Beha io al Pa e n Analysis o Secu e Mig a ion and Communica ions in
eComme ce using C yp og aphic P o ocols on a Mobile MAS Pla o m
S. Pozo, R. M. Gasca, R. Ceballos
Compu e Languages and Sys ems Depa men
ETS Ingenie ía In o má ica, Uni e si y o Se ille
A da. Reina Me cedes S/N, 41012 Se ille, Spain
[email p o ec ed], {ceballos,gasca}@lsi.us.es
Abs ac
Mobile Mul i-Agen Sys ems (MAS) sys ems can be
used wi h eal success in a g owing numbe o
eComme ce applica ions nowadays. Secu i y has been
iden i ied as nume ous imes by di e en esea che s
as a op c i e ion o he accep ance o mobile agen
adop ion. In his pape we p esen an in-dep h analysis
o beha io pa e ns o a mobile MAS pla o m when
using di e en c yp og aphic p o ocols o assu e
communica ion and mig a ion in eg i y and
con iden iali y. Di e en use case scene ies o
eComme ce applica ions as well as many o he aspec s
ha e been s udied, such as o e head, di e en
communica ion pa e ns, di e en loads and
bandwid h issues. This wo k is also ex ensible o o he
mobile and non-mobile MAS pla o ms. The esul s
ob ained can be used and should be aken in o accoun
by designe s and implemen e s o secu e mobile and
also non-mobile agen pla o ms and agen s.
1. In oduc ion
Mobile Mul i-Agen Sys ems (MAS) can be used
wi h eal success in a g owing numbe o eComme ce
applica ions nowadays. Fo example, he e a e a lo o
esea ch e o s in agen coope a ion and nego ia ion
p o ocols and hei applica ions, such as coope a ing
agen s in p oblem esolu ion and nego ia ing agen s in
eComme ce. MAS a e also being used wi h g ea
success in Tele-Assis ance pla o ms o elde ly ca e
[21, 22].
Mos nego ia ion and coope a ion p o ocols ha a e
widely deployed o e big ne wo ks o o e In e ne
ha e been used wi h mobile MAS. Secu i y has been
iden i ied nume ous imes by di e en esea che s as a
op c i e ion o he accep ance o mobile agen
adop ion [1]. The e a e e en many domains o
applica ion which ha e speci ic secu i y equi emen s
imposed by legisla ion, like he applica ions used in
heal h ca e and eComme ce en i onmen s. The e is
also a ade o be ween us and secu i y, and he
o e head c ea ed by c yp og aphic p o ocols ha
should be aken in o accoun when designing
eComme ce applica ions, since i can a ec se e ely i s
use ulness.
In his pape we p esen an in-dep h s udy o
beha io pa e ns o a mobile MAS pla o m when
using c yp og aphic p o ocols o assu e
communica ion and mig a ion in eg i y and
con iden iali y. Di e en eComme ce use case
scene ies as well as many o he aspec s ha e been
s udied, such as o e head, di e en communica ion
pa e ns, di e en loads, and bandwid h issues. This
wo k, as we a e going o expose below, is also
ex ensible o o he mobile and in some cases non-
mobile MAS pla o ms. The es sys em has been
implemen ed om a p e ious wo k [11], which is
based on he possibili y o using secu e unnels o
mobile MAS.
The es o he pape is o ganized as ollows: In
sec ion 2 we p esen a b ie in oduc ion o secu i y
p oblems o pla o ms and mobile agen s, and ela ed
wo ks. In sec ion 3 we expose an o e iew o a secu e
unnel app oach in which we ha e based ou es
sys em. In sec ion 4 we p opose ou es ing
me hodology and desc ibe he es ing en i onmen ,
and p esen he esul s ob ained, discussions, and
conclusions abou hem. We conclude his wo k in
chap e 5 wi h gene al conclusions and u u e esea ch
wo ks sugges ed du ing his wo k.
2. Mobili y and secu i y issues in mobile
MAS. Rela ed wo ks
Mobile so wa e agen s a e goal-di ec ed p og ams
capable o suspending hei execu ion s a e on one
pla o m and mo ing o ano he , whe e hey esume
hei execu ion. S a ic agen s ha e he same
cha ac e is ics excep ha hey a e no mobile, i.e. hey
canno mo e om he sys em hey a e unning on
(which is also he sys em which c ea ed hem.)
