework
has
bee
n
suggested
to
al
le
v
iate
the
DD
oS
at
t
ac
k
and
to
provide
av
ai
l
abi
lit
y
of
c
lou
d
resourc
es
to
it
s
u
sers.
The
fra
m
ework
int
roduc
es
thre
e
scre
eni
ng
te
sts
VIS
UA
LCO
M,
IMG
COM
,
and
AD
-
IMG
COM
to
pre
ven
t
the
attac
k
an
d
two
queu
es
with
c
ert
a
in
constra
in
ts
to
det
e
ct
the
at
t
ac
k
.
The
resul
t
of
our
fra
m
ework
show
s
an
improvem
ent
an
d
bet
t
er
ou
tc
om
es
and
prov
ide
s
a
r
ec
over
ed
fr
om
at
tack
det
e
ct
ion
wi
th
h
igh
availa
b
il
i
t
y
rat
e
.
Also,
th
e
p
erf
orm
anc
e
of
t
he
queui
ng
m
odel
has
b
ee
n
ana
l
y
sed
.
Ke
yw
or
d
s
:
Alle
viate
f
ram
ewor
k
Cl
oud
s
ec
ur
it
y
Distrib
uted
d
e
nial o
f
s
e
rv
ic
e
Qu
e
uing
m
od
e
l
Syst
e
m
av
ai
la
bili
t
y
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Seifedi
ne Kad
r
y,
Dep
a
rtm
ent o
f M
at
hem
a
ti
cs and
Com
pu
te
r S
ci
ence,
Faculty
of scie
nce,
Be
irut Ara
b U
niv
e
rsity
, Leb
a
non
.
Em
a
il
:
s.k
adr
y
@b
a
u.
e
du.lb
1.
INTROD
U
CTION
Cl
oud
com
pu
ti
ng
is
a
n
em
ergi
ng
fiel
d
that
pro
vid
es
virt
ual
so
luti
ons
t
o
th
e
business
a
nd
oth
e
r
use
rs.
It
offer
s
se
veral
serv
ic
es
to
custom
ers
with
un
i
qu
e
featur
es
li
ke
im
pr
ov
e
d
scal
abili
ty
,
avail
abili
ty
and
m
anag
eabil
it
y
of
re
source
s
on
dem
and
.
T
he
us
er
’s
data
an
d
ap
plica
ti
on
s
are
store
d
on
c
loud
stora
ge
a
nd
the
us
ers
ca
n
acc
ess
them
a
t
a
nyw
her
e
a
nd
at
any
tim
e.
Eve
n
thou
gh
t
he
cl
oud
e
nv
i
ronm
ent
has
var
i
ou
s
adv
a
ntage
s,
it
has
var
i
ous
ris
ks
a
nd
chall
e
nges
i
n
the
cas
e
of
sec
ur
it
y,
since
it
sto
res
the
use
r
’s
per
s
on
al
,
confide
ntial
an
d
crit
ic
al
data.
Th
us
,
tr
us
tw
ort
hin
ess,
se
r
vice
avail
abili
ty
a
nd
se
ns
it
ive
da
ta
protect
ion
a
re
the
i
m
po
rtant
c
on
c
ern
s
for
cl
ou
d
us
ers
.
I
nter
national
Data
Corp
or
at
io
n
(
I
DC)
cond
ucted
a
s
urvey
in
Aug
us
t
2008
with
the
cl
oud
us
e
rs.
Acc
ord
ing
t
o
the
surv
ey
,
secu
rity
is
the
m
ajo
r
barrier
in
us
in
g
cl
oud
en
vir
on
m
en
t
[1
]
.
The
cl
oud
se
rvi
ce pr
ovide
rs
s
hould
gu
a
ra
ntee connecti
vity
, av
ai
la
bili
ty
, a
nd
sec
ur
it
y t
o
the c
loud u
s
ers
.
I
f
this
gu
a
ra
ntee
is
c
om
pr
om
ise
d,
then
the
us
e
r
or
the
or
gan
iz
at
ion
will
suffe
r
ridic
ulously
[
2].
Den
ia
l
of
Ser
vice
(DoS
)
at
ta
cks
and
Distri
bu
te
d
Den
ia
l
of
Ser
vice
(
DDoS)
at
ta
cks
are
tw
o
m
ai
n
netw
ork
centre
d
vu
l
nerabil
it
ie
s
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
5
,
Oct
ober
20
19
:
4163
-
4175
4164
ta
rg
et
in
g
the
web
se
r
ve
rs
and
the
cl
ou
d
se
rv
e
rs
to
m
ake
res
ourc
e
inacce
ssi
ble
to
a
uth
e
ntic
us
e
rs
.
The
pr
e
sence
of
a
D
D
oS
at
t
ack
was
id
enti
fied
i
n
J
un
e
19
98
an
d
t
his
is
consi
der
e
d
a
s
t
he
fi
rst
oc
cu
rrence
in
web
hist
or
y
[3]
.
As
t
he
data
in
the
cl
oud
is
avail
able
t
o
t
he
le
gitim
at
e
us
ers
at
a
ny
ti
m
e
an
d
si
nce
it
can
be
acce
ssed
f
ro
m
anyw
her
e
,
this
at
ta
ck
is
i
ncr
e
asi
ng
eve
ry
ye
ar.
Since
the
da
ta
are
distrib
ut
ed
in
cl
oud
a
nd
we
b
serv
e
rs,
se
ver
a
l
netwo
r
k
at
ta
ck
too
ls
are
de
velo
ped
a
nd
re
adily
avail
able
to
insti
gate
the
at
ta
ck.
Re
f.
[4
]
is
consi
der
e
d
t
o
be
the
first
t
ool
wh
ic
h
has
a
s
et
of
pro
gr
am
s
wr
it
te
n
i
n
C.
Gen
e
rall
y,
this
too
l
was
us
e
d
widely
by
hack
e
rs
to
la
un
c
h
D
D
oS
a
tt
acks.
Seve
ral
DDoS
at
ta
ck
too
ls
are
c
urrent
ly
avail
able
to
si
m
ulate
eno
rm
ou
s
pack
et
re
quest
s
co
nc
urren
tl
y
to
victi
m
serv
er
,
wh
ic
h
pro
vid
es
una
vaila
ble
serv
ic
e
to
the
le
gitim
a
te
us
er
s
.
Howe
ver,
due
to
the
im
pr
ov
e
m
ent
in
te
ch
no
l
og
y,
botnet
s
are
intr
oduc
ed
to
c
omm
en
ce
the
D
D
oS
at
ta
cks.
The
m
al
war
e
is
plante
d
in
th
e
group
of
syst
e
m
s
wh
ic
h
bec
om
es
zo
m
bie
or
botnet
th
at
in
sti
gate
si
m
ultan
eo
us
request
to
the
victim
to
create
the
at
ta
ck
tra
ff
ic
.
If
t
his
at
ta
ck
c
on
ti
nues
f
or
a
lo
ng
e
r
peri
od
,
it
eve
n
ex
cl
ude
s
web
s
pid
e
rs
an
d
we
b
craw
le
r
s
fr
om
visit
ing
the
web
sit
e
w
hich
le
ads
to
the
reducti
on
in
pag
e
ra
nk
i
ng
for
the
par
ti
cu
la
r
sit
e.
So
,
t
he
us
er
m
ay
no
t
s
how
i
nt
erest
in
visit
in
g
the
sit
e,
sinc
e
the
pa
ges
ar
e
not
sho
wn
by
the
search
engine
s
du
e
to
l
ow r
a
nkin
g [5
]
.
The
ob
j
ect
ive
of
this
pap
e
r
is
to
detect
and
analy
se
Distrib
uted
D
enial
of
Ser
vice
(DDoS
)
at
ta
cks
in
a
cl
oud
com
puti
ng
e
nv
i
ronm
ent.
T
he
pro
po
sed
fr
am
ewo
r
k
can
be
us
e
d
i
ns
te
ad
of
im
age
reCA
PTCH
A
a
nd
oth
e
r
m
e
tho
ds
al
ong
with
no
CAPTC
H
A.
The
pro
pose
d
pr
e
ven
ti
on
str
at
egies
identif
y
bo
ts
from
hu
m
ans
thr
ough
scree
ning
te
sts
w
hi
ch
ca
n
be
pe
rfor
m
ed
after
no
CAP
TCH
A
reC
AP
TC
H
A
a
nd
t
he
de
te
ct
ion
strat
egies
m
i
tig
at
es
the
D
D
oS
at
ta
cks
su
c
h
as
SYN
flo
od,
ACK
flo
od,
L
ay
er
7
D
D
oS
a
tt
ack,
Sm
ur
f
at
ta
cks,
and
oth
e
rs.
Th
e
per
f
or
m
ance
of
the
syst
e
m
is
co
m
par
ed
with
oth
e
r
m
eth
ods
an
d
pr
oves
to
be
bette
r
with
a
detect
ion rate
of 98%.
2.
DIS
T
RIBUT
ED D
E
NI
AL
OF SE
RV
I
CE
A
TT
A
CK
Mostl
y,
in
this
ty
pe
of
at
ta
ck
,
the
ta
r
get
sys
tem
is
floo
de
d
with
inc
om
ing
m
essages
an
d
f
orces
the
syst
e
m
to
m
ove to a b
us
y st
at
e, the
reb
y
de
nying
se
r
vice to l
egitim
at
e u
sers
. A ty
pical
DD
oS
att
ack
sce
na
rio
is
sh
ow
n
in
Fi
g
ur
e 1
.
Fig
ure
1. A
t
yp
ic
al
scenar
i
o
f
or
distrib
uted de
nial o
f
se
rv
ic
e
at
ta
ck
Fo
r
a
su
cce
ssf
ul
D
D
oS
at
ta
ck,
t
he
at
ta
c
ker
us
es
a
t
wo
st
age
process
[
6]
.
In
the
fi
rst
sta
ge,
t
he
at
ta
cker
e
xp
l
oits
a
vulne
rab
il
it
y
and
plants
a
Tr
oj
a
n
horse
or
a
ny
m
al
war
e
on
a
ta
r
get
m
achine.
T
he
m
al
wa
r
e
m
ay
no
t
be
noti
ceable
since
it
do
e
s
no
t
c
ause
a
ny
h
ar
m
to
the
ta
r
ge
t
syst
e
m
.
Now,
t
he
ta
r
get
m
achin
e
beco
m
es
the
DDoS
m
ast
er
or
botm
ast
er.
Eit
her
the
bot
m
ast
er
or
t
he
at
ta
cker
ide
ntifie
s
ot
her
vulnera
ble
A
tt
a
c
k
e
r
w
it
h
Bo
t
M
a
ste
r
B
o
tn
et
V
ictim
S
tag
e
1
.
A
tt
a
c
k
e
r
p
lan
ts a M
a
lw
a
re
St
a
g
e
2
.
Zo
m
b
ie
s
a
tt
a
c
k
th
e
V
ictim
Zo
m
b
ie
s
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A n
ew
fr
amew
or
k t
o all
evi
ate DDoS v
uln
e
rabil
it
ie
s
…
(
A. Sarava
nan)
4165
syst
e
m
s
and
in
fects
them
with
the
m
al
war
e.
Each
of
these
syst
e
m
s
no
w
be
com
es
zom
bi
e
or
bot.
T
his
gro
up
of
bots
is
cal
le
d
a
botn
et
.
Ne
ver
t
heless,
the
se
syst
e
m
s
carr
y
ou
t
their
nor
m
al
wo
r
k
sinc
e
they
are
una
war
e
of
the
resid
ent
z
om
bie.
In
due
c
ourse,
t
he
at
ta
cker
i
den
ti
fies
a
victim
and
with
the
help
of
botm
as
te
r,
s
ends
a
sign
al
to
al
l
the
zom
bies
to
l
aun
c
h
t
he
at
ta
ck
on
the
victi
m
.
No
w
,
the
vi
ct
i
m
enco
un
te
r
s
a
nu
m
ber
of
at
ta
ck
s
from
al
l zo
m
bi
es at t
he
sa
m
e
po
i
nt o
f
ti
m
e.
More
ov
e
r,
the
zom
bies
m
a
y no
t use th
e sam
e att
ack.
Each
zom
bie
m
ay
u
se d
if
fere
nt f
l
ooding att
acks.
A
s
in [
5], DDoS att
acks
are cate
go
rized
i
nto
t
wo br
oad cat
egories:
-
Net
w
ork
Ce
ntr
ic
D
D
oS at
ta
ck
(
Lay
er
-
3 at
ta
ck)
-
Applic
at
ion C
entric
DDoS at
ta
ck (Lay
er
-
7 at
ta
ck)
Gen
e
rall
y,
net
work
an
d
tr
ans
port
cent
ric
D
DoS
at
ta
cks
a
r
e
carrie
d
ou
t
t
o
e
xh
a
us
t
se
rver’
s
res
ource
s
by
arr
ay
in
g
a
n
en
or
m
ou
s
num
ber
of
pa
c
kets
of
TC
P,
UDP,
ICMP
prot
oco
ls.
T
hes
e
are
nam
ed
as
flood
at
ta
cks.
Lay
er
7
at
ta
ck
e
xp
l
oi
ts
the
vulne
rabi
li
t
ie
s
of
a
pp
li
c
at
ion
le
vel
pro
tocols
a
nd
de
pl
et
es
victim
se
rv
e
r'
s
resou
rces
us
in
g HTTP
and
ot
her ap
plica
ti
ons.
3.
RELATE
D
W
ORK
The
de
fe
ns
e
a
gainst
D
D
o
S
a
tt
ack
can
be
pro
vid
e
d
at
var
i
ou
s
sta
ges.
S
om
ani
et
al
.
categ
ori
zed
the
def
e
ns
e
a
gains
t
DDoS
as
at
t
ack
pre
ven
ti
on
,
at
ta
ck
detect
ion
a
nd
at
ta
ck
recovery
[7
]
.
Figure
2
de
pic
ts
the
def
e
ns
e
m
echan
ism
against
th
e
D
DoS
at
ta
ck.
A
detai
le
d
s
urvey
ab
out
the
detec
ti
on,
pre
ve
ntion,
a
nd
rec
ov
e
ry
m
et
ho
ds wit
h t
heir pr
os an
d
c
on
s
has al
s
o be
en descri
bed by
them
[
8].
Fig
ure
2. De
fe
ns
e m
echan
ism
ag
ai
nst
the
D
DoS att
ack
As
a
pre
ven
ti
ve
m
easur
e,
“
Gr
a
phic
al
Turi
ng
Test
”
ca
n
be
us
e
d
to
di
sti
ng
uis
h
m
ac
hin
es
from
a
hu
m
an.
T
hes
e
are
inten
de
d
to
be
unrea
da
ble
by
intel
li
ge
nt
m
achines,
or
e
ve
n
the
sc
reen
rea
der
s
c
annot
unde
rstan
d
the
m
.
Ba
sic
a
ll
y,
t
he
use
of
c
halle
ng
e
res
pons
e
syst
e
m
m
a
y
help
in
pr
e
ve
ntin
g
the
at
ta
ck
.
O
ne
of
the
m
os
t
com
m
on
i
m
ple
m
e
ntati
on
s
as
a
pr
e
ve
nti
on
str
at
egy
is
a
Tu
r
ing
Te
st
in
th
e
form
of
CA
PTCH
A
i
m
ages
[9
]
.
T
hi
s
protoc
ol
is
us
ua
ll
y
con
sid
ered
as
one
of
the
m
os
t
ideal
m
et
ho
ds
in
t
he
cl
ass
of
c
ha
ll
eng
e
respo
ns
e
syst
e
m
s.
It
trie
s
to
per
cei
ve
wh
et
her
the
us
e
r
is
a
bot/
at
ta
cker
m
achine.
T
his
ty
pe
of
p
r
oto
c
ol
m
ay
al
so
inclu
de
gr
aph
ic
al
te
st
[10],
Text
P
uzzles
[11],
Crypto
Pu
zzl
es
an
d
P
r
oof
-
of
-
Wor
k
[
12
]
to
pre
ven
t
the
bot
dr
i
ven
at
ta
ck
to
occ
ur.
Seve
r
al
distor
ti
on
or
no
ise
,
s
uch
as
wav
i
ness
an
d
horizo
ntal
stroke
were
al
so
add
e
d
to
escal
at
e
the
com
plexit
y
of
breakin
g
the
CA
PTCH
A
with
a
com
pu
te
r
pro
gr
am
.
The
reCAPTC
HA
a
nd
i
m
age
reCAP
TCH which is an
enha
ncem
ent o
f
CA
PTCH
A
supp
li
es the w
ebsite
s
w
it
h
i
m
ages o
f
w
ords
that ar
e h
ar
d
to
rea
d
f
or
opti
cal
char
act
er
r
ecognit
ion
(
O
CR
)
software
,
as
a
chall
en
ge
to
the
c
li
ents.
Im
age
identific
at
ion
CAPTCH
A
is
al
so
widely
use
d
in
rece
nt
days.
Var
i
ou
s
m
et
ho
ds
li
ke
Nam
ing
CAP
TCHA
an
d
A
no
m
al
y
Detect
ion
CA
P
TCHA
a
re
al
so
in
us
e.
T
he
m
ai
n
dow
ns
ides
of
the
a
bove
m
et
hods
are
the
gr
a
phic
s
ge
nerat
ion
and
sto
ra
ge
s
pa
ce
over
hea
d.
Conv
er
sel
y,
te
xt
pu
zzl
es
can
al
so
be
us
e
d
to
ide
ntify
the
bot
syst
em
,
but
the
lim
it
at
ion
is
OCR
at
ta
cks.
Ap
art
from
challen
ge
re
spo
ns
e
syst
e
m
,
oth
er
m
et
ho
ds
ha
ve
al
so
been
pro
pose
d
to
pr
e
ve
nt the D
DoS att
ack
with
restrict
ed
acc
ess [13
]
.
Howe
ver,
sev
eral
m
e
th
ods
hav
e
t
he
po
s
sibil
it
y
of
puzzl
e
accum
ulati
on
at
ta
ck
a
nd
even
puzzl
e
gen
e
rati
on
an
d
sp
ace
to
store
the
i
m
ages
are
add
it
iona
l
overh
ea
d.
Cl
ie
nt
pu
zzl
e
m
echani
s
m
is
i
m
ple
mented
in
[
14
]
.
T
he
s
econd
sta
ge
is
an
at
ta
ck
det
ect
ion
.
With
t
he
he
lp
of
s
om
e
detect
i
on
m
et
ho
ds,
the
at
ta
ck
sign
at
ur
es
are
detect
ed.
The
at
ta
ck
m
ay
be
at
the
init
ia
l
sta
ge
without
a
ny
inj
ect
io
n
or
it
m
ay
be
at
th
e
final
sta
ge
wh
e
re
i
t
has
al
rea
dy
aff
ect
e
d
the
syst
e
m
.
Several
m
et
ho
ds
ha
ve
been
pr
opose
d
us
i
ng
a
no
m
aly
detect
ion
[
15]
,
web
beh
a
vior
[
16
]
,
trace
bac
k
m
et
ho
d
[17],
thres
hold
filt
ering
wh
ic
h
inclu
des
hop
c
ount
[18]
,
request
co
unt
[19]
an
d
c
onfi
den
ce
base
d
filt
erin
g
[
18]
.
A
Q
ueu
e
m
od
el
to
de
te
ct
the
DDo
S
at
ta
ck
is
pro
posed
[
20
]
.
The
c
ov
a
riance
a
naly
sis
m
od
el
has
been
al
so
im
p
lem
ente
d
to
de
te
ct
the
at
ta
ck
[21]
.
C
h
a
l
l
e
n
g
e
R
e
sp
o
n
se
P
R
E
V
E
N
T
I
O
N
D
E
T
E
C
T
I
O
N
R
E
C
O
V
E
R
Y
A
t
t
ac
k
e
r
U
se
r
C
h
a
l
l
e
n
g
e
R
e
sp
o
n
se
D
r
o
p
P
a
c
k
e
t
s
V
i
c
t
i
m
S
e
r
v
e
r
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
5
,
Oct
ober
20
19
:
4163
-
4175
4166
To
detect
m
odern
botnet
-
li
ke
m
al
war
e
base
d
on
a
no
m
al
ou
s
pa
tt
ern
s
i
n
a
netw
ork,
e
nt
ropy
-
base
d
net
work
ano
m
al
y detec
ti
on
m
et
ho
d i
s e
xp
la
ine
d
i
n [22
]
.
The
aut
hors
L
on
ea
,
et
al
.
pr
op
os
ed
a
m
eth
od
that
com
bin
es
the
repo
rt
giv
e
n
by
th
e
intru
si
on
detect
ion
syst
em
s
dep
loye
d
i
n
virtu
al
m
achines
wit
h
a
dat
a
f
us
io
n
a
ppr
oa
ch
[1
]
.
B
ut
m
any
of
t
hese
m
et
hods
do
not
ha
ve
s
uppo
rt
f
or
ef
fici
ency
a
nd
s
om
e
m
et
ho
ds
a
re
not
scal
able.
T
he
t
hir
d
sta
ge
is
a
r
eco
ver
y
sta
ge
i
n
wh
ic
h
the
m
eth
ods
a
re
to
im
plem
ent
in
the
victim
serv
er
to
serv
e
it
s
use
r
or
to
rec
over
from
the
a
tt
acks.
Seve
ral
m
et
ho
ds
i
nclu
ding
m
igrati
on
[
23
]
an
d
bac
kup
r
eso
ur
ces
[
24
]
are
s
uggeste
d
by
a
few
rese
arch
e
rs
.
But
at
this
sta
ge,
im
ple
m
ent
at
ion
of
a
ny
m
et
hod
le
ad
s
to
an
ov
e
rh
e
ad
of
reserve
d
resour
ces
an
d
c
os
t
s
to
th
e
victim
serv
er
.
Re
centl
y,
a
pr
otect
ion
poli
cy
to
dy
nam
ic
all
y
instal
l
secur
i
ty
app
li
cat
ion
s
across
the
c
on
trolle
r
and
switc
he
s
has
been
pr
opos
e
d
by
Ha
n
et
al
.
tha
t
ide
ntifie
s
the
acc
ur
at
e
l
ocati
on
of
the
botnet
[25].
Suppor
t
Vect
or
Ma
chi
ne
(
S
VM)
al
gorith
m
has
been
use
d
in
creati
ng
the
m
od
el
f
or
cl
assifi
cat
ion
of
D
oS
at
ta
cks
an
d
norm
al
network
be
hav
i
or
s
[
26]
.
T
he
s
urve
y
on
va
rio
us
distrib
uted
de
nial
-
of
-
se
r
vice
at
ta
ck
,
pr
e
ve
ntion,
a
nd
m
itigati
on
te
chn
i
qu
e
s
ha
s
be
en
discusse
d
in
Ma
hja
bin
et
al
.
[
27
]
.
Th
us,
f
ro
m
the
li
te
ratur
e
su
r
vey,
it
is
cl
ear
that
the
m
ai
n
lim
it
ation
s
in
pr
e
ven
ti
on
st
rategies
a
re
grap
hics,
i
m
age,
puzzl
e
and
te
xt
gen
e
rati
on
as
well
as
the
sto
r
age
s
p
ace
ove
r
head.
A
ddit
iona
ll
y,
so
m
e
m
eth
ods
pa
ve
the
way
f
or
oth
e
r
at
ta
cks
su
c
h
as
OCR
and
pu
zzl
e
accu
m
ula
ti
on
at
ta
cks.
I
n
case
of
a
tt
ack
detect
ion,
few
m
et
ho
ds
do
no
t
sup
port
m
or
e
eff
ic
ie
ncy
an
d
scal
abili
ty
.
In
the
la
st
case
of
at
ta
ck
rec
overy,
the
reserved
resou
rces
and
t
he
costs
to
the
victim
ser
ver a
re m
or
e.
4.
PROP
OSE
D M
ET
HO
D
The
m
ai
n
go
a
l
of
the
syst
e
m
is
to
m
i
ti
gate
the
DDoS
at
ta
ck
by
pr
ov
i
ding
the
preven
ti
ve
an
d
detect
ive
m
easur
es
.
T
he
over
al
l
process
of
t
he
pro
po
s
ed
syst
e
m
is
giv
e
n
be
low
in Figure
3
. Q
ueu
i
ng
Mo
del
is
us
e
d
on
the
se
rv
e
r
side.
A
Fi
nite
Ca
pacit
y
Ma
rkov
ia
n
Q
ue
uing
Mo
del
M/
M/
1/K
is
us
ed
[
28
]
.
It
is
a
sing
le
serv
e
r
queue
w
it
h
a
que
ue
siz
e
K.
T
he
se
rv
e
r
has
tw
o
queu
es,
P
ro
ce
ssin
g
Qu
e
ue
a
nd
Wa
it
ing
Q
ue
ue.
A
ll
the
request
from
t
he
cl
ie
nt
is
ver
i
fied
an
d
it
is
store
d
in
the
processin
g
que
ue
.
If
the
proces
s
ing
que
ue
is
ful
l
and
if
certai
n
co
nd
it
ion
s
m
et
,
the
request
will
be
store
d
on
the
wait
ing
qu
e
ue
.
The
m
ai
n
goal
of
intr
oduci
ng
t
he
wait
ing
qu
e
ue i
s to
process all
the p
ackets l
at
er,
instea
d of
dr
oppi
ng
the s
us
pici
ou
s
pac
ke
ts. A
lso
based on
th
e
nu
m
ber
of
pa
ckets
in
the
qu
e
ue,
t
he
pr
even
ti
on
m
ec
han
ism
var
ie
s.
The
th
ree
e
nh
a
nce
d
CAP
TCHA
m
echan
ism
s
are
introd
uced
nam
el
y
VI
SUA
LCOM,
IM
G
COM,
A
D
-
IM
GCOM.
If
t
he
processi
ng
qu
eue
is
fu
ll
,
the
n
the
intric
at
e
m
et
ho
d
IMGC
OP
or
AD
-
IM
GCOM
is
giv
en
as
a
chall
eng
e
t
o
the
us
er,
el
se
the
si
m
ple
m
et
ho
d V
ISU
ALCOM is
u
se
d.
Fig
ure
3. The
ov
e
rall
arc
hitec
ture of
the m
itigati
on
fr
am
ew
ork
for DD
oS
v
ul
ner
a
bili
ti
es
4.1.
Queuin
g
m
odel
Tw
o
qu
e
ues
ar
e
pr
op
os
e
d
in
this
fr
am
ewo
r
k.
All
the
pac
kets
or
re
quest
s
arr
ive
d
at
the
serv
er
a
re
check
e
d
with
detect
ion
co
ns
t
raints
.
I
niti
al
ly,
the
pac
kets
wait
ing
to
be
processe
d
a
re
store
d
i
n
a
pro
cessi
ng
qu
e
ue.
H
ow
e
ve
r,
if
a
ny
of
th
e
const
raints
m
et
or
if
the
proces
sin
g
que
ue
is
fu
ll
,
the
n
t
he
re
quest
s
are
store
d
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A n
ew
fr
amew
or
k t
o all
evi
ate DDoS v
uln
e
rabil
it
ie
s
…
(
A. Sarava
nan)
4167
on
the
wait
ing
qu
e
ue.
Lat
e
r,
the
re
qu
est
s
are
m
ov
ed
to
the
processi
ng
Queue
f
or
f
ur
t
her
process
.
Th
us
,
al
l
the
requests
a
re
proces
sed
an
d
even
a
s
uspic
iou
s
re
quest
is
al
so
proces
se
d
bu
t
with
som
e
delay
or
wh
e
n
t
he
serv
e
r
is
idle.
The
proc
essin
g
queue
is
fi
nite
with
Po
iss
on
arr
i
val
an
d
expo
nen
ti
al
ser
vice
an
d
the
wait
in
g
qu
e
ue
is
infi
ni
te
.
T
her
e
are
so
m
e
assu
m
ptions
to
be
fo
ll
owe
d
in
im
ple
m
enting
the
queui
ng
m
od
el
.
The
se
rv
ic
in
g
m
et
ho
d
ca
n
ac
cept
an
d
sto
re
k
re
quest
s
m
e
rely
.
The
ar
riva
l
rate
of
a
re
quest
s
(t
he
num
ber
of
requests
ar
rive
d
pe
r
un
it
ti
m
e)
is
de
note
d
by
λ.
T
he
ef
fe
ct
ive
arr
i
val
of
the
request
(
the
rate
of
r
eq
uests
enterin
g
in
the
syst
e
m
)
is
give
n
by
λ
e
.
T
hes
e
requests
a
re
store
d
in
t
he
queue
.
k
is
the
m
axi
m
u
m
num
ber
of
serv
ic
es
i
n
the
syst
e
m
.
The
ra
te
of
t
he
re
que
st
no
t
e
nteri
ng
the
que
ue
is
giv
en
by
λ
b
.
H
oweve
r,
t
hese
re
qu
e
sts
are
stor
e
d
in
t
he
wait
in
g
qu
eue.
T
hus
λ=
λ
e
+
λ
b
.
The
se
rvi
ce
rate
of
the
requests
(the
nu
m
ber
of
re
quest
s
serv
ic
e
d
pe
r
unit
tim
e)
is
den
ote
d
by
μ.
An
a
ve
rag
e
num
ber
of
re
quest
s
in
the
sy
stem
is
giv
en
by
L
s
.
The
num
ber
of
re
quest
s
in
the
qu
e
ue
is
gi
ven
by
L
q
.
T
he
a
ver
a
ge
w
ai
ti
ng
ti
m
e
of
the
request
be
fore
com
pleti
on
of
it
s
req
uest
is
giv
en
by
W
s
.
T
he
requests
are
pr
oce
ssed
in
First
Com
e
Fi
rst
Serv
e
d
sche
du
li
ng
.
In
the
queue
,
t
he
requests
wa
it
fo
r
a
ti
m
e
W
q
for
t
he
se
rv
ic
e.
ρ
is
t
he
util
i
zat
ion
facto
r
w
hich
determ
ines
the
pro
portion o
f
ti
m
e that t
he
se
r
ver is b
us
y se
r
vicing re
quest
s
[
29
]
.
4.2.
Preven
tion
str
at
e
gy
Since
sit
es
m
ay
us
e
C
AP
TC
HA
s
as
par
t
of
the
prel
im
inary
re
gistrati
on
proce
dure,
or
as
a
par
t
of
ever
y
l
og
i
n
process,
the
chal
le
ng
e
can
com
plete
ly
blo
c
k
a
ccess
to
the
m
achines
by
dist
inguishi
ng
the
m
fr
om
a
hu
m
an.
This
sect
ion
ex
plains
ab
ou
t
thr
e
e
te
sts
na
m
el
y
VI
S
UALC
O
M,
IMGCOM,
and
A
D
-
IM
G
COM.
Visu
al
c
om
pr
e
hensi
on
(
V
ISU
ALCOM
)
ca
n
be
a
ppli
ed
by
pro
vid
in
g
t
he
visu
al
scene
or
pictu
re
t
o
the
us
er
.
The
n
the
quest
ion
s
re
ga
rd
i
ng
the
visu
al
sce
ne
can
be
give
n
as
a
chall
en
ge
to
the
use
r.
The
us
e
r
gi
ve
s
the
answer
as
a
re
sp
onse
.
The
m
ai
n
adv
a
ntag
e
o
f
this
m
et
ho
d
is
that
storag
e
sp
ace.
I
n
nor
m
al
Gr
aph
ic
al
Tur
i
ng
Test
,
eac
h
te
st
nee
ds
one
or
m
or
e
i
m
ages
wh
ic
h
inc
rease
s
the
s
pace
co
m
plexit
y.
Th
us
‘
n’
us
e
rs
nee
d
‘
n’
or
m
or
e
than
n
i
m
ages
to
unde
rgo
thei
r
T
ur
i
ng
Test
.
H
ow
e
ve
r,
in
this
pro
pose
d
m
et
ho
d,
wit
h
a
si
ng
le
i
m
age,
four
to
fi
ve
qu
est
ion
s
can
be
fr
am
ed.
So
a
sing
le
im
age
can
serv
e
‘m
’
us
er
s
wh
e
re
m
is
the
nu
m
ber
of
f
r
a
m
ed
qu
e
sti
on
s
with
a
par
ti
cular
im
age.
The
ne
xt
propose
d
m
et
hod,
Im
age
Com
pleti
on
(IM
GCOP)
is
com
plex
wh
e
n
com
par
e
d
to
the
VI
S
U
ALCOM.
In
this
m
et
ho
d,
a
sing
le
im
age
i
s
div
ide
d
into
par
ts.
T
he
inc
om
plete
i
m
age
al
on
g
w
it
h
the
par
ts
a
r
e
giv
e
n
as
a
chall
eng
e
t
o
the
us
er
.
The
us
e
r
is
suppose
d
to
drag
a
nd
dro
p
the
par
ts
to
m
ake
a
com
plete
i
m
age.
If
t
he
i
m
age
m
at
ches
with
the
one
in
the
databa
s
e,
the
n
the
use
r
is
authe
ntica
te
d.
