A icle
E alua ion o De ice-Independen In e ne
Spa ial Loca ion
Dan Komosny 1,*, Paul Pang 2, Mi alem Mehic 3and Mi osla Voznak 3
1Depa men o Telecommunica ions, B no Uni e si y o Technology, B no 61600, Czech Republic
2Depa men o Compu ing, Uni ec Ins i u e o Technology, Auckland 1025, New Zealand;
[email p o ec ed]
3Depa men o Telecommunica ions, Technical Uni e si y o Os a a, Os a a 70833, Czech Republic;
[email p o ec ed] (M.M.); mi osla [email p o ec ed] (M.V.)
*Co espondence: [email p o ec ed].cz; Tel.: +420-54114-6973
Academic Edi o s: No be Ba elme and Wol gang Kainz
Recei ed: 27 Feb ua y 2017; Accep ed: 24 May 2017; Published: 27 May 2017
Abs ac :
De ice-independen In e ne spa ial loca ion is needed o many pu poses, such as da a
pe sonalisa ion and social beha iou analysis. In e ne spa ial da abases p o ide such loca ions based
he IP add ess o a de ice. The ee o use da abases a e na i ely included in o many UNIX and Linux
ope a ing sys ems. These sys ems a e p edominan ly used o e-shops, social ne wo ks, and cloud
da a s o age. Using a cons uc ed g ound u h da ase , we comp ehensi ely e alua e hese da abases
o null esponses, e u ned coun y/ egion/ci y, and dis ance e o . The c ea ed g ound u h
da ase di e s om o he s by co e ing ci ies wi h bo h low and high popula ions and main aining
only de ices ha ollow he ule o one IP add ess pe ISP (In e ne Se ice P o ide ) and pe ci y.
We de ine wo new pe o mance me ics ha show he e ec o ci y popula ion and us wo hiness
o he esul s. We also e alua e he da abases agains an al e na i e measu emen -based app oach.
We s udy he easons behind he esul s. The da a e alua ed comes om Eu ope. The esul s may be
o use
o enginee s
, de elope s and esea che s ha use he knowledge o geog aphical loca ion o
ela ed da a p ocessing and analysis, such as ma ke ing.
Keywo ds:
spa ial loca ion; de ice independen ; IP add ess; geoloca ion; da abase; loca ion-awa e;
ci y popula ion; in e ne
1. In oduc ion
This pape deals wi h he spa ial loca ion o In e ne de ices. I ocuses on de ice-independen
loca ion whe e he loca ion o a de ice is es ima ed emo ely by i s known IP add ess and no
locally by he de ice i sel , such as using a GPS module. The e o e, he pape ocuses on an In e ne
de ice-independen spa ial loca ions ha
•does no use GPS o o he global posi ioning sys ems,
•
does no use e es ial adio-based loca ion sys ems, such as iangula ion in WiFi o mobile
cellula ne wo ks,
•does no use loca ion in o ma ion en e ed by use s.
The de ice-independen se ice p o ide s o do no communica e wi h he de ice loca ed.
The loca ion is es ima ed based on he knowledge o de ice IP add ess. Such IP-based loca ion
is used o a g ea numbe o loca ion-awa e se ices and applica ions, including web con en
and social ne wo k pe sonaliza ion [
1
]; use beha iou analysis [
2
] (including isi o maps
o websi es); load balancing by edi ec ing use s o geog aphically close da a/ esou ce eplicas [
3
];
geog aphical-based da a collec ion om a la ge numbe o use s [
4
]; spam il e ing by sende loca ion;
ISPRS In . J. Geo-In . 2017,6, 155; doi:10.3390/ijgi6060155 www.mdpi.com/jou nal/ijgi
ISPRS In . J. Geo-In . 2017,6, 155 2 o 17
de ec ion o ID and passwo d sha ing; de ec ion o c edi ca d online aud [
5
]; and law en o cemen
on media dis ibu ion by deli e y es ic ions.
Compa ed wi h In e ne de ice-dependen loca ion, he se ice p o ide s di ec ly communica e
wi h he de ice loca ed o ob ain i s loca ion. The sha ed loca ion is ypically ob ained om
GPS o WiFi iangula ion. A use ypically has o ag ee wi h sha ing hei loca ion. This may
happen when
an applica ion/se ice
is being ins alled and a use accep s he legal ag eemen s.
Applica ions o In e ne
de ice-dependen loca ion a y a lo . Typical examples a e inding nea by
poin s o in e es s, inding nea es social media con ac s, a ic epo s, municipal anspo
schedules, e c.
In his pape , we e alua e he pe o mance o In e ne de ice-independen spa ial loca ion.
We ocus on ee o use loca ion da abases as hey a e commonly used o unning In e ne se ices
and applica ions. The ee loca ion da abases a e na i ely included in many open-sou ce UNIX and
Linux ope a ing sys ems. These sys ems a e p edominan ly used o hos ing e-shops, social ne wo ks,
and cloud
da a s o age. The web analy ic and ma ke sha e company W3Cook moni o s he op
million se e s mon hly and shows ha a ound 97% o he web se e s a e hos ed by open-sou ce
Linux ope a ing sys ems [
6
]. The ee loca ion da abases a e also a ailable h ough so wa e packaging
sys ems ( pm, deb) o such ope a ing sys ems.