The e a e di e en kinds o mobili y: s ong, in
which code and execu ion s a e a e mo ed; and weak,
whe e only he code is mo ed. Tha is, e e y ime he
agen a i es a a di e en des ina ion, i uns back
om he beginning, o om a ma ke exp essly placed
o his pu pose.
Mobili y can also be classi ied as single hop and
mul i-hop. Mos secu i y p oblems which a ise in
mobile mul i-agen pa adigm occu wi h mul i-hop
mobile agen s. This is due o he ac ha only he
home pla o m ( he pla o m whe e he agen was i s
c ea ed) is us ed. Tha is, i we a e no su e ha a
mobile agen comes om a us ed sou ce, on which
we know ha nei he he code no he s a e ha e been
al e ed, we canno know whe he his agen will ca y
ou an unplanned ac i i y on he des ina ion pla o m.
The mos secu e loca ion o an agen is i s home
pla o m. Al hough nei he agen s no home pla o ms
a e in ulne able, a numbe o con en ional echniques
can be applied o cons uc adequa e de enses. Each
ime an agen mig a es, secu i y isks a ise, and so i is
needed a way o ansmi his us ed en i onmen o
o he pla o ms whe e agen s may a el. The g ea es
p oblem wi h mul i-hop MAS is jus he us
ela ionship which can be es ablished in single-hop
MAS be ween wo pla o ms hanks o he secu i y
mechanisms de i ed om Clien /Se e a chi ec u e.
These us ela ionships a e no ansi i e, no ha e o
be bila e al.
The e a e some p ac ical solu ions o secu ing
communica ions and mig a ion in mul i-hop MAS, bu
he as majo i y o hem include es ic ions on
i ine a ies. Fo example, i is possible o se up a
es ic ed i ine a y in which all pla o ms ha e mu ual
us in all o he s, so agen s can mo e eely be ween
hem (each pla o m signs and/o enc yp s he agen
be o e i mig a es).
In an ea lie wo k [11], we p oposed a di e en
way o secu e agen s based on secu e unnels. We
ob ained a secu e agen communi y usable in
p oduc ion eComme ce, heal h ca e and Tele-
Assis ance en i onmen s wi h exis ing echnology,
wi h li le impac in ne wo k opology, and wi hou
ha ing o modi y exis ing agen applica ions o agen
pla o ms. Speci ically we we e looking o a sys em
ha p o ide in eg i y, con iden iali y, da a o igin
au hen ica ion, MAS independence, compliance wi h
s anda ds, exis ence o c yp og aphic accele a ion
ha dwa e, cos -sa ing and euse o exis ing so wa e
and ha dwa e in as uc u e. Only communica ions and
mig a ion we e p o ec ed. No di ec p o ec ion o
malicious agen s o pla o ms was p o ided, as he e
a e o he echniques o ha pu poses, such as aul
isola ion [2], sa e code in e p e a ion [20], s a e
app aisal [3], pa h his o ies [4, 5], p oo ca ying code
[6], execu ion acing [7], en i onmen al key
gene a ion [8], compu ing wi h enc yp ed unc ions
[9], code ob usca ion [10], and o he s.
Agle s [13] ha e been used o ou e e ence
implemen a ion in he labo a o y, since i is Open
Sou ce, c oss-pla o m, easy o de elop, has weak
mobili y ( he maximum mobili y a Ja a-coded MAS
could ha e), high accep ance and ela i ely good
documen a ion [14].
We ha e ocused ou esea ch in he
pa ame e iza ion o beha io pa e ns o secu ed
mobile agen s in eComme ce: beha io wi h di e en
agen loads, beha io modi ica ion by he use o
di e en comp ession algo i hms, beha io wi h
di e en communica ion pa e ns… concluding wi h a
pe o mance analysis. O he issues ela ed o he use o
ou pa icula app oach o secu ing communica ions
and mig a ion in mobile MAS ha e a isen, such as
scalabili y, NAT and p oxying issues. All o hese
colla e al issues, as well as an in-dep h s udy o he
applicabili y o he c yp og aphic p o ocols used in his
wo k, ha e been discussed in mo e de ail in a p e ious
wo k [11].
The conclusions de i ed om his pape a e easily
applicable o o he mobile MAS pla o ms, as ou
secu i y mechanism [11] is applicable in a wide a ie y
o hem wi hou ha ing o modi y pla o ms o agen s
hemsel es. Mo eo e , he c yp og aphic p o ocols
used a e based on s anda ds, and can easily be
encoun e ed implemen ed a pla o m le el in many
mobile MAS pla o ms.
3. O e iew o secu e unnels
Tunneling is he capabili y o encapsula ing one
p o ocol wi hin ano he , using his second p o ocol o
a e se ne wo k nodes. A secu e unnel encapsula es
an insecu e p o ocol (like FTP o HTTP) wi hin a
secu e one (like SSL [19] o TLS [15]). Tunnels may
also be used o bypass i ewalls, and a e also
ulne able o denial o se ice a acks, since hey use a
public and un us ed ne wo k as ansmission media.
We a e going o gi e a e y b ie explana ion o
c yp osys ems applied o mobile MAS pla o m,
because an in-dep h s udy is gi en in [11], whe e we
de eloped a us ed agen communi y wi h applica ions
which gene a e applica ion independen secu e unnels.
• S unnel [16] is an applica ion which ac s as SSL 3
se e and/o clien , p o iding a secu e SSL-based
secu e unnel (w appe ) o insecu e p o ocols o
applica ions wi h he only need o he ins alla ion o
he applica ion in each o he sys ems ha needs o
secu e. S unnel is dis ibu ed unde a GPL license
and has e sions o Mic oso Windows, some
la o s o UNIX and many o he OS. S unnel also
suppo s c yp og aphic accele a o ha dwa e and
clien and se e au hen ica ion wi h X.509 digi al
ce i ica es (as SSL does).
• SSH [17] is a p o ocol o secu ing ne wo k se ices
o e an insecu e ne wo k. I is adi ionally used o
p o ec ing insecu e UNIX p o ocols such as elne ,
login, e c. Mo eo e SSH can be used o secu e
o he se ices c ea ing a w appe a ound hem using
a local po edi ec ion scheme e y simila o he
one used by S unnel. SSH is widely accep ed by he
scien i ic communi y as being a us ed secu i y
p o ocol. SSH a chi ec u e is e y simila o SSL o
TLS one, and p o ides basically he same
unc ionali y.
• Zebedee [18] is ano he Open Sou ce applica ion
used o c ea e secu e unnels wi h implemen a ions
in Windows, UNIX, Linux, Ja a and Ruby.
Zebedee ( om i s documen a ion) has a small
memo y oo p in and low wi e p o ocol o e head.
Zebedee uses a plain Di ie-Hellman p o ocol o i s
(weak) key ag eemen p ocess and a symme ic key
c yp og aphic algo i hm, Blow ish. Zebedee does
no p o ide any ea u es o da a in eg i y. We ha e
included his applica ion in ou es s o e i y he
claims abou he o e head compa ed o o he
p o ocols.
4. Beha io pa e ns o c yp og aphic
p o ocols in eComme ce applica ions
Di e en beha io pa e ns ha e been analyzed,
such as
• Beha io in single-hop and mul i-hop wi h di e en
mig a ion and communica ion pa e ns
• Beha io wi h di e en agen loads
• Beha io modi ica ion by he use o di e en
comp ession algo i hms (be o e enc yp ion phase)
We ha e also done a comple e pe o mance
analysis, since pe o mance could be a disc imina o y
ac o in se e al en i onmen s, and mo e especially in
eComme ce. Mo eo e , some issues ha e also been
de ec ed du ing he es s, as we a e going o explain
below.
4.1. En i onmen
Du ing ou expe imen s he e ha e been used wo
compu e s connec ed be ween a dedica ed 100Mbi
E he ne swi ch isola ed om he es o he ne wo k.
All esul s ha e been aken when he machines we e
ully dedica ed and wi h a esh ins alla ion o he
ope a ing sys em. Bo h machines a e Pen ium 4
1,7GHz p ocesso wi h a 400MHz FSB and 256Mb
DDR266 RAM. Bo h we e unning SuSE Linux 9.0
ully pa ched (a he ime o he w i ing o his pape )
and compiled o i586 a chi ec u e. Linux Ke nel
2.4.21-99, JDK 1.4.2b28 and Agle s 2.0.2 we e he
base so wa e used.
• Agle s 2.0.2. Agle s HTTP p oxy ea u e ha e been
used o o e come a unnel dodging p oblem [11], as
we hink his is he way Agle s a e going o be used
in a eal en i onmen . No embedded pla o m
au hen ica ion has been used in he es s.
• OpenSSL 0.9.7b. This lib a y p o ides algo i hms
used by o he applica ions and p o ocols, in his
case S unnel and OpenSSH. De aul con igu a ion
was used.
• S unnel 4.04 PTHREAD. 128-bi AES enc yp ion
ha e been used ins ead o 3DES, since AES is as e
han 3DES and 3DES is also being eplaced wi h
AES. 1024-bi RSA ce i ica e based au hen ica ion
scheme and no comp ession ha e also been used,
since S unnel does no suppo i , as i uses SSL.
S unnel was a bi di icul o con igu e because o
i s lack o up o da e documen a ion.
• OpenSSH 3.7.2.1p2-18. 128-bi AES enc yp ion,
1024-bi RSA ce i ica e based au hen ica ion and
no comp ession ha e been used. We go some
s abili y p oblems when ans e ing big agen s
(bigge han 3Mb). We will explain mo e abou i
la e on his chap e , as well as beha io changes
when using comp ession.
• Zebedee 2.5.2 wi h libBlow ish 0.9.5a, libBzip2
1.0.1 and zlib 1.1.4. 128-bi Blow ish and SHA1
hashes (160 bi ) ha e been used. Also comp ession
has been deac i a ed. No embedded au hen ica ion
mechanism has been used because o unusabili y
easons explained be o e in his pape and in o he
wo ks [11].
4.2. Me hodology and pa ame e s
Nine a ia ions o he same es applica ion ha e
been used, a ying in he numbe o hops he agen
a e ses, and in he load he agen ca ies on. The
applica ion is composed o a mobile agen ha
mig a es o a pla o m and hen goes back o i s home
pla o m. The agen does no compu e any hing a any
pla o ms.
Fo beha io ecogni ion wi h di e en
communica ion pa e ns, he numbe o i ine a y laps
pe o med by he agen changes be ween 1, 10 and
100. Tha is, i he numbe o i ine a ies is se o 10,
hen he agen goes o he des ina ion and hen o i s
home pla o m en imes.
Fo beha io ecogni ion wi h di e en loads, he
agen was loaded wi h h ee di e en da a ec o s
(0Kb, 100Kb o 1Mb) using OpenO ice 1.1
sw645mi.dll ile om i s Mic oso Windows e sion,
in o de o ge some deg ee o andomness in da a
con en . The size o he compiled Agle i sel was
3,7Kb.
Comp ession was also used o es beha io changes
and o e head incu ed on i s usage, as i is common o
use comp ession on e y low bandwid h links
( ypically used in B2C eComme ce applica ions). Two
comp ession algo i hms (Zlib and BZip2) we e used in
o de o es he ela ion be ween di e en
communica ion pa e ns, agen loads, and
c yp og aphic p o ocols.
Special in e es has been pu in a compa a i e
analysis o each beha io pa e n de ec ed and i s
de i a es. Each es has been epea ed h ee imes. The
execu ion ime p esen ed in he pe o mance analysis is
he a e age o he h ee esul s. We hink ha hese
es s co e he majo i y o si ua ions p esen in B2C
and B2B eComme ce applica ions.
4.3. Resul s and discussion
In his sec ion we p esen he h ee main use case
eComme ce scene ies which ha e been conduc ed,
showing om only one i ine a y o a hund ed o hem,
wi h esul s o di e en agen loads. Finally beha io
changes due o he use o comp ession a e also ci ed.
Wi h hese es s nea all possibili ies ega ding B2C
and B2B communica ions a e co e ed, since mig a ion
is a o m o communica ion in which an agen can do
compu a ions a des ina ion pla o ms. Since ou es
agen s do no compu e any hing, a mig a ion can be
hough as a message in e ms o communica ion ime,
adding he cons an ime aken o se ialize and
eins an ia e he agen . Also he size o an unloaded
agen (3,7Kb in ou unloaded es agle ) could be
simila o a FIPA-ACL small message, since i uses a
complex syn ax, and da a can be embedded in o
messages. Tes ing a ange o possible leng hs in he
0Kb-3Kb ange is i ele an as is going o be exposed
below, since he di e ences in beha io a e oo small
o be aken in o accoun . Again, we mus no e ha he
esul s p esen ed in his sec ion a e di ec ly applicable
o o he agen pla o ms, including pla o ms wi h he
c yp og aphic p o ocols ci ed below embedded in
hem.