As
li
ke
the
e
xi
sti
ng
syst
em
,
a
sing
le
im
age
is
us
ed
,
but
th
e
com
plexity
i
s
increase
d
th
r
ough
wh
ic
h
eve
n
th
e
intel
li
gen
t
bo
t
cannot
gi
ve
a
cor
rect
res
pons
e
.
A
no
t
he
r
m
e
tho
d,
w
hich
is
a
var
ia
ti
on
of
IMGCO
P
is
I
m
age
Com
pletio
n
wit
h
A
nom
al
y
detection
(AD
-
IMGC
O
M).
It
is
si
m
ilar
to
the
IM
G
COM,
wh
e
re
the
pa
rts
of
t
he
im
age
are
gi
ven
wit
h
ad
diti
onal
an
om
aly.
Thus
the
use
r
has
t
o
identify
the
a
nom
aly
and also
drag
a
nd dr
op the
part
s co
r
rectl
y t
o m
ake th
e
or
i
gin
al
im
age.
Thi
s is sho
w
n i
n F
igure
4.
Fig
ure
4. Sam
ple A
D
-
IM
GCO
M Tu
rin
g
Test
to ide
ntify t
he hu
m
an
an
d
a bot with
5 pa
rts and a
n
i
ncom
ple
te
i
m
age
. Th
e
fou
rth part
is an
anom
al
y
4.3.
Det
ec
tion
s
tr
ateg
y
Eve
n
after
t
he
chall
eng
e
res
pons
e
ver
ific
at
ion
,
t
her
e
is
a
po
s
sibil
it
y
fo
r
flo
od
i
ng
at
ta
ck
.
Th
us
,
t
his
sect
ion
pro
vide
s
the
detect
io
n
m
et
ho
ds
t
o
i
den
ti
fy
t
he
at
ta
cks.
Each
pac
ke
t
is
analy
zed
t
o
ide
ntify
the
at
ta
ck.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
5
,
Oct
ober
20
19
:
4163
-
4175
4168
Seve
ral
par
am
et
ers
are
us
e
d
to
analy
ze
th
e
pa
ckets
li
ke
IP
address
,
TTL,
et
c..
Thr
ee
var
i
ables
Re
qu
e
st
Count
C
R
,
Sour
ce
Co
un
t
C
S
,
S
YN
F
la
g
Co
unt
C
F
a
re
m
ai
ntained
by
the
ser
ver.
The
request
c
ount
f
or
eac
h
source
is
store
d
i
n
C
R
.
Si
m
il
arly
,
new
pac
kets
a
rr
i
vin
g
co
ntin
uous
l
y
from
diff
ere
nt
s
ources
is
st
or
e
d
i
n
C
S
.
Ne
xt,
th
e
pack
et
s
a
rr
i
ving
co
ntin
uous
ly
with
S
YN
flag
on
is
m
ai
ntain
ed
i
n
C
F
to
m
on
it
or
S
YN
flo
od
at
ta
ck
.
Als
o,
th
e
thres
ho
l
d
valu
e
fo
r
eac
h
var
i
able
is
set
to
identify
the
DDo
S
at
ta
ck.
The
f
ollow
i
ng
ar
e
va
rio
us
co
nd
it
io
ns
that
are to
b
e
c
hec
ke
d
to
m
ov
e the
p
ac
kets to
the
processi
ng que
ue
a
nd w
ai
ti
ng
qu
e
ue.
-
A
Tim
e
To
Liv
e
(TTL
)
is
an
e
igh
t
bit
fiel
d
to
sp
eci
fy
the
m
axim
u
m
l
ifet
i
m
e
of
a
n
IP
packet
.
From
so
ur
c
e
to
destinat
io
n,
the
pac
ket
will
pass
t
hroug
h
s
ever
al
routers
and
e
ach
route
r
dec
re
ases
t
he
TTL
val
ue
of
an
IP
pack
et
by
one.
As
in
[
30
]
the
ser
ver
m
ain
ta
ins
IP2HC
t
able.
Eac
h
ar
ri
ved
pac
ket
is
ver
ifie
d
with
the
IP
2HC t
able w
it
h
the cal
culat
ed
HC and in Access Co
ntr
ol
List
(
ACL),
if
there is a m
a
tch
, th
e
n
the
pac
ket
unde
rgoes
the n
ext
se
c
ur
it
y
check,
el
se
it
is
sp
oo
fed.
In
s
uc
h
case,
the
pac
kets
can
be
ad
ded
to
the
w
ai
ti
ng
qu
e
ue
a
nd the
ACL ca
n be
updated wit
h
t
he parti
cula
r
I
P a
ddress.
-
If
se
ver
al
pac
ke
ts
arr
ive
d
fro
m
the
sa
m
e
so
ur
ce
,
then
t
he
count
(
request
count
C
R
)
will
be
m
ai
ntained
.
If
t
he
c
ount
e
xceeds
the
th
r
esh
old
value
T
R
,
the
n
the
pa
ckets
are
sto
r
ed
in
the
wait
ing
queue
el
se
it
unde
rgoes
t
he next c
hec
k.
-
Si
m
il
arly
,
if
sever
al
ne
w
pa
ckets
a
rr
ive
d
con
ti
nu
ously
f
ro
m
diff
e
re
nt
so
urces
,
a
gain
the
c
ount
cal
le
d
so
urce
c
ount
C
S
,
is
m
ai
nt
ain
ed
an
d
up
dated.
A
gain
if
it
excee
ds
t
he
t
hr
es
hold
T
S
th
en
th
e
pack
et
s
are
m
ov
ed
to t
he wai
ti
ng
queu
e.
-
The
ne
xt
pa
ra
m
et
er
is
SY
N
flag.
If
se
ver
al
pack
et
s
ar
rive
d
co
ntin
uous
ly
with
SYN
fla
g
on,
the
c
ount
C
F
will
be
m
ai
nta
ined.
If
the
c
ount
e
xcee
ds
th
e
thr
es
ho
l
d
T
F
then
t
he
pac
ke
t
will
be
m
ov
ed
to
t
he
wait
in
g
qu
e
ue.
The
pack
et
s
s
tore
d
in
th
e
wait
ing
queue
are
pr
ocesse
d
afte
r
a
m
ini
m
u
m
delay
.
T
he
de
te
ct
ion
al
gorithm
is sh
own
i
n
Fi
gure
5.
Fig
ure
5. Pro
pose
d
detect
ion
al
gorithm
4.4.
Constr
ucti
on
of IP
2HC
t
ab
l
e
The
m
app
ing
betwee
n
the
I
P
address
an
d
the
Hop
Co
un
t
is
m
ai
ntained
by
the
serv
er.
The
Borde
r
Gateway
Proto
col
gen
e
rall
y
m
ai
ntains
the
HC
to
oth
er
hosts
f
or
w
hich
it
nee
ds
to
c
omm
un
ic
at
e.
W
he
nev
e
r
i
t
receives
t
he
pa
cket
f
ro
m
the
pa
rtic
ular
IP
a
ddress
it
ver
i
fies
with
the
IP2
H
C
ta
ble.
I
f
the
r
e
is
no
m
at
ch
fo
r
t
he
par
ti
cula
r
IP
a
ddress,
it
broa
dcasts
the
Ro
ut
e
Re
qu
est
RR
EQ
pac
ket
to
the
neig
hbou
rs
with
the
pa
rtic
ular
IP
address
as
the
Desti
nation
A
ddress
(
D
A)
.
On
rec
ei
ving
the
request,
the
neighb
ours
ve
rify
the
DA
w
it
h
it
s
own
IP
a
ddres
s.
I
f
the
destin
at
ion
a
ddress
i
s
not
their
own
,
they
f
ur
t
her
f
orwa
r
d
to
the
neig
hbouri
ng
r
ou
te
rs.
The
inte
nd
e
d
ho
st
,
on
receiv
ing
t
he
re
quest
,
sen
ds
t
he
Ro
ute
Re
ply
RR
EP
pac
ket
c
onta
ining
t
he
r
ou
te
,
ho
p
count
a
nd
oth
e
r
in
f
or
m
at
ion
to
t
he
s
ource
in
t
he
sam
e
route
as
RR
EQ
bu
t
in
re
ver
se
directi
on.