We use a il e ed g ound u h da ase ha we cons uc ed. I includes de ices coming om
Eu ope wi h a known co ec loca ion. These de ices a e mobile phones wi h WiFi. Ou g ound u h
da ase di e s om hose used in he ela ed esea ch by co e ing di e en ypes o coun y a eas,
especially including less inhabi ed places. The da ase s used in he ela ed wo k ypically come om
la ge and majo ci ies. An example o such a commonly used da ase is Plane Lab [
7
]. The de ices om
la ge ci ies a e gene ally loca ed wi h a be e accu acy. Howe e , o gene al use, he de ices om
lesse popula ed places should also be conside ed. We p o ide a de ailed desc ip ion o he da ase
c ea ed, including he ci y popula ion numbe s.
An une en dis ibu ion o he g ound u h de ices among he ISPs in luences loca ion accu acy.
The eason is ha de ices wi h an IP add ess assigned ia he same ISP p o ide a e mapped
o he same
o a small se o loca ions. We he e o e p ocess ou g ound u h da ase o main ain only
de ices ha ollow he ule o one IP add ess pe ISP and pe ci y. This way, he esul s should be
e o - ee as hey a e no in luenced by a mul iplica ion o he same esul s o he de ices wi h an IP
add ess ob ained h ough he same ISP.
We pa icula ly e alua e hese me ics: null esponses e sus e u ned loca ions, co ec ly
esol ed coun y/ egion/ci y, and dis ance e o . We addi ionally de ine wo new me ics ha show
he pa icula
pe o mance di e ences. These me ics a e ‘accu acy dependence on ci y popula ion’
and ‘ us wo hiness o he loca ions’. We also compa e he ee da abases wi h an al e na i e
measu emen -based loca ion.
The pape is s uc u ed as ollows: Sec ion 2desc ibes he backg ound on de ice-independen
spa ial loca ion based on IP add esses. Sec ion 3gi es a lis o he ela ed esea ch and su eys
in IP
geoloca ion. We summa ize he p esen ed esul s and discuss hei e alua ion. Sec ion 4p esen s
in de ail he p inciple o IP geoloca ion. A desc ip ion o he ee o use da abases used in his esea ch
is also included. In Sec ion 5, we desc ibe ou me hod o c ea ing he g ound u h da ase . The used
me hodology o loca ion pe o mance e alua ion is p esen ed in Sec ion 6. The esul s and commen s
on he indings a e desc ibed in Sec ion 7. Finally, we conclude he pape .
2. Backg ound on De ice-Independen Spa ial Loca ion
The e a e se e al me hods used o de ice-independen loca ion based on IP add esses.
We summa ize
hem in Table 1. The i s wo a e used by he gene al public. These me hods can
be accessed eely o h ough a paid subsc ip ion. The las me hod is ypically dedica ed o legally
au ho ized en i ies only.
ISPRS In . J. Geo-In . 2017,6, 155 3 o 17
The i s me hod—da abase lookup—deals wi h sea ching he global loca ion da abases.
A geoloca ion
da abase main ains blocks o con inuous IP add esses. Each such block has a loca ion
assigned. The assigned loca ions a e ob ained h ough se e al sou ces ha we desc ibe in Sec ion 4.
A de ice loca ion is es ima ed by sea ching he da abase o he co esponding block o IPs and i
a ma ch is ound, he geog aphical posi ion s o ed o he block is e u ned [8–10].
The second me hod—In e ne measu emen —is based on measu emen and analysis o da a
communica ion. Typically, communica ion la ency is measu ed om a se o se e s called landma ks
wi h known loca ions o he de ice loca ed. The la encies a e con e ed o a geog aphical loca ion
o he de ice using se e al echniques, such as dis ance mul ila e a ion [11,12].
The las me hod—ISP p i a e da a—is based on p i a e eco ds o In e ne se ice p o ide s (ISP).
An ISP leases IP add esses o he de ices. Coun y-speci ic laws speci y wha in o ma ion has o be
eco ded and how long i has o be main ained by ISPs. ISP subsc ibe s p o ide hei de ails including
pos al add esses in he billing con ac s. The loca ion o a de ice can be acked down by linking his
in o ma ion. The legal de ails may a y ac oss coun ies. The police and jus ice se ices a e usually
au ho ised o be gi en he loca ion upon a o mal eques .
The accu acy o de ice-independen loca ion is gene ally low, usually in a ange o ens o
hund eds o kilome es. The accu acy o he ISP in e nal loca ion in o ma ion a ies acco ding
o he eco ds kep .
Table 1. O e iew o de ice-independen loca ion based on IP add esses.
How Accu acy Typically Used by
Da abase lookup Low Non-de ice use (e e ybody)
In e ne measu emen Low Non-de ice use (e e ybody)
ISP p i a e da a High Non-de ice use (au ho ized)
The e a e o he mino me hods o de ice-independen loca ion, such as da a-mining
o web
pages and o he In e ne esou ces o spa ial in o ma ion [
13
]. Ano he me hod is based
on an enhancemen
o he DNS (Domain Name Sys em) se ice called DNS LOC [
14
]. This se ice
p o ides he geog aphical loca ion o a domain name. The disad an age o his solu ion is a poo
co e age o In e ne add ess space, and i is no widely used.
Di e en me hods a e used o de ice-dependen loca ion. These me hods a e usually based on
global posi ioning sys ems (GPS), measu ing adio signal s eng h (RSSI), ime o a i al (TOA),
and angle o a i al (AOA) [
15
,
16
]. The accu acy o de ice-dependen loca ion is highe han
de ice-independen . Some p inciples a e sha ed wi h de ice-independen loca ion, such as dis ance
mul ila e a ion.
3. Rela ed Wo k
In his sec ion, we e iew he ela ed wo k and summa ize he esul s wi h a ocus on hese
poin s o in e es : loca ion accu acy, loca ion e iciency (null esponses e sus e u ned loca ions),
and he up- o-da eness
p oblem. The p oblem dealing wi h he cu en ness o he s o ed loca ion da a
is speci ic o any da abase-based IP geoloca ion and is caused by he non-s a ic IP add ess space since
new assignmen s and ealloca ions o IP add esses con inually happen.
Re . [
17
] s udied he accu acy o he common loca ion da abases—MaxMind, IPligence, Geoby es,
Hos IP, Digi al En oy/Ne Acui y, and IP2Loca ion. The s udy indica ed ha he loca ion pe o mance
a ied a lo — om 20 o 80% o he co ec ci y es ima ions. The au ho s o his s udy used a 100 km
ange o he ci y-le el de ini ion, whe eas o he esea che s used a 40 km ange, such as [
8
,
9
].
The esul s wi hin he coun y-le el we e a ound 80% and be e . The s udy also in ol ed a s a is ic o
he success ul loca ion que ies. Some da abases did no loca e abou 30% o he IP add esses eques ed
(we le he Hos IP da abase as i epo ed a much highe pe cen age o he unsuccess ul loca ion
a emp s han he o he s). The au ho s o he s udy emphasised ha his es showed whe he an en y
ISPRS In . J. Geo-In . 2017,6, 155 4 o 17
o an IP add ess exis s and i does no say any hing abou he accu acy o he loca ions ob ained.
The s udy also compa ed he loca ion accu acy epo ed by he da abase endo s. The epo ed
coun y-le el accu acy was o e 97% and he ci y-le el accu acy was o e 80%. These alues we e o a
g ea e accu acy compa ed o he da a collec ed in he s udy.
Re . [
8
] e alua ed he accu acy o se e al loca ion da abases including Digi al En oy/Ne Acui y,
MaxMind/GeoIP, MaxMind/GeoLi e, IPligence and Hos IP. They ound ha o e 95% o he loca ions
we e co ec a he coun y-le el. O he co ec loca ions, 70–90% we e wi hin he ci y-le el (40 km).
Re . [
9
] conside ed i e loca ion da abases—MaxMind, IP2Loca ion, In oDB, Hos IP,
and So wa e77
.
The au ho s ound ha he loca ion accu acy o he coun y-le el was abo e 96%. They also no iced ha
he loca ion accu acy was qui e low o he ci y-le el (40 km), app oxima ely 20%.
The CAIDA (Cen e o Applied In e ne Da a Analysis) echnical epo [
10
] s udied IP
geoloca ion accu acy a he coun y-le el g anula i y. The epo ocused on he RIR (Regional
In e ne Regis e s) da abase and compa ed i agains he MaxMind/GeoLi e da abase. Bo h da abases
pe o med app oxima ely he same—95% o he IP add ess space was loca ed iden ically a
he coun y-le el.
The con inual e olu ion o he IP add ess space (i.e., new IP add esses assignmen s o
ealloca ions) is a signi ican p oblem o any da abase-based IP geoloca ion. The endo s claim
o lose accu acy o a couple o pe cen in each mon h wi hou e lec ing hese changes. Fo example,
IP2Loca ion epo s up o 5% o he IP eco ds being upda ed each mon h [
18
]. Re . [
17
] showed ha
app oxima ely 1–2% o he loca ions was changed each mon h o he da abases IPligence, Hos IP,
IP2Loca ion, and Ne Acui y.
The ela ed wo k epo ed qui e di e en esul s, mainly dealing wi h loca ion accu acy.
We explain his by he use o he di e en g ound u h da ase s wi hou il e ing he edundan
o w ong en ies. This consequen ly misin e p e s he esul s. Some s udies also used ‘ i ual’
g ound u h da a by c ea ing g oups o close IP add ess called PoPs (Poin s o P esence) [
17
].
The PoPs we e de e mined by an analysis o he ne wo k opology s uc u e and communica ion
la ency measu emen s. Each PoP was assigned a loca ion de i ed as a comp omise o he di e en
loca ion esul s om he loca ion da abases. Such e alua ion esul ed in a be e loca ion accu acy
o he ci y-le el in con as wi h accu acy being e alua ed using o he g ound u h da ase s.
4. De ice-Independen Loca ion Using IP Geoloca ion Da abases
IP geoloca ion da abases a e commonly used o de ice-independen loca ion
because o he con enien
access and he sho ime o he loca ion p ocess when compa ed
o al e na i e measu emen -based loca ion. The loca ion da abases g oup con inuous IP add esses in o
blocks, usually s o ing he i s and las add ess o he block (o he block size gi en by a ne wo k
mask). The de ined blocks ha e loca ion in o ma ion assigned. This loca ion in o ma ion is de i ed
om a ious sou ces, such as om In e ne egis ies managed by he In e ne Assigned Numbe s
Au ho i y (IANA), In e ne a ic measu emen and analysis, mining o public online da a ( ypically
web pages), in o ma ion om GPS de ices, and olun ee s p o iding hei loca ions. An example
o communi y-based IP geoloca ion da abase is Hos ip [
19
]. Mo e sou ces a e usually combined o
imp o e he loca ion in o ma ion alidi y.
4.1. Cons uc ion o IP Geoloca ion Da abases
An example cons uc ion o a geoloca ion da abase is shown in Lis ings 1and 2. The lis ings
a e based on he ee o use da abase GeoLi eCi y by MaxMind. The da abase uses he CSV o ma .
Bo h lis ings we e simpli ied o a be e unde s anding. The i s lis ing shows a sample con en
o he i s
ile, whe e he blocks o he con inuous IP add esses a e de ined. The block is delimi ed by
he i s IP add ess and i s size is gi en by he ne wo k mask. The nex i ems in he lis a e he block ID
and he bound loca ion in o ma ion in he o m o la i ude and longi ude.
ISPRS In . J. Geo-In . 2017,6, 155 5 o 17
177.87.232.0,119,3077706,50.2391,12.7486
288.101.213.0,120,3073254,49.5985,18.1452
Lis ing 1: Blocks o con inuous IP add esses wi h assigned loca ion in o ma ion.
13077706 , EU ,Eu ope ,CZ , Czechia ,KA , Ka lo a sky egion , Chodo
23073254 , EU ,Eu ope ,CZ , Czechia ,MO , Mo a skoslezsky egion , Kop i nice
Lis ing 2: Loca ion in o ma ion o blocks o con inuous IP add esses.
The second lis ing shows o he loca ion in o ma ion assigned o he block o he con inuous IP
add esses. I s a s wi h he block ID ollowed by he con inen code and name, coun y code and
name, egion code and name, and ci y name.
4.2. Loca ion In o ma ion P o ided
The geoloca ion da abases a e known o a la ge di e si y o he loca ion accu acy as we show
in Sec ion 3. An example o such loca ion di e si y o an In e ne de ice is shown in Figu e 1. I shows
loca ions ob ained om se en ee o use loca ion da abases:
•GeoLi e2 Ci y by MaxMind [20], in he pape e e edas GeoLi e2,
•IP add ess o ci y (low esolu ion) by DB-IP [21], in he pape e e ed as IP oCi y,
•DB11.LITE by IP2Loca ion [22],
•Li e F ee by IPligence [23],
•hos ip [19],
• eegeoip [24],
•so wa e77 [25].
The map in he igu e shows ha he es ima ed loca ions a e some imes a apa . Some es ima ed
loca ions poin ed o he capi al ci y o he coun y (P ague)—IP oCi y, and DB11.LITE. O he da abases
es ima ed only he co ec coun y—Li e F ee and so wa e77. In his case, he geog aphical cen e o he
coun y was used. The o he wo da abases es ima ed he co ec egion, bu he w ong ci y—GeoLi e2,
and eegeoip. Finally, he hos ip da abase e u ned a loca ion ou side he co ec coun y.
The loca ion e o s a e summa ized in Table 2. The co ec loca ion was a la i ude 48.97
and longi ude 16.61. The loca ion e o s a ied om 7 km (GeoLi e2) o 201 km (DB11.LITE).
Some da abases
e u ned di ec ly he coo dina es o he es ima ed loca ion. In he cases when he
coo dina es we e no e u ned di ec ly, we de i ed hem as he geog aphical cen e o he e u ned
ci y/ egion/coun y, espec i ely. In he able, we ma ked hese cases as ‘*’.
Table 2. Accu acy o loca ions ob ained om se en ee o use IP geoloca ion da abases.
Vendo /Da abase La , Lon E o [km] No e
MaxMind/GeoLi e2 48.98, 16.52 7 Co ec egion
DB-IP/IP oCi y 50.08, 14.47 * 190 Capi al ci y
IP2Loc./DB11.LITE 50.09, 14.42 201 Capi al ci y
IPligence/Li e F ee 49.75, 15.5 * 118 Cen e o coun y
hos ip 50.0, 18.47 177 W ong coun y
eegeoip 49.0, 16.86 19 Co ec egion
so wa e77 49.75, 15.5 * 118 Cen e o coun y

ISPRS In . J. Geo-In . 2017,6, 155 6 o 17
Figu e 1. Loca ions ob ained om se en ee o use IP geoloca ion da abases (map shows egions).
The da abases a y in wha in o ma ion hey p o ide. Table 3gi es an o e iew o he loca ion
in o ma ion p o ided by each da abase. GeoLi e2, DB11.LITE, and eegeoip p o ide a ull lis
o geog aphical in o ma ion, i.e., coun y, egion, ci y, and coo dina es. The da abase IP oCi y does no
p o ide he coo dina es and he da abase hos ip does no p o ide he egion. The da abases Li e F ee
and so wa e77 p o ide only coun y.
Table 3. Loca ion in o ma ion p o ided by IP geoloca ion da abases.
Vendo /Da abase Coun y Region Ci y Coo dina es
MaxMind/GeoLi e2 x x x x
DB-IP/IP oCi y x x x -
IP2Loca ion/DB11.LITE x x x x
IPligence/Li e F ee x - - -
hos ip x - x x
eegeoip x x x x
so wa e77 x - - -
Geoloca ion da abases can be accessed emo ely o locally. Remo e access is p e e ed o a low
numbe o loca ion lookups as each lookup gene a es a ic and is delayed. In his case, he loca ion
se ice p o ide main ains he da abase o be upda ed. Usually, he numbe o loca ion que ies is
eco ded and he se ice p o ide cha ges he cus ome s based on hese numbe s.
5. Cons uc ion o G ound T u h Da ase
We cons uc ed he g ound u h da ase by a long- e m collec ion o he co ec loca ions
o mobile WiFi de ices. Fo each de ice, we s o ed i s public IP add ess, ISP, coun y, egion, ci y,
ci y popula ion
, and he geog aphical coo dina es. The coo dina es we e ob ained h ough in-buil GPS.
The ne wo k- ela ed in o ma ion was ob ained using a de eloped applica ion un on he collec ion
de ices. The de ices we e ypically connec ed h ough WiFi access poin s wi h NAT (Ne wo k
Add ess T ansla ion). The applica ion he e o e esol ed he public IP add ess o he WiFi access
poin . Conside ing ha de ice-independen IP loca ion wo ks wi h a esolu ion o kilome es and
he WiFi ypical co e age is o a hund ed o me e s, we linked he us ed GPS loca ion o he de ice
ISPRS In . J. Geo-In . 2017,6, 155 7 o 17
wi h he public IP add ess o he WiFi access poin . The applica ion un on he de ices epo ed
he da a o ou collec ion se e . The epo ed da a was p ocessed
o esol e
ci y, egion, and coun y.
We ob ained his in o ma ion using e e se geocoding
o he coo dina es
. The ISP o he de ices was
aken om he RIPE NCC (Reseaux IP Eu opeens Ne wo k Coo dina ion Cen e) WHOIS da abase
based on he public IP add ess.
Finally, he demog aphic
in o ma ion o ci ies was ob ained h ough
Na u al Ea h [26] using de eloped sc ip s in Py hon.
The da abases may e u n he same loca ion o a se o de ices belonging o he same ISP. This has
a de imen al e ec on loca ion accu acy e alua ion as he esul s a e shi ed owa ds he alues
o such de ices. We he e o e pa icula ly ocused on he ISP geog aphical dis ibu ion. We inspec ed
he ISPs in each ci y in ol ed in he g ound u h da ase . I we ound mo e de ices wi hin a ci y
and belonging o he same ISP, we il e ed ou he addi ional de ices in he same ci y o lea e only
one ISP/ci y de ice in he da ase . Fo he de ices ound in he coun y and belonging o he same
ISP, we used a minimal dis ance o 10 km be ween he de ices. I he dis ance was smalle , we again
il e ed ou he addi ional de ices. We se his dis ance empi ically a e inspec ing he ela ion
o he
g ound u h de ices (co ec loca ions) and he es ima ed loca ions. An example o such an a ea
gi en by his dis ance is shown in Figu e 2. The igu e shows he co ec loca ions (OK-sign icon) and
he es ima ed loca ions (in o-sign icon). The pa icula pai s—co ec and es ima ed loca ions—a e
connec ed wi h a line. The highligh ed a ea labelled as ‘same e o a ea’ shows he place a which he
addi ional un il e ed nodes we e es ima ed o he same loca ion (connec ed wi h he line).
Figu e 2. Same e o a ea.
We collec ed he g ound u h de ices o h ee yea s and eco ded mo e han 3000 loca ions
in o al. F om he esul o ISP geog aphical dis ibu ion p ocessing, we ob ained da a o app ox.
600 de ices in 234 di e en ci ies in Eu ope. The e o e, he p esen ed esul s should no be in luenced
by using mul iple de ices belonging o he same ISP in a ci y.
ISPs may eassign he same IP add ess o a di e en de ice which migh be (o migh no be)
a apa om he geog aphical loca ion o he p e ious de ice. As desc ibed in he ela ed wo k,
he da abases
change up o 1–5% o hei loca ion eco ds each mon h. Fo example, IP2Loca ion
epo s up o 5% o he IP eco ds being upda ed each mon h [
18
], and Re . [
17
] showed ha
app oxima ely 1–2% o he loca ions was changed each mon h.
We app oached his p oblem by loca ing he nodes on a one-by-one basis when new nodes
we e added o he g ound u h. The pa icula s eps we e: (i) a us ed loca ion was epo ed by
ISPRS In . J. Geo-In . 2017,6, 155 8 o 17
a communi y membe who con ibu ed o he g ound u h da ase . A mobile GPS de ice wi h
ou de eloped applica ion was used. The GPS loca ion and he ele an ne wo king da a was
immedia ely ans e ed o ou collec ion se e . (ii) On he collec ion se e , once a day ano he
de eloped applica ion was un. This applica ion loca ed he newly epo ed IP add esses using
he loca ion da abases. The applica ion also calcula ed he loca ion e o , ci y/ egion/coun y
co ec ness,
and o he
ele an da a. The esul s we e hen me ged wi h he p e iously s o ed
da a. This way, we wo ked wi h he up- o-da e co ec loca ions and elimina ed he p oblem o IP
add ess e-alloca ions. We also upda ed he locally-s o ed da abases on a mon hly basis o keep
hem up- o-da e.
The da abases ypically loca e he de ices om la ge ci ies wi h a be e accu acy. We he e o e
also ocused on small ci es and, as a esul , he g ound u h da ase co e s a ange o ci ies wi h
a di e en popula ion. The ci y popula ion de ails a e shown in Figu e 3. The median ci y popula ion
in he da ase is 17,000. The smalles ci y has 64 inhabi an s and he la ges ci y has 1.8 million people.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
100 1000 10000 100000 1e+06
Cumula i e p obabili y o g. . nodes
Ci y popula ion
17000 people
Figu e 3. G ound u h de ices in ci ies wi h di e en popula ion.
6. Me ics o Pe o mance E alua ion
We app oach he pe o mance e alua ion by he use o an ex ended se o me ics. We also
de ine ou addi ional me ics o pa icula ly s udy he impo an p ope ies o he da abases.
Whe e applicable, we compa e he da abases wi h an al e na i e measu emen -based loca ion.
Fi s , we analyse whe he a loca ion o he eques ed de ice is e u ned o no , i.e.,
loca ion e iciency. The e u ned loca ion can be coun y, egion, ci y, o coo dina es. The loca ion
e iciency is also compa ed wi h he measu emen -based loca ion. I is no di e en ia ed be ween he
co ec and w ong loca ions e u ned and we do no measu e he dis ance e o in his case.
Nex , we deal wi h loca ion accu acy. We wo k wi h he accu acy as being disc e e and
quan i a i e. Wi h disc e e accu acy, we inspec i he coun y/ egion/ci y e u ned is co ec .
The da abases di e in he p o ided disc e e loca ion in o ma ion as shown in Table 3. Fo quan i a i e
accu acy, we calcula e he e o dis ance be ween he co ec and es ima ed coo dina es. We conside
hese wo pa icula poin s o ob aining he coo dina es:
(i) Some o he da abases do no e u n he coo dina es di ec ly. Table 4shows ou app oach
in ob aining hem in such cases. In he able, we ma ked he di ec ly e u ned coo dina es
wi h he
le e ‘a’. The le e ‘b’ deno es ha we calcula e he coo dina es as he geog aphical cen e o a ci y
e u ned. I a ci y is no e u ned, he cen e o he e u ned egion is used. I a egion is no e u ned,
we conside he coo dina es o be unknown. The le e ‘c’ deno es he si ua ion when he i ual
coo dina es o he cen e o a coun y a e e u ned.
ISPRS In . J. Geo-In . 2017,6, 155 9 o 17
Table 4. Sou ce o es ima ed de ice coo dina es.
Vendo /Da abase Coun y Region Ci y
MaxMind/GeoLi e2 - - a
DB-IP/IP oCi y - b b
IP2Loca ion/DB11.LITE - - a
IPligence/Li e F ee c - -
hos ip - - a
eegeoip - - a
so wa e77 c - -
a—di ec ly e u ned; b—cen e o ci y o egion; c—cen e o coun y (no included in e alua ion).
(ii) The measu emen -based loca ion does no e u n he coo dina es, bu i esul s
in a geog aphica
l a ea whe e he de ice is loca ed [
27
]. Wi h his ype o loca ion, communica ion
la ency om a se o landma ks (wi h known geog aphical posi ion) is measu ed o he de ice loca ed.
As he e exis s a posi i e co ela ion be ween la ency and geog aphical dis ance [
28
], he measu ed
la encies a e con e ed o geog aphical dis ances ha de ine he adius o he g ea -ci cles a ound each
landma k. The in e sec ion o he ci cles delimi s he a ea o he de ice loca ion as shown in Figu e 4.
The igu e is a sample ou pu om ou de eloped measu emen loca ion sys em. Ou landma k
so wa e is un on Plane Lab [
7
] se e s si ua ed in Eu ope [
29
]. We use he cen e o g a i y o he
in e sec ed a ea o ob ain he speci ic coo dina es, as i was in oduced in [27].
Figu e 4. Used la ency-based geoloca ion.
Ou i s me ic is accu acy dependence on ci y popula ion. As discussed be o e, he da abase
endo s claim o loca e de ices om highly popula ed a eas wi h a be e accu acy. Howe e ,
he low
and high popula ion densi y di e ences a e no gi en o s udied in he ela ed wo k we ound.
We di ide
he ci ies in he g ound u h da ase in o wo se s—small (less han 17,000 people) and la ge
(mo e han 17,000 people) based on he median ci y popula ion shown in Figu e 3.
Ou second me ic is he us wo hiness o he da abases. As he esul s in Sec ion 7show,
some da abases e u n a loca ion o almos any de ice. I seems he e o e ha he da abase pe o ms
well. Howe e , some o he e u ned esul s a e no co ec . This migh lead o alse conclusions
abou he loca ion pe o mance. The us wo hiness o he da abases indica es how much he esul s
can be us ed.
ISPRS In . J. Geo-In . 2017,6, 155 16 o 17
•
The communi y-p o ided loca ion in o ma ion is no a eliable sou ce o IP geoloca ion.
This in o ma ion does no co e he whole IP add ess space, which is indica ed by low loca ion
e iciency esul s. In addi ion, loca ion accu acy is low when his in o ma ion is used.
•
The al e na i e la ency-based loca ion p o ides wo se esul s compa ed wi h he da abase-based
loca ion. In pa icula , we obse ed ha loca ion e iciency was poo . The eason is ha
la ency measu emen s a e o en blocked by ne wo king de ices en ou e o by he end de ices.
We obse ed his o o e 50% o he de ices loca ed.
As can be seen om he esul s, he cu en app oaches may p oduce inco ec spa ial da a.
This makes he dependen In e ne loca ion-awa e se ices p one o e o s. Fu he esea ch in o
de ice-independen loca ion may be mo i a ed by a be e use o he a ailable In e ne esou ces o
public sa e y. New widely used In e ne -based communica ion echnologies, such as online messaging
and VoIP (Voice o e IP), may be used in public sa e y p o ided ha he loca ions a e accu a e o a
speci ic le el. Fo example, wi h VoIP, he loca ion o he calle in eme gency is needed o be known in
cases when he loca ion canno be sha ed due o speci ic ci cums ances. These may be houseb eaking,
kidnapping, and heal h inju ies no allowing he calle o speak, o he calle may be o ced o abandon
he call. Ano he issue ela ed o public sa e y a e hoax calls. Wi h be e knowledge in he ield o
de ice-independen spa ial loca ions, hoax calls could be de ec ed by a compa ison o he calle ’s
loca ion o he epo ed place o ouble [32].
Acknowledgmen s:
Resea ch desc ibed in his pape was inanced by he Na ional Sus ainabili y P og am unde
g an LO1401. Fo he esea ch, he in as uc u e o he SIX (Senso , In o ma ion and Communica ion Sys ems)
Cen e was used.
Au ho Con ibu ions:
Dan Komosny p oposed he idea o he pape . He designed he expe imen s and w o e
he p og ams needed. He also analysed he da a. Paul Pang p o ided he in o ma ion abou he accu acy
and e iciency o spa ial loca ions. He also con ibu ed o he opic dealing wi h loca ion-awa e applica ions.
Mi alem Mehic and Mi osla Voznak p o ided da a o spa ial loca ions by co e ing a ious ci ies and coun ies.
Con lic s o In e es : The au ho s decla e no con lic o in e es .
Re e ences
1.
Chia a, D.; Gio anni, P.; Sebillo, M.; To o a, G.; Vi iello, G. Geoma ke ing Policies and Augmen ed Reali y
o Ad e isemen Deli e y on Mobile De ices. In P oceedings o he 17 h In e na ional Con e ence
on Dis ibu ed Mul imedia Sys ems. Knowledge Sys ems Ins i u e, Flo ence, I aly, 18–20 Oc obe 2011;
pp. 78–83.
2.
Ba nes, R.; Lepinski, M.; Coope , A.; Mo is, J.; Tscho enig, H.; Schulz inne, H. An A chi ec u e o Loca ion
and Loca ion P i acy in In e ne Applica ions; Reques o Commen s: 6280 (Bes Cu en P ac ice); In e ne
Enginee ing Task Fo ce: F emon , CA, USA, 2011.
3.
Yen, N.Y.; Huang, R.; Ma, J.; Jin, Q.; Shih, T.K. In elligen ou e gene a ion: disco e y and sea ch o co ela ion
be ween sha ed esou ces. In . J. Commun. Sys . 2013,26, 732–746.
4.
Bu ge , R.; Komosny, D.; Ka hi a elu, G. Topology Awa e Feedback T ansmission o Real-Time Con ol
P o ocol. J. Ne w. Compu . Appl. 2012,35, 723–730.
5.
MaxMind. F aud De ec ion h ough IP Add ess Repu a ion and a Mu ual Collabo a ion Ne wo k; Whi e Pape ;
MaxMind: Wal ham, MA, USA, 2011.
6. W3cook. A ailable online: www.w3cook.com (accessed on 26 May 2017).
7. Plane Lab. A ailable online: www.plane -lab.o g (accessed on 26 May 2017).
8.
Hu ake , B.; Fomenko , M.; Cla y, K. Geocompa e: A Compa ison o Public and Comme cial Geoloca ion Da abases;
Technical Repo ; Coope a i e Associa ion o In e ne Da a Analysis (CAIDA): La Jolla, CA, USA, 2011.
9.
Poese, I.; Uhlig, S.; Kaa a , M.; Donne , B.; Gueye, B. IP geoloca ion da abases: Un eliable? ACM SIGCOMM
Compu . Commun. Re . 2011,41, 53–56.
10.
Zande , S. How Accu a e Is IP Geoloca ion Based on IP Alloca ion Da a? Technical Repo 120524A; Swinbu ne
Uni e si y o Technology: Haw ho n, VIC, Aus alia, 2012.