4.3.1. Beha io depending on agen communica ion
and mig a ion pa e ns. We e alua ed he beha io
o an agle mig a ing only om one pla o m o
ano he and i s way back o he home pla o m. Then
he agle disposed i sel . Then we es ed all
c yp og aphic p o ocols explained in an ea lie sec ion
(Figu e 1) o his beha io pa e n ( he agle in clea ,
S unnel, SSH and Zebedee). Then a en hop pa e n
mig a ion was used, and inally a hund ed hop
mig a ion pa e n. No comp ession was used, and he
agle was no loaded wi h any ex a da a (bu he size
o i s compiled code was 3,7Kb).
Fo one i ine a y, di e ences a e oo small o be
ep esen a i e, bu some in e es ing esul s mus be
no ed. The e is a minimal di e ence be ween he
beha io s o he agle using no c yp og aphic p o ocol
and he one unneled by S unnel (SSL). Bu he ime
aken by SSH is nea ly wice he ime aken by he
clea agle o mig a e and come back. Finally, he ime
aken by a Zebedee unneled agle is h ee imes he
one o he clea agle . This basic es has gi en a sigh
abou he beha io al changes c ea ed by he o e head
o he di e en p o ocols wi h small loads.
Su p isingly, S unnel (SSL) and SSH (which also used
RSA au hen ica ion) c ea ed less o e head han
Zebedee wi h no au hen ica ion. Fo he en and a
hund ed i ine a ies es s, hese esul s can be seen wi h
g ea de ail, as o e head di e ences a e mo e
p onounced.
As i can be seen wi h mo e de ail in he able
(Figu e 1), he e is no di e ence in p ac ical e ms
be ween beha io o a clea ex agle and o a SSH o
S unnel ones as he numbe o i ine a ies inc eases. I
can be hough ha compu e s used in he es we e
powe ul enough o enc yp he mig a ing agle wi h no
pe o mance hi . Bu i ha we e ue, hen Zebedee
esul s would be di e en , wi h less o e head. Clea ly
Zebedee uses a poo designed p o ocol. The
di e ences be ween Zebedee and he o he wo ones
a e oo big. These di e ences a e no ela ed o he
small di e ences in pe o mance o AES (used by
S unnel and SSH) and Blow ish (Blow ish is used by
Zebedee, being only a bi slowe han Two ish),
acco ding o a wide ange o s udied cases [12, 23] in
32 bi pla o ms wi h C coded algo i hms.
Figu e 1. Beha io depending on he numbe
o i ine a ies ( ime in milliseconds)
4.3.2. Beha io depending on agen load pa e ns.
We e alua ed he beha io o an agle wi h di e en
loads. We es ed i wi h no load (3,7Kb was he size o
compiled code), 100Kb and 1Mb. We es ed all
c yp og aphic p o ocols explained in an ea lie sec ion.
Th ee scene ies we e used in o de o gi e a
compa a i e analysis o di e ences in beha io due o
di e en communica ion and mig a ion pa e ns
desc ibed in he p e ious sec ion, wi h one i ine a y,
en and a hund ed ones (Figu es 2, 3, and 4).
The i s scene y (Figu e 2) ep esen s he one
i ine a y beha io pa e n s udied in he p e ious
sec ion, bu wi h di e en loads. Fo no ex a load, he
esul s a e equal o he ones o he p e ious sec ion o
one i ine a y.
When he agen was loaded wi h 100Kb, he
p o ocol o e head becomes mo e e iden due o he
ac ha , e en wi h mo e da a o enc yp , he mig a ion
imes ha e no inc eased oo much. This can be seen
wi h mo e de ail in SSH es s, whe e he beha io s o
0Kb and 100Kb agle ha e minimal di e ences
(301ms Vs. 314ms). Fo he es o he p o ocols,
beha io di e ences a e mo e e iden . Bu o all o
hem, beha io di e ences be ween unloaded agle s
and 100Kb loaded ones a e minimal. A load inc ease is
necessa y o know abou beha io al changes due o
enc yp ion o e head a he han p o ocol o e head.
Figu e 2. Beha io depending on agen load.
One i ine a y ( ime in milliseconds)
I is di icul o hink ha ypical eComme ce agen
applica ions would use big messages o big pieces o
code because o he na u e o mobile agen pa adigm
and eComme ce applica ions, bu we ha e also s udied
a 1Mb loaded agle . In his case, he o e head is
c ea ed by he enc yp ion algo i hm and no by he
p o ocol o e head. The e is a se e e beha io al change
be ween a 100Kb agle and a 1Mb one in all cases.
This di e ence in quan i a i e e ms is a mul iplica i e
ac o o app oxima ely 10 in ime, which is consis en
wi h he inc ease in load ( om 100Kb o 1Mb).
Mo eo e he e is an impo an ac o no e: a 1Mb
load, he imes be ween di e en c yp og aphic
p o ocols a e minimal (including he agle used wi h no
1 10 100
Clea 174 667 4792
S unnel 210 974 6231
SSH 301 845 5008
Zebedee 540 2420 21108
0Kb 100Kb 1Mb
Clea 174 225 3312
S unnel 210 319 3512
SSH 301 314 3522
Zebedee 540 633 3520

enc yp ion), like i i we e a pe o mance bo leneck.
This bo leneck could be due o many ex e nal, non
pla o m and non algo i hm ela ed ac o s, like
memo y bandwid h limi a ion, da a oo big o i in
cache, ansmission media limi a ion, e c.
The second scene y (Figu e 3) ep esen s he en
hop i ine a y pa e n s udied in he p e ious sec ion,
bu wi h di e en loads. Fo no ex a load, he esul s
a e equal o he ones o he p e ious sec ion o en
i ine a ies.
The beha io o hese agle s is e y simila o he
ones o Figu es 1 and 2 bu in a di e en , highe scale.
E en he beha io al di e ences be ween S unnel and
SSH o one and en i ine a ies a e he same o he
same numbe o i ine a ies, wi h an unloaded agle and
wi h a 100Kb agle . No e ha a load inc ease does no
change beha io o he agle s in o he way han he
inc ease in i ine a ies did. Fo a hund ed o i ine a ies
(Figu e 4), he esul s a e e en he same, bu again a a
bigge scale.
Figu e 3. Beha io depending on agen load.
Ten i ine a ies ( ime in milliseconds)
Figu e 4. Beha io depending on agen load. A
hund ed o i ine a ies ( ime in milliseconds)
Figu e 5. De ailed iew o SSH p o ocol
beha io ( ime in milliseconds)
The pe o mance bo leneck also emains he e o
1Mb load, bu a a highe le el. The jump be ween
100Kb and 1Mb is oo big o analyze wi h de ail he
causes o he hea y beha io al change. Because o
ha , some es s wi h agle s loaded in he ange 128Kb-
512Kb we e conduc ed. We ha e es ed his pa e n
(Figu e 5) wi h SSH, as his has been he p o ocol wi h
a mo e s able beha io and wi h be e pe o mance
om all o he s udied ones.
0Kb 100Kb 1Mb
Clea 667 1155 17921
S unnel 974 2168 18249
SSH 845 1660 18125
Zebedee 2420 2857 17896
0Kb 100Kb 1Mb
Clea 4792 9709 151796
S unnel 6231 13431 170393
SSH 5008 11961 163912
Zebedee 21108 30341 165208
0Kb 845
100Kb 1660
512Kb 11309
1Mb 18125
2Mb 22629
We disco e ed ha only om 512Kb o highe ones
p oduces he se e e beha io al change, a scale change
o a highe one (no e he 2Mb es ). Fo loads ewe
han 512Kb only easonable changes ake place. In ou
opinion his con i ms ha he pe o mance bo leneck
should be due o ex e nal, non pla o m and non
c yp og aphic algo i hm ela ed ac o s, like memo y
bandwid h limi a ion, da a oo big o i in cache,
ansmission media limi a ion, e c. Ano he scale
change would be possible a e en highe han 2Mb
loads, bu we ha e no es ed i .
4.3.3. Compa a i e beha io analysis depending on
mig a ion and load pa e ns. A g aphical
compa a i e analysis (Fig. 6) is necessa y o no e he
beha io al changes be ween he di e en s udied
beha io pa e ns. When looking a he compa a i e
g aphic, ep esen ing one, en o a hund ed i ine a ies,
wi h di e en load and c yp og aphic p o ocols in
hem, he h ee g aphics appea s o be equal.
Ne e heless he ime scale is di e en . When
inc easing he numbe o i ine a ies and agle load a
he same ime, he di e ences in execu ion ime
be ween he ciphe ed agle s and he plain ones widen.
An inc ease in i ine a y numbe is also mo e c i ical
han an inc ease in e ms o agen size, excep o
eally big agen s, whe e he pe o mance bo leneck
a ises.
4.3.4. Beha io al changes when using comp ession.
The use o no o comp ession is a con o e sial issue,
because some c yp og aphic p o ocols de ine
comp ession as a desi able ea u e (SSL, SSH) a he
han a equi ed one. O cou se, in e ope abili y is
gua an eed du ing he nego ia ion phase be ween
pee s. The use o comp ession is an in e es ing issue o
es , since comp ession is usually used in e y low
bandwid h links, as elephone lines o cell phones.
In ou es s (wi h a high bandwid h link)
comp ession ha e been ac i a ed and deac i a ed o
Figu e 6. Compa a i e analysis o all s udied
beha io pa e ns
es i s impac in all he s udied pa e ns. The esul s
a e no con incing, wi h be e esul s in some cases
and wo se in o he s. Since he use o comp ession
canno be changed dynamically, his issue is no easy
o sol e no he ocus o his wo k. Only SSH (Zlib)
and Zebedee (BZip2) we e es ed using comp ession,
as S unnel does no suppo comp ession.
• One i ine a y. Pe o mance only ge s be e i agle
size is less han 100Kb. I he agle is loaded wi h
mo e da a, pe o mance ge s wo se.
• Ten i ine a ies. Only SSH imp o es pe o mance
wi h he use o comp ession, bu in e y pa icula
cases. SSH can use a wide ange o symme ic key
c yp osys ems. The wo as e ones a e AES and
Blow ish. Pe o mance only imp o es when using
Blow ish and loads unde 100Kb. Blow ish is bi a
slowe algo i hm han AES, so he pe o mance
imp o emen could be due o he ac ha Blow ish
has o enc yp less da a because i was comp essed.
This should be ue i he comp ession phase is
as e han he enc yp ion one, as usually is, and
only i he comp ession a io is enough o jus i y he
ime spen in ha phase, as he e is no way o know
dynamically how well o bad a ile is going o
comp ess (bu we can es ima e i ). Ne e heless, he
same ac should apply o AES: wi h he use o
comp ession ansmission ime should also imp o e.
Bu his is no he case, as comp ession only
imp o es ansmission ime using Blow ish. This is
a e y s ange issue we ha e no sol ed.
• A hund ed i ine a ies. Pe o mance always go
wo se esul s in any case.
We hink ha he e a e enough easons no o use
comp ession wi h any c yp og aphic p o ocol in a mid
o high bandwid h link, mainly due o he e ogenei y
issues be ween di e en p o ocols and lack o s abili y
a di e en si ua ions. Beha io es ing o agen s wi h
a e y low bandwid h equi es a o ally new analysis,
since he e a e a lo o new a iables o ake in o
accoun : sys em a chi ec u e (cell phone, PDA, e c.),
memo y cons ain s, p og amming language design
and implemen a ion issues, OS cons ain s, line
quali y, e c.
4.3.5. Conclusions abou eComme ce beha io
pa e n analysis. Du ing his in-dep h analysis and
expe imen a ion we ha e assu ed some well known
ac s, such as ha is as e o send a big agen han
sending a high numbe o small agen s ( he same
applies o messages, Figu es 2, 3, 4). O he no so well
know ac s ha e also been disco e ed, such as ha he
di e ence in execu ion ime o a clea agen and a
unneled one sho en when sending a bu s o
messages o doing a bu s o mig a ions (Figu e 2).
F om ou poin o iew, due o s abili y and
pe o mance du ing ou es s wi h di e en pa e ns,
SSH2 is he bes sui ed p o ocol ( om he s udied
ones) o secu ing communica ions and mig a ion in
B2B and B2C eComme ce applica ions. Mo eo e , he
use o c yp og aphic p o ocols does no suppose a
signi ica i e pe o mance penal y. We ecommend
using 128 bi AES (AES ha e a signi ica i e
pe o mance penal y a highe s eng hs) and a leas
1024 bi RSA key o exchange. Table 1 e lec s a
summa y o all he s udied aspec s.
5. Conclusions and u u e wo k
Mobile mul i-agen sys ems in oduce a g ea deal
o secu i y issues. Some o hem a e new and o he s
a e e y simila o clien -se e pa adigm ones.
Secu i y has been iden i ied as nume ous imes by
di e en esea che s as a op c i e ion o he adop ion
o mobile agen s in eal applica ions.
We ha e p esen ed an in-dep h s udy o beha io
pa e s o a mobile MAS pla o m when using
c yp og aphic p o ocols o assu e communica ion and
mig a ion in eg i y and con iden iali y in eComme ce
applica ions. Di e en eComme ce use case scene ies
as well as many o he aspec s ha e been s udied, such
as p o ocol o e head, communica ion pa e ns, load
pa e ns, comp ession use and bandwid h issues,
co e ing a wide ange o eComme ce eal si ua ions.
This wo k is also ex ensible o o he mobile and in
some cases non-mobile MAS pla o ms.
Du ing his in-dep h s udy we ha e assu ed some
well known ac s. O he no so well known ac s ha e
also been disco e ed, such as he pe o mance
bo leneck (and scale change) Vs. bu s -sending o
agen s and messages. F om ou poin o iew, due o
s abili y and pe o mance du ing ou es s wi h
di e en pa e ns, SSH2 is he bes sui ed p o ocol
( om he s udied ones) o secu ing communica ions
and mig a ion. Ano he signi ican conclusion is ha
he use o c yp og aphic p o ocols does no suppose a
signi ica i e pe o mance penal y compa ed wi h no
using hem a all.
The esul s ob ained can be used and should be
aken in o accoun by designe s and implemen e s o
secu e mobile and non-mobile agen pla o ms and
agen s, as hese esul s can also be applied o message
ansmission, because o he implica ions o beha io al
changes c yp og aphic p o ocols, communica ion,
mig a ion, load and comp ession pa e ns ha e in
mobile agen pa adigm.
The use o comp ession is a con o e sial issue,
since we ha e ob ained e y in e es ing, and in some
cases con adic o y esul s. These esul s sugges ed us
ha ano he in-dep h analysis o beha io al changes in
cons ained de ices wi h e y low bandwid h (and low
quali y in some imes) links is necessa y, as MAS
pla o ms a e being de eloped o cons ained de ices,
wi h cons ained memo y, p ocessing powe and
ene gy consump ion. Beha io al analysis o hese
mobile MAS pla o ms a e a e y in e es ing opic o
u u e esea ch, since adi ional public key
c yp og aphy (i.e. based on ac o iza ion p oblems)
canno be used due i s high compu a ion powe needs.
Di e en c yp osys ems, such as ellip ic cu e ones,
need o be used.
6. Acknowledgemen s
This wo k was unded in pa by he Fi h F amewo k
IST p og am o he Eu opean Commission unde he
TeleCARE IST-2000-27607, and by he Spanish
company SKILL Technology G oup SL. The au ho s
hank he con ibu ion o he TeleCARE conso ium
membe s.
Pa e n Resul s
Communica ion and mig a ion
beha io pa e ns
• I is p e e able o send a bu s o agen s o
mig a ions han isola ed ones
• Bad p o ocol design could lead in a eally big
pe o mance hi
Load beha io pa e ns
• Fo small loads p o ocol o e head is mo e
e iden , wi h ew changes in pe o mance
be ween clea and enc yp ed agen s
• As load inc eases, p o ocol o e head becomes
less e iden . Pe o mance hi de i es om he
c yp og aphic algo i hm i sel
• Fo e y high loads (>=512Kb), he e is se e e
beha io change and a scale change in
measu ed ime (e en o he clea agen )
Communica ion, mig a ion and
load blend beha io pa e ns
• In gene al, any inc ease in numbe o
i ine a ies o messages sen , o in agen load
p oduces an inc ease in execu ion ime, bu a
di e en scales
• Inc easing bo h numbe o i ine a ies and agen
load widen he di e ence in execu ion ime
be ween clea and ciphe ed agen s mo e han
in wha e e isola ed case
• Inc easing he numbe o messages sen o he
numbe o mig a ions necessa y o ans e
da a o a des ina ion a he han sending a big
block o da a, dec eases pe o mance, excep
o he pe o mance bo leneck case
Comp ession in a mid o high
bandwid h media beha io
pa e n
• Imp o es pe o mance o isola ed and small
messages o mig a ions. The pe o mance is
wo se in he es o cases
• A s udy in e y low bandwid h link is
necessa y o es beha io al changes in o he
en i onmen s
Table 1. Summa y o eComme ce beha io
analysis esul s