Als
o,
as
i
n
DMIPS
[
31
]
,
a
ver
ific
at
io
n
ste
p
can
be
car
ried
out
by
se
nding
a
que
ry
to
the
host
an
d
by
set
ti
ng
the
re
trie
ve
d
Hop
C
ount
a
s
t
he
T
TL
value
.
Af
te
r
the
re
ply
has
ar
rive
d,
th
e
source
the
n
updates
the
IP2
HC
ta
ble
with
the
I
P
address
and
H
op Co
unt.
On
ce
the
pac
ke
t
has
a
rr
i
ve
d,
the
IP
a
ddres
s
is
m
at
ched
with
th
e
I
P2H
C
ta
ble.
If
the
re
is
a
m
at
ch,
the
HC w
il
l
be
cal
culat
ed
f
rom
the
TTL
val
ue
an
d
c
om
par
ed
with
t
he
Hop
Co
unt
in
the table
ref
e
rr
e
d
a
s
CH
2
.
The
H
op
Co
unt
val
ue
ca
n
be
directl
y
cal
culat
ed
f
r
om
the
T
TL
val
ue
of
the
rec
e
ived
pac
ket
as
the
interm
ediary
ro
ute
r
decr
ease
s
the
TT
L
value
of
t
he
packet
befor
e
f
orw
ard
i
ng
it
to
th
e
subse
qu
e
nt
r
ou
t
e
r
.
Since,
the
at
ta
cker
can
not
m
od
i
fy
the
val
ue
s
of
t
he
nu
m
ber
of
hops
r
equ
i
red
f
or
a
pack
et
to
reac
h
it
s
Fo
r
each p
acket
Extract the
IP
ad
d
ress, T
T
L,
SYN
f
lag
;
Up
d
ate
Req
u
est Co
u
n
t C
R
,
So
u
rce
C
o
u
n
t C
S
,
SYN
Flag
Co
u
n
t C
F
;
CH
1
= Co
m
p
u
te H
C f
ro
m
the T
TL
;
CH
2
= Acc
ess
HC
f
o
r
th
e I
P add
ress f
ro
m
IP2HC
;
If
(
CH
1
= CH
2
)
th
e
n
Els
e if
(
C
R
<
T
R
)
th
en
Else if
(
C
S
<
T
S
)
th
en
Else if
(
C
F
<T
F
)
th
en
S
to
re
th
e pack
e
t
in th
e Pr
o
cess
in
g
Qu
eu
e
Else
Sto
re
the p
acket
in
the W
aitin
g
Que
u
e
End
I
f
End
For
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A n
ew
fr
amew
or
k t
o all
evi
ate DDoS v
uln
e
rabil
it
ie
s
…
(
A. Sarava
nan)
4169
destinat
io
n,
e
ve
n
th
ough
t
here
is
a
possi
bili
t
y
of
m
od
ify
in
g
the
fiel
ds
in
th
e
IP
h
ea
der.
T
he
cal
cula
ti
on
of
H
op
Count is
giv
e
n i
n
(
1).
Hop
C
ount C
H
1
= In
it
ia
l TTL
Value
–
Final
TTL
Value
(1)
The
Fin
al
TTL
Value
is
the
on
e
e
xtracte
d
from
the
received
pack
et
.
T
he
recei
ver
ca
lc
ulate
s
the
In
it
ia
l
TTL
Value
wh
ic
h
is
m
or
e
chall
eng
ing.
Luc
kily
,
m
axim
u
m
m
od
ern
O
Ss
em
plo
y
on
ly
a
lim
i
t
ed
a
nd
certai
n
init
ia
l
TTL
val
ues,
30,
32,
60,
64,
128,
a
nd
255.
And
si
nce
the
m
axi
m
u
m
num
ber
of
hops
betwe
e
n
any
two
node
s
on
the
inter
ne
t
is
m
or
e
than
30
hops
,
the
re
cei
ver
can
cal
culat
e
the
In
it
ia
l
TTL
Value
as
the
sm
a
ll
est
value
that
is
la
rg
e
r
th
an
the
Final
T
TL
Val
ue.
F
or
exam
ple,
if
th
e
Final
TTL
V
al
ue
is
108,
th
en
the
In
it
ia
l
TTL
Va
lue
will
be
m
ini
m
u
m
of
(
128,
255)
w
hich
is
128.
The
pro
pose
d
m
et
ho
d
de
te
ct
s
var
io
us
DDo
S
at
ta
cks
in
t
he
cl
oud
se
rv
e
r.
I
f
se
ver
al
pac
ke
ts
arr
i
ved
f
rom
the
sam
e
so
ur
ce
,
a
nd
if
it
exceed
s
the
th
reshold
value,
the
pac
ke
ts
are
m
ov
ed
to
the
wait
ing
qu
e
ue.
T
his
set
s
a
lim
it
fo
r
ea
ch
cl
ie
nt
a
config
ur
a
ble
nu
m
ber
of
request
to
the
serv
e
r
w
hich
detect
s
la
ye
r
7
ap
plica
ti
on
l
ay
er
DDoS
at
t
acks.
Also
if
sever
al
new
pa
ckets
arr
ive
d
c
onti
nu
ou
sly
from
different
s
ources
a
nd
the
c
ou
nt
e
xceeds
the
gi
ve
n
th
res
ho
l
d,
the
pack
et
s
are
m
ov
ed
to
the
wait
ing
qu
e
ue.
This
conditi
ons
will
detect
and
m
itigate
t
he
sm
ur
f
at
ta
cks,
ACK
fl
ood
at
ta
cks.
Gen
e
rall
y,
the
se
two
c
onditi
on
s
detect
vari
es
lsy
er
3,
la
ye
r
4
a
nd
la
ye
r
7
D
DoS
at
ta
cks.
T
he
SYN
flo
od
at
ta
cks
can
be
detect
ed
an
d
m
itigated
by
m
on
it
or
ing
the
co
un
t
of
the
pac
kets
ar
riving
with
S
Y
N
flag
on
.
Th
us
the
pr
o
po
sed
m
et
ho
d o
ut
perform
s in
m
i
ti
gating DDoS
vu
l
ner
a
bili
ti
es.
5.
RESU
LT
S
A
ND AN
ALYSIS
The
e
valuati
on
of
a
ny
detect
ion
m
et
ho
d
is
extrem
el
y
i
m
p
or
ta
nt
befo
re
dep
l
oym
ent
in
a
real
-
tim
e
netw
ork.
T
hus
the
exp
e
rim
ental
analy
sis
us
es
a
sing
le
ser
ver
a
nd
15
cl
i
ents
to
ge
ner
at
e
the
netw
ork
traff
ic
and
at
ta
ck
tra
f
fic.
Net
wa
g
to
ol
[
32
]
is
us
e
d
to
produce
t
he
know
n
D
D
oS
at
ta
cks
su
c
h
a
s
a
TCP
S
Y
N
a
tt
ack,
sm
ur
f
at
ta
ck
et
c.
Additi
on
a
ll
y,
to
identify
the
pack
et
s
and
acce
ss
a
ll
it
s
head
er
i
nfor
m
at
ion
,
a
pack
et
captu
rin
g
to
ol
JPCap
is
em
plo
ye
d
[33].
Gen
e
rall
y,
for
transm
itti
ng
and
ca
pturin
g
the
pac
kets
f
ro
m
the
netw
ork,
JPCa
p
will
be
a
per
f
ect
cho
ic
e
wh
i
ch
is
an
open
s
ource
j
a
va
Lib
r
ary.
The
le
giti
m
at
e
traff
ic
ha
s
bee
n
us
e
d
as
a
trai
ni
ng
data
with
100
D
DoS
at
ta
cks.
The
ove
r
al
l
pr
e
ven
ti
on
and
detect
io
n
rate
f
or
the
pr
opose
d
m
et
ho
d
al
ong
with
t
he
e
xisti
ng
m
et
ho
ds
suc
h
as
pu
s
hbac
k
[34],
distanc
e
base
d
an
d
C
2
DF
[33]
have
been
analy
sed
a
nd t
he
c
om
par
ison
is sh
own
i
n
Fi
gure
6.
Fig
ure
6. Per
f
orm
ance A
naly
s
is wit
h D
Do
S
Detect
ion
Ra
te
T
he
e
xi
s
t
i
ng
P
us
hb
a
c
k
m
e
t
ho
d
i
s
l
e
s
s
e
f
f
e
c
ti
ve
s
i
nc
e
i
t
us
es
hi
gh
c
om
pu
ta
t
i
on
po
w
e
r
a
n
d
e
xe
c
ut
i
on
t
im
e
.
T
he
di
s
ta
nc
e
ba
s
e
d
m
e
t
ho
d
d
oe
s
no
t
pe
r
f
or
m
e
f
f
ec
t
i
ve
ly
i
n
c
a
se
of
r
e
c
ov
e
r
y
ph
a
s
e
a
nd
r
e
s
ou
r
c
e
ut
i
l
iz
a
ti
on
.
T
he
s
e
t
ti
ng
s
of
C
2
D
F
m
e
t
ho
d
s
uf
f
e
r
s
f
r
om
l
ow
a
c
c
ur
a
c
y
i
n
de
te
c
t
i
ng
t
he
D
D
o
S
a
t
t
ac
ks
.
T
he
gr
a
p
h
i
n
F
i
gu
r
e
7
s
ho
w
s
t
he
c
om
pa
r
is
on
of
t
he
f
a
l
s
e
ne
ga
t
i
ve
r
a
t
e
for
a
l
l
t
he
e
xi
s
t
in
g
m
et
ho
ds
a
n
d
pr
o
p
os
e
d
a
D
D
o
S
m
it
i
ga
ti
on
a
pp
r
oa
c
h.
A
l
s
o,
t
he
a
na
l
y
s
i
s
ha
s
be
e
n
m
a
de
to
c
om
pa
r
e
t
he
dr
o
po
ut
r
a
t
e
f
or
t
he
p
r
o
po
s
e
d
a
n
d
e
xi
s
t
i
ng
a
l
go
r
i
t
hm
s
by
va
r
y
i
ng
t
he
n
um
be
r
of
r
e
q
ue
s
t
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
5
,
Oct
ober
20
19
:
4163
-
4175
4170
Fig
ure
7. Per
f
orm
ance
a
naly
sis wit
h D
DoS
f
al
se n
e
gative
r
at
e
T
he
d
r
o
p
r
a
t
e
i
s
t
he
f
r
a
c
t
i
o
n
o
f
t
he
nu
m
be
r
of
pa
c
ke
t
s
dr
op
pe
d
t
o
t
h
e
t
ot
a
l
nu
m
be
r
of
pa
c
ke
t
s
.
F
i
gu
r
e
8 r
e
pr
e
s
e
nt
s
t
he
dr
op
o
u
t
ra
t
e c
om
pa
r
i
son
.
F
r
om
t
he
ana
l
y
s
i
s
,
i
t i
s c
le
a
r
t
ha
t
t
he
dr
op
ou
t
ra
t
e
i
s
m
i
ni
m
um
f
or
t
he
pr
op
os
e
d
m
i
ti
ga
t
i
on
f
r
a
m
e
w
or
k
w
he
n
c
om
pa
r
e
d
w
i
t
h
t
he
e
xi
s
t
i
ng
m
e
t
ho
ds
.
Fig
ure
8. Per
f
orm
ance A
naly
s
is wit
h D
Do
S
Dropo
ut Rat
e
A
l
s
o,
t
he
pe
r
f
o
r
m
a
nc
e
e
va
l
u
a
ti
on
ha
s
be
e
n
m
a
de
ba
s
e
d
on
t
he
e
xe
c
ut
i
on
t
i
m
e
f
or
t
he
pr
o
po
s
e
d
m
od
e
l
a
s
a
pa
r
am
et
e
r
.
T
he
t
im
e
t
o
f
e
t
c
h
t
he
im
a
ge
s
a
nd
t
o
ve
r
i
f
y
t
he
r
e
s
po
ns
e
i
s
a
na
l
y
z
e
d
f
or
t
he
pr
op
os
e
d
m
et
ho
d.
T
a
bl
e
1
s
h
ow
s
t
he
a
na
l
y
s
i
s
of
c
ha
l
l
e
ng
e
r
e
s
p
on
s
e
s
y
s
t
em
w
it
h
t
he
e
xe
c
ut
i
on
t
im
e
of
t
hr
e
e
m
e
t
ho
ds
.
H
e
nc
e
,
t
he
a
ve
r
a
ge
e
xe
c
ut
i
on
t
im
e
f
or
V
I
S
U
A
L
C
O
M
i
s
29
.
2
m
s
,
th
a
t
of
I
M
G
C
O
P
i
s
42
.
6
m
s
a
nd
A
D
-
I
M
G
C
O
P
is
48
.
6
m
s
.
Wi
t
ho
ut
t
he
pr
op
os
e
d
pr
e
ve
nt
i
on
t
e
c
hn
i
q
ue
s
,
t
he
e
x
e
c
ut
i
on
t
im
e
of
t
he
s
y
s
t
em
i
s
14
.
4
m
s
.
Table
1.
C
om
par
iso
n of
exec
ut
ion
ti
m
e w
it
h
and w
it
ho
ut pr
opos
e
d pr
e
ve
ntion
m
et
ho
ds
Test No
Execu
tio
n
T
i
m
e
in
m
illiseco
n
d
s
W
ith
VISUA
LCO
M
W
ith
IM
GCO
P
W
ith
AD
-
I
MGCO
P
W
ith
o
u
t Pr
ev
en
tio
n
tech
n
iq
u
es
W
ith
VISUA
LCO
M
1
25
41
46
12
25
2
31
48
53
14
31
3
27
39
45
18
27
4
29
42
48
15
29
5
32
46
49
16
32
6
34
41
50
14
34
7
27
42
51
15
27
8
28
43
48
13
28
9
30
42
47
12
30
10
29
42
49
15
29
Av
erage
2
9
.2
4
2
.6
4
8
.6
1
4
.4
2
9
.2
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A n
ew
fr
amew
or
k t
o all
evi
ate DDoS v
uln
e
rabil
it
ie
s
…
(
A. Sarava
nan)
4171
T
he
a
ve
r
a
ge
t
i
m
e
de
l
a
y
i
s
of
14
.
8,
28
.
2
a
nd
34
.
2
m
il
li
s
e
c
on
ds
r
e
s
pe
c
t
i
ve
l
y,
f
o
r
t
he
pr
op
os
e
d
m
et
ho
ds
w
hi
c
h
a
r
e
ne
gl
i
gi
bl
e
w
he
n
t
h
e
c
on
s
e
q
ue
nc
e
of
t
he
D
D
oS
a
t
t
a
c
ks
i
s
c
on
s
i
d
e
r
e
d.
T
he
g
r
a
p
h
i
n
F
i
g
ur
e
9,
c
l
e
a
r
ly
de
pi
c
t
s
t
he
t
im
e
t
a
ke
n
f
o
r
t
he
pr
o
po
s
e
d
m
et
ho
d.
Fig
ure
9. Com
par
is
on of
e
xec
ution t
i
m
e for pr
e
ve
ntion m
eth
ods
A
m
on
g
t
he
t
hr
e
e
pr
e
ve
nt
i
on
m
e
c
ha
ni
sm
AD
-
I
M
G
C
O
P
t
a
ke
s
m
or
e
t
im
e
a
nd
t
h
us
i
t
i
s
a
l
on
e
c
om
pa
r
e
d
w
i
t
h
ot
he
r
s
t
r
a
t
e
gi
e
s
s
uc
h
a
s
r
e
C
A
P
T
C
H
A
a
nd
im
a
ge
r
eC
A
P
T
C
H
A
.
F
r
om
Fi
gu
r
e
10
,
i
t
i
s
cl
e
a
r
t
ha
t
t
he
e
xe
c
ut
i
on
t
im
e
i
s
m
i
ni
m
al
w
he
n
c
om
pa
r
e
d
w
i
t
h
t
he
ot
he
r
e
xi
s
t
i
ng
t
e
c
hn
i
qu
e
s
.
A
l
s
o,
t
h
e
pr
o
po
s
e
d
m
et
ho
ds
r
e
qu
i
r
e
l
e
s
s
s
t
or
a
ge
s
pa
c
e
w
he
n
c
om
pa
r
e
d
w
i
t
h
t
he
ot
he
r
s
a
nd
i
t
i
s
di
s
c
us
s
e
d
a
t
t
he
e
nd
o
f
t
he
s
e
c
t
i
on
.
T
hu
s
i
f
no
C
A
P
T
C
H
A
i
s
no
t
t
oo
s
ur
e
,
t
he
n
t
he
pr
op
os
e
d
m
et
ho
d
c
a
n
be
us
e
d
i
ns
t
e
a
d
of
ot
he
r
e
xi
s
t
i
ng
t
e
c
hn
i
q
ue
s
.
Fig
ure
10. C
om
par
ison
of
ex
ecuti
on
ti
m
e fo
r pr
e
ven
ti
on m
et
hod wit
h
e
xis
ti
ng
m
et
ho
ds
T
o f
i
nd
ou
t
t
he
pe
r
f
o
r
m
a
nc
e
o
f
t
he
pr
oc
e
s
s
i
n
g q
ue
ue
,
t
he
a
n
a
l
y
si
s
ha
s
be
e
n
m
a
de
by
c
ha
ng
i
ng
t
he
s
i
z
e
of
t
he
q
ue
ue
a
nd
t
he
a
ve
r
a
ge
w
a
i
t
i
ng
t
im
e
of
t
he
s
y
s
t
em
i
s
c
a
l
c
ul
a
te
d.
T
he
A
r
r
i
va
l
r
a
t
e
of
pa
c
ke
t
s
i
s
λ
=
9
pa
c
ke
t
s
/m
i
n.
T
he
a
ve
r
a
ge
s
e
r
vi
c
e
t
im
e
f
or
a
s
i
ng
l
e
pa
c
ke
t
i
s
10
s
e
c
o
nd
s
.
T
he
n
t
he
s
e
r
vi
c
e
r
a
t
e
μ
=
1/
1
0
pa
c
ke
t
s
/
s
e
c
=
6
pa
c
ke
t
s
/m
i
n.
t
hu
s
t
he
U
t
i
l
i
z
at
io
n
va
l
ue
ρ
=
9/
6
=
1.
5.
T
he
c
a
l
c
ul
a
t
i
on
of
a
n
um
be
r
of
r
e
qu
e
s
t
s
i
n
t
he
qu
e
ue
,
w
a
i
t
i
ng
t
im
e
i
n
t
he
qu
e
ue
a
nd
t
he
t
im
e
t
a
ke
n
t
o
c
o
m
pl
e
te
t
he
r
e
qu
e
s
t
a
r
e
gi
ve
n
i
n
T
a
bl
e
2.
Table
2.
Wait
ing t
i
m
e and com
ple
ti
on
ti
m
e
cal
culat
ion
for vari
ou
s
que
ue si
ze
Qu
eu
e Size
Ef
f
ectiv
e ar
rival r
a
te
(pack
ets/
m
in
)
No
of
r
eq
u
ests
waitin
g
in q
u
eu
e
Ti
m
e
to co
m
p
lete
th
e service (
in
m
in
)
W
aitin
g
ti
m
e in th
e
q
u
eu
e
(in
m
in
)
5
5
.8
3
.4
0
.76
0
.59
10
5
.9
8
1
.5
1
.3
15
5
.9
13
2
.3
2
.1
20
6
18
3
.1
3
25
6
23
4
3
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
5
,
Oct
ober
20
19
:
4163
-
4175
4172
T
hu
s
w
he
n
t
he
qu
e
ue
s
i
z
e
i
s
s
m
a
ll
,
t
he
pe
r
f
o
r
m
a
nc
e
of
t
he
s
y
s
t
em
i
s
be
tt
er
,
s
i
nc
e
t
he
w
a
i
t
i
ng
t
im
e
i
s
sm
a
ll
f
or
sm
al
l
qu
e
ue
s
t
ha
n
t
he
l
on
g
qu
e
ue
s
.
T
he
pe
r
f
or
m
a
nc
e
a
na
l
y
si
s
i
s
sho
w
n
i
n
F
i
gu
r
e
11
.
T
he
qu
e
ue
s
i
z
e
i
s
f
i
xe
d
a
t
10
.
T
he
pr
op
os
e
d
f
r
am
e
w
or
k
ha
s
a
c
a
pa
bi
l
i
ty
t
o
de
t
e
c
t
a
bo
ve
9
5%
of
a
t
t
a
c
ks
.
H
o
w
e
ve
r
,
w
he
n
t
he
t
hr
e
s
ho
l
d
i
s
de
c
r
e
a
s
e
d,
t
he
a
cc
ur
a
c
y
i
s
i
nc
r
e
a
s
e
d
e
ve
n
be
t
te
r
.
T
hi
s
i
s
s
ho
w
n
i
n
T
a
bl
e
3.
T
he
t
a
bl
e
s
ho
w
s
t
he
a
c
c
ur
a
c
y
a
nd
e
r
r
o
r
r
a
t
e
w
i
t
h
t
he
t
hr
e
s
h
ol
d
va
l
ue
be
t
w
e
e
n
5
a
nd
10
.
Fig
ure
11. Per
f
or
m
ance
analy
sis of the
que
ue
w
it
h va
ried
si
ze
Table
3.
Wait
ing t
i
m
e and com
ple
ti
on
ti
m
e
cal
culat
ion
for vari
ou
s
que
ue si
ze
Thresh
o
ld
Value
Co
rr
ectly
classif
ie
d
attacks
Inco
rr
ectl
y
cl
ass
if
ied
attacks
10
95
5
9
9
5
.4
4
.6
8
9
6
.2
3
.8
7
97
3
6
9
7
.8
2
.2
5
9
8
.6
1
.4
T
he
a
c
c
ur
a
c
y
a
nd
e
r
r
o
r
r
a
t
e
f
or
t
he
a
b
ov
e
im
pl
em
e
nt
at
i
on
i
s
s
h
ow
n
gr
a
ph
i
c
a
l
l
y
us
i
ng
t
he
c
ha
r
t
i
n
F
i
gu
r
e
12
.
A
l
s
o,
t
he
pr
op
os
e
d
m
e
t
ho
d
i
s
a
na
ly
z
ed
w
i
t
h
m
em
or
y
s
pa
c
e
a
s
a
no
t
he
r
pa
r
am
et
e
r
.
F
or
V
I
S
U
A
L
C
O
M
,
s
i
ng
l
e
im
a
ge
w
i
t
h
5
qu
e
s
t
i
on
s
f
or
e
a
c
h
im
a
ge
i
s
t
a
ke
n
i
nt
o
a
c
c
ou
nt
w
hi
c
h
c
a
n
s
e
r
ve
5
us
e
r
s
.
Fig
ure
12. Res
ult
analy
sis wit
h varyin
g
t
hr
es
ho
l
d values
S
im
i
l
a
r
ly
,
I
M
G
C
O
P
s
t
or
e
s
a
c
om
pl
et
e
im
a
ge
w
i
t
h
i
t
s
pa
r
t
.
I
n
o
ur
e
x
pe
r
i
m
en
t
,
e
a
c
h
im
a
ge
i
s
s
pl
i
t
i
nt
o
6
pa
r
t
s
a
nd
gi
v
e
n
a
s
a
c
ha
ll
e
ng
e
t
o
t
he
c
li
e
nt
.
H
ow
e
ve
r
,
a
s
i
ng
l
e
im
a
ge
ca
n
s
e
r
ve
up
t
o
15
us
e
r
s
w
i
t
h
di
f
f
e
r
e
nt
c
om
bi
na
t
i
on
s
.
T
he
t
hi
r
d
m
et
ho
d
A
D
-
I
M
G
C
O
P
ne
e
ds
t
h
e
s
am
e
m
em
ory
a
s
I
M
G
C
O
P
a
l
on
g
w
i
t
h
sm
a
ll
a
dd
i
t
i
on
a
l
m
em
or
y
t
o
s
t
or
e
th
e
a
no
m
a
ly
im
a
ge
a
s
w
e
l
l
.
T
he
e
xi
s
t
i
ng
m
et
ho
d
us
e
d
f
or
c
om
pa
r
i
s
on
i
s
t
ha
t
a
n
im
a
ge
r
e
C
A
P
T
C
H
A
i
n
w
hi
c
h
a
g
r
o
up
of
i
m
a
ge
th
a
t
i
s
s
e
l
e
c
te
d
ba
s
e
d
o
n
t
he
gi
ve
n
c
h
a
l
l
e
ng
e
.
T
hi
s
e
xp
e
r
i
m
en
t
a
l
r
e
s
ul
t
i
s
s
how
n
i
n
T
a
bl
e
4
.
F
i
g
ur
e
13
s
ho
w
s
t
he
gr
a
ph
i
c
a
l
r
e
pr
e
s
e
nt
a
t
i
on
of
t
he
c
om
pa
r
i
s
on
.
Evaluation Warning : The document was created with Spire.PDF for Python.