ISPRS In . J. Geo-In . 2017,6, 155 17 o 17
11.
Mo a ek, P.; Komosny, D.; Simek, M.; S eda, J.; Handl, T. Vi aldi and o he localiza ion me hods.
In P oceedings o he 32nd In e na ional Con e ence on Telecommunica ions and Signal P ocessing,
Gyo -Moson-Sop on, Hunga y, 26–27 Augus 2009; pp. 214–218.
12.
Mo a ek, P.; Komosny, D.; Bu ge , R.; S eda, J.; Handl, T.; Ja oso a, L. S udy and pe o mance o localiza ion
me hods in IP based ne wo ks: Vi aldi algo i hm. J. Ne w. Compu . Appl. 2011,34, 351–367.
13.
Guo, C.; Liu, Y.; Shen, W.; Wang, H.; Yu, Q.; Zhang, Y. Mining he Web and he In e ne o Accu a e IP
Add ess Geoloca ions. In P oceedings o IEEE In e na ional Con e ence on Compu e Communica ions
2009, Rio de Janei o, B azil, 19–25 Ap il 2009; pp. 2841–2845.
14.
Fio eze, T.; Heijenk, G. Ex ending DNS o suppo geocas ing owa ds VANETs: A p oposal. In P oceedings
o Vehicula Ne wo king Con e ence (VNC) 2010, Je sey Ci y, NJ, USA, 13–15 Decembe 2010; pp. 271–277.
15.
Mo a ek, P.; Komosny, D.; Simek, M.; Jelinek, M.; Gi bau, D.; Laza o, A. In es iga ion o adio channel
unce ain y in dis ance es ima ion in wi eless senso ne wo ks. Telecommun. Sys . 2013,52, 1549–1558.
16.
Mo a ek, P.; Komosny, D.; Simek, M.; Gi bau, D.; Laza o, A. Ene gy Analysis o Recei ed Signal S eng h
Localiza ion in Wi eless Senso Ne wo ks. Radioenginee ing 2011,20, 937–945.
17. Sha i , Y.; Zilbe man, N. A Geoloca ion Da abases S udy. IEEE J. Sel. A eas Commun. 2011,2, 2044–2056.
18. IP2Loca ion. A ailable online: www.ip2loca ion.com/da a-accu acy (accessed on 26 May 2017).
19. Hos ip. A ailable online: www.hos ip.in o (accessed on 26 May 2017).
20.
GeoLi e Legacy Downloadable Da abases. A ailable online: de .maxmind.com/geoip/legacy/geoli e
(accessed on 26 May 2017).
21. Da abase downloads. A ailable online: db-ip.com/db (accessed on 26 May 2017).
22. F ee IP Geoloca ion Da abase. A ailable online: li e.ip2loca ion.com (accessed on 26 May 2017).
23. IPligence Li e F ee. A ailable online: www.ipligence.com/ ee-ip-da abase (accessed on 26 May 2017).
24. F eegeoip Ne . A ailable online: www. eegeoip.ne (accessed on 26 May 2017).
25.
IP o Coun y Da abase (IPV4 and IPV6). A ailable online: so wa e77.ne /geo-ip (accessed on 26 May 2017).
26. Na u al Ea h. A ailable online: www.na u alea hda a.com (accessed on 26 May 2017).
27.
Gueye, B.; Zi iani, A.; C o ella, M.; Fdida, S. Cons ain -based geoloca ion o in e ne hos s. IEEE/ACM
T ans. Ne w. 2006,14, 1219–1232.
28.
Pe cacci, R.; Vespignani, A. Scale- ee beha io o he In e ne global pe o mance. Phys. B Condens. Ma e .
2003,32, 411–414.
29.
Komosny, D.; P uzinsky, J.; Ilko, P.; Polasek, J.; Masek, P.; Koca epe, O. On Geog aphic Coo dina es o
Plane Lab Eu ope. In P oceedings o 38 h In e na ional Con e ence on Telecommunica ions and Signal
P ocessing (TSP), Vysocany P ague, Czech Republic, 9–11 July 2015; pp. 642–646.
30.
Chun, B.; Culle , D.; Roscoe, T.; Ba ie , A.; Pe e son, L.; Waw zoniak, M.; Bowman, M. Plane Lab: an o e lay
es bed o b oad-co e age se ices. ACM SIGCOMM Compu . Commun. Re . 2003,33, 3–12.
31.
Komosny, D.; M do ic, S.; Ilko, P.; G ej ak, M.; Pospichal, O. Tes ing In e ne applica ions and se ices using
Plane Lab. Compu . S and. In e aces 2017,53, 33–38.
32.
Voznak, M.; Rezac, F. The implemen a ion o SPAM o e In e ne elephony and a de ence agains his
a ack. In P oceedings o he 32nd In e na ional Con e ence on Telecommunica ions and Signal P ocessing,
Gyo -Moson-Sop on, Hunga y, 26–27 Augus 2009; pp. 26–27.
c
2017 by he au ho s. Licensee MDPI, Basel, Swi ze land. This a icle is an open access
a icle dis ibu ed unde he e ms and condi ions o he C ea i e Commons A ibu ion
(CC BY) license (h p://c ea i ecommons.o g/licenses/by/4.0/).