Disco e ing he E he eum2 P2P Ne wo k
Mikel Co es-Goicoechea
Ba celona Supe compu ing Cen e
Ba celona, Spain
[email p o ec ed]
Leona do Bau is a-Gomez
Ba celona Supe compu ing Cen e
Ba celona, Spain
[email p o ec ed]
Abs ac —Achie ing he equilib ium be ween scalabili y, sus-
ainabili y, and secu i y while keeping decen aliza ion has
p e ailed as he a ge solu ion o decen alized blockchain
applica ions o e he las yea s. Se e al app oaches ha e been
p oposed by mul iple blockchain eams o achie e i , E he eum
being among hem. E he eum is on he pa h o a majo p o ocol
imp o emen called E he eum 2.0 (E h2), implemen ing Sha ding
and in oducing he P oo -o -S ake (PoS). As he change o
consensus mechanism is a delica e ma e , his imp o emen will
be achie ed h ough di e en phases, he i s o which is he
implemen a ion o he Beacon Chain. As E he eum1, E h2 elies
on a decen alized pee - o-pee (p2p) ne wo k o he message
dis ibu ion. Up o da e, we es ima e ha he e a e a ound 5.000
nodes in he E h2 main ne geog aphically dis ibu ed. Howe e ,
he opology o his one s ill p e ails unknown. In his pape , we
p esen he esul s ob ained om he analysis we pe o med on
he E h2 p2p ne wo k. Desc ibing he opology o he ne wo k,
as possible haza ds ha his one implies.
Index Te ms—Blockchain, E he eum2, E h2, Beacon Chain,
Sha ding, Moni o ing, P oo o S ake, P2P Ne wo ks, Scaling
I. INTRODUCTION
E he eum [1] is a ma u e decen alized web in as uc u e
ha has p o ed eliabili y and had led he g ow h o decen-
alized applica ions. I ea u es an ecosys em o decen alized
applica ions based on a gene al pu pose i ual machine [2]
and i s dedica ed p og amming language [3]. This ecosys em
holds ce ain cha ac e is ics, which has gene a ed a la ge
communi y o de elope s ha ha e p oposed new me hods o
ackle some o he limi a ions a ec ing he E he eum p o ocol.
The E he eum Founda ion oge he wi h he E he eum com-
muni y ha e been wo king on he de elopmen o he second
gene a ion o E he eum, E he eum 2.0 (E h2). The app oach
p oposed in he E h2 p o ocol ackles bo h: he scalabili y chal-
lenge by in oducing sha ding, and he sus ainabili y challenge
by changing he consensus p o ocol o P oo -o -S ake (PoS).
This ambi ious p ojec is di ided in o di e en phases because
o i s complexi y. The i s phase (phase 0) is he Beacon Chain
[4] launched on Decembe 2020.
E h2 aims o c ea e a majo ne wo k o nodes ha will
p o ide he in as uc u e o sha ding. The di e en de elope
eams ha e been wo king o e he las wo yea s o gene a e
E h2 clien implemen a ions, each clien a ge ing a speci ic
ype o use . This a ie y o clien s, oge he wi h he low
compu ing-powe equi ed by he E h2 p o ocol, o e s o mos
o he use s he possibili y o join he ne wo k.
In con as o he E he eum1 clien , he E h2 one is di ided
in wo main pa s, he Beacon Node and he Valida o . The
beacon node, o node, pe o ms all kind o in e ac ions wi h
he E h2 ne wo k. I es ablishes he connec ions wi h o he
nodes, exchanges he messages, downloads he blocks and
main ains he iew o he Beacon Chain as he iew o he
Sha ds ha he alida o is pe o ming a , building among
hem he en i e ne wo k. On he o he hand, he alida o is
in cha ge o he logical pa , gene a ing he a es a ions on he
ecei ed blocks, as well as p oposing he blocks when needed.
To pe o m i s du ies, he alida o needs o ollow he cu en
s a e o he Chain, and o be in cons an communica ion wi h
a Beacon Node ha p o ides such in o ma ion.
A p ojec wi h his complexi y le el comple ely elies on he
heal hiness o he ne wo k. To p opaga e he messages be ween
he nodes o he ne wo k, E h2 elies on he GossipSub
p2p p o ocol [5]. The p o ocol is based on he exchange o
messages and me ada a be ween pee s dis ibu ed in meshes,
achie ing he message p opaga ion wi hin he a ge ed ime-
ma gins and minimizing was e ul bandwid h consump ion.
Gi en he E h2 ne wo k equi emen s and he in e ac ion
be ween he clien s, i is impo an o moni o c i ical poin s o
possible a acks. In his pape , we p esen a comple e analysis
o he p2p ne wo k o he E h2 main ne . The analysis o e s
a gene al o e iew o he ne wo k composi ion, becoming he
i s analysis on he ecen ly launched E h2 main ne .
The emainde o his pape is o ganized as ollows. Sec-
ion II discusses he backg ound and ela ed wo k. Sec ion III
exposes he p o ocols in ol ed on he E h2 p ojec . Sec ion IV
explains he me hodology used o ou s udy and e alua ion.
Sec ion V shows and analyzes he esul s ob ained by ou
ool, including he clien and geog aphical dis ibu ion ind-
ings ha his wo k p esen s. In Sec ion VI we discuss he
possible haza ds ha he E h2 ne wo k could ace. Finally,
Sec ion VII concludes his wo k and p esen s some possible
u u e di ec ions.
II. BACKGROUND
Dis ibu ed applica ions among p2p ne wo ks ha e gained
popula i y wi h he pass o ime o e he las wo decades.
S a ing in he 2000s, dis ibu ed ile sys ems eme ged wi h
he launch o he i s p2p ne wo k, Gnu ella [6]. Since back
hen, he idea o p opaga ing in o ma ion h ough he ne wo k
by using p2p connec ions among pee s has se ed as he base
o de elop mode n dis ibu ed blockchain applica ions such as
Bi coin, E he eum o Mone o.
P e ious wo ks [7] [8] [9] poin ed ha decen alized p2p
ne wo ks by na u e en s o sel -o ganize h ough he usage o
1
© 2021 IEEE. Pe sonal use o his ma e ial is pe mi ed. Pe mission om IEEE mus be ob ained o all o he uses, in any cu en o u u e media, including
ep in ing/ epublishing his ma e ial o ad e ising o p omo ional pu poses,c ea ing new collec i e wo ks, o esale o edis ibu ion ose e s o lis s, o euse
o any copy igh ed componen o his wo k in o he wo ks. DOI: 10.1109/BRAINS52497.2021.9569801
me hods like he Dis ibu ed Hash Tables (DHT) [10], acqui -
ing unique p ope ies and layou s ha can di ec ly a ec he
pe o mance, eliabili y, scalabili y, and some imes anonymi y
o he applica ions ha ely on hem. They emphasized he
impo ance ha he opology o he o e lay ne wo k had,
p oposing me hods ha would help o dig and disco e he
opology o he i s p2p ne wo ks.
As discussed in [11], he size o he p2p ne wo ks con-
s an ly g ows as he size o he in o ma ion ha ge s sha ed,
lea ing he p e ious app oaches and me hods ou da ed. In his
app oach, he au ho s included in he analysis non-in usi e
echniques able o ack: he a ailable nodes o e ime, he
a ic olume, and he hos dis ibu ion in he ne wo k.
Wi h he popula iza ion o blockchain echnology o e he
las decade, new dis ibu ed p o ocols based on blockchain
echnology appea ed, e.g. Bi coin and E he eum. The im-
pac o hese decen alized p o ocols has augmen ed in he
indus ial and academic ields as shown in [12], igge ing
ema kable a en ion on he p2p ne wo ks ha hey ely on.
P e ious wo ks [13] [14] [15] p o ide compelling insigh s
in o he ac ual p2p ne wo ks o he mos p ominen Blockchain
p o ocols, Bi coin and E he eum. In hese wo ks, he au ho s
show me hods o ack he numbe o nodes on he o e lay,
dig in o he pee disco e y o e he ne wo k, and e en some
ligh message exchange o in o ma ion among he pee s. In
his pape , we p esen he p2p ne wo k analysis o e he E h2
main ne pe o med wi h ou de eloped ne wo k c awle ool
ha we name A mia ma.
The p oposed app oach le e ages some p e ious ideas in
[16], [17] and [18], such as ne wo k c awle s as a ool o
exploi possible ne wo k weaknesses ha migh be a haza d
o he in eg i y o he p o ocol. As explained in hose pape s,
he in o ma ion abou he connec ions be ween he nodes ha
o m he ne wo k is as impo an as he in o ma ion abou he
nodes hem-sel . In his app oach, we gene a e a s udy on he
new E h2 p2p ne wo k, including he analysis o :
• he disco e ed opology o he E h2 p2p ne wo k o e lay.
• he dis ibu ion be ween he clien ypes ha a e a ailable
o pa icipa e in he ne wo k.
•ne wo k a ic insigh s om he GossipSub p o ocol.
III. ETH2NETWORK ECOSYSTEM
On a la ge decen alized blockchain ha in ol es so many
nodes and clien s, he secu i y and in eg i y o he p o ocol
elies on he heal hiness o he ne wo k. Being able o debug
om inside all he occu ing e en s, o e s he possibili y o
ge an in-dep h iew o he ne wo k s a us and migh e en help
o p e en pe o mance issues as well as mo e se ious ne wo k
eliabili y p oblems o e en secu i y ulne abili ies. Escaping
om he complex and edious wo k o gene a ing speci ic
es cases o each o he clien s, he communi y ha e ied
o de elop ools o in e ac easily wi h he ne wo k, wi hou
ha ing o ope a e as a ull clien .
A. Rumo
Rumo is an in e ac i e shell sc ip ool o debugging and
es ing he in e ac ion be ween he E h2 clien s. O iginally
s a ed as a E h2 ne wo king ool w i en in Go, Rumo was
designed as an al e na i e o become he common pla o m
o es ing he di e en E h2 nodes in a less e o and
ime-consuming way. By o e ing a se o commands ha
can eplica e mos o he E h2 ne wo k p o ocol in e ac ions,
Rumo o e s he abs ac ion om he code implemen a ion
ha makes se ing a cus om node, possible by simply ipping
a ew commands.
The pla o m is based on an en i onmen close o he
Bash shell ha eases he es building labou , whe e some o
he Bash syn ax (e.g., a iables and con ol low s a emen s)
a e accep ed. Along wi h he sc ip ing capabili ies, Rumo
includes a complex sys em o sub-en i onmen s (known in
Rumo as ac o s) ha allows o gene a e di e en nodes wi h
di e en p ope ies simul aneously. The di e en nodes dis-
ibu ed along he di e en ac o s sha e he common en i on-
men , which pe o m as link be ween hem. This en i onmen
se les he possibili y o he pee s o exchange in o ma ion,
needed o pe o m a synch oniza ion o ac ions based on
cons an ly occu ing e en s. The go e nance o e he node
specs and i s conduc makes Rumo he app op ia e ool o
es ing and debugging he E h2 clien s and hei beha iou
unde di e en condi ions. The a ie y o commands ha
Rumo o e s is s ill on a de eloping s age, bu i al eady
includes mos o E h2 ne wo king specs such:
•Gene a e a ENR iden i y o he node needed o pa ici-
pa e on some o he E h2 p o ocols.
•The E h pee disco e y p o ocol d 5.1.
•Gene a e a pee da abase o pee s o e o he connec ed
and disco e ed pee s.
•Es ablish a connec ion wi h a gi en pee .
•Exchange S a us and Me ada a o he Beacon S a e wi h
he connec ed pee s.
•Lis en and publish messages on he GossipSub p2p
p o ocol used in E h2.
•Send and espond o RPCs om and o he pee s.
The di e en hos s con igu a ions and beha iou s a egies
can be deployed in di e en ways. I he desi ed es is no
oo long, i can be launched om he umo shell by yping
he commands in he desi ed o de . Fo mo e epe i i e asks
o es s, he pla o m o e s he possibili y o ead he lis o
commands om a gi en ile (e.g., sc ip . umo ). Fo mo e
complex asks, Rumo also o e s he op ion o add a es -plan
in o a se ice, allowing he es -case o beha e as a se e .
B. Node Disco e y P o ocol 5
As a decen alized pla o m, E h2 ies o a oid any poin
o cen aliza ion inside he p oposed p o ocol. The ne wo king
a ea, and mo e speci ically he pee disco e y, is no an
excep ion o his ule. The implemen ed GossipSub p o ocol,
despi e being a p o ocol o ien ed o a message p opaga ion
on decen alized applica ions, does no o e any kind o
pee disco e y se ice. Lea ing ha applica ion in cha ge o
his ask. E h2 de eloped i s own node disco e y p o ocol,
Disco e y 5 (d 5), cu en ly on i s 5.1 e sion [19].
2
D 5 ocuses on he E he eum Node Reco ds (ENR) ex-
change along he pee s. These node eco ds a e eco ded in
a DHT. This DTH o e s he possibili y o sample and sea ch
along he gene a ed da abase, allowing o easily upda e he
node eco d o a speci ic pee i modi ica ions a e de ec ed. As
en y poin o he d 5 p o ocol, he nodes can s a sea ching
o o he s ha publicly ad e ise hemsel es on speci ic opics,
o as an al e na i e ha imp o es he pe o mance, a i s
connec ion wi h boo -nodes is possible.
The pee disco e y se ice is essen ial o he ne wo k
in gene al. New pee s joining he ne wo k need connec ions
om whe e synch onize he chain. Pee s ha a e al eady
on he ne wo k also need o upda e om ime o ime hei
in o ma ion sou ce, ensu ing ha hey a e p ope ly ollowing
he head o he chain. Fu he mo e, o inc ease he esilience
o he ne wo k o Sybil a acks, d 5 p o ides a lis o nodes
on he ne wo k ha he node could connec a any poin .
IV. METHODOLOGY
The app oach p oposed in his pape ackles he lack o
in o ma ion abou he pe o mance o he GossipSub p o ocol
and he pee s on he E h2 main ne . The de eloped ool,
A mia ma, is buil on op o Rumo and i o e s a simple
ye powe ul me hod ha p o ides meaning ul da a abou he
p2p ne wo k and he E h2 clien s. The ool is di ided in o wo
pa s, he A mia ma c awle and he A mia ma analyze , as
shown in igu e 1 and desc ibed bellow.
A. A mia ma C awle
A mia ma C awle is he da a ga he e pa o he ool.
The c awle has been based on he E h2 clien debugging
ool Rumo , and i s pe o mance is linked o i s abili y o
disco e and pee wi h nodes om he E h2 ne wo k. The
c awle is a speci ic use case o Rumo , desc ibed in Figu e 1.
On op o Rumo , we build he cus om hos , ha oge he wi h
he modi ica ions compiled in i s eposi o y [20], p o ides a
simpli ica ion o he da a ga he ing, da a p ocessing and es
ep oducibili y p ocesses. The c awle can be di ec ly launched
om he ool, building au onomously he necessa y en i on-
men o ope a e be o e i execu es he chain o commands ha
will ini ialize he cus om hos .
Fig. 1: Scheme o A mia ma. A mimia ma addi ions o Rumo
in blue.
The ini ializa ion p ocess s a s wi h he hos se up assign-
ing he IP and po s ha will be used by he hos . Followed
by he ENR gene a ion, he E h2 ENR iden i y ge s gene a ed
om he hos in o and by speci ying he ne wo k ha we
wan o belong o. Once we ad e ise on he ENR o he
c awle ha we pa icipa e on he E h2 main ne [21], a Beacon
Node is aked. Claiming o be a he genesis s a e p e en s
he c awle om being penalized o no o e ing se ices
om he beacon chain, such as block synch oniza ions. To
success ully ake an E h2 clien and being accep ed by he
es o he pee s, he cus om hos has been se up o be a
he genesis s a e o he Beacon Chain. A e se ing he hos ,
he GossipSub p2p p o ocol is ini ialized, joining he main
opics om he E h2 specs. The PubSub p o ocol [22] Rumo
GossipSub implemen a ion has been modi ied [20] o s o e all
he da a ga he ed om he pee s we ge connec ed o, ge ing
om each o hem:
1) In o ma ion abou he pee : Pee Id, Node Id, Clien
Type and Clien Ve sion, Pubkey, Mul iAdd ess, IP,
Coun y, Ci y and La ency.
2) The connec ion/disconnec ion e en s wi h he imes amp
o each e en .
3) A Coun e o e e y message we ha e ecei ed om each
pee on he i e main GossipSub opics o E h2:
•BeaconBlock
•BeaconAgg ega eAndP oo
•Volun a yExi
•P opose Slashing
•A es e Slashing
Once he hos is ully ini ialized, he pee disco e ing
p o ocol d 5 [19] and he connec all Rumo se ices a e
launched. This way, all he pee s ha we a e able o ind
h ough he d 5 p o ocol a e eco ded in o a pee s o e and
we a emp o connec wi h hem. F om all o hose pee s
ha we es ablish a connec ion, he c awle would s a o ge
he messages om he GossipSub opics p e iously joined,
gene a ing as well he me ics p e iously men ioned. To sa e
he me ics, a new ucn ionali y has been added, o expo ing
he eco ded da a in o he p e iously de ined es -p ojec .
B. A mia ma Analyze
While A mia ma c awle is ocused on he in e ac ion wi h
he p2p ne wo k, eco ding in o a da abase all he e en s
and pee in o ma ion, A mia ma analyze ge s in cha ge o
pe o ming an analysis on he aw in o ma ion ga he ed. A he
momen , he aw da a pa sing ge s done by he c awle i sel ,
expo ing pe iodically he inal me ics in o a cs ile. This
expo ed me ics compiles he indi idually p ocessed da a o
each o he nodes. Du ing his pa sing p ocess, he o ma o
se e al i ems ge s so ed o modi ied, aiming o ease u he he
da a analysis. One o hose o ma modi ica ions co esponds
o he coun o he eco ded connec ions and disconnec ions
e en s, ha se es o ob ain he o al connec ion ime o
each pee . The analysis o he me ics ge s pe o med in a
subsequen momen when he use execu es he commands on
he ool. Fo ha he Analyze will wo k o e he me ics
om he gi en p ojec - olde . In he analysis, he analyze
pe o ms some il e ing and analysis me hods p oducing as
3
Fig. 2: Connec ion s a us dis ibu ion o
he disco e ed pee s.
Fig. 3: E o dis ibu ion on ac i e con-
nec ion a emp s.
Fig. 4: Amoun o ime connec ed wi h
each pee .
esul human- eadable g aphs. The ob ained compila ion o
g aphs and hei insigh s will be discussed in sec ion V.
V. ANALYSIS
In o de o es he capabili ies o he ool, we pe o med an
expe imen in which we le he c awle un o se e al days
collec ing as much da a as possible. The c awle has been se
on a cloud se e wi h limi ed esou ces, ying o eplica e a
low-compu a ional powe en i onmen . The machine hos ing
he ool has 2 In el Co es a 2.3GHz and 7GB o RAM, 50 GB
o s o age, and 250 Mbps o ne wo k bandwid h. The exac
unning pe iod o he expe imen is om 2021-06-19 o 2021-
06-26. The discussion o he esul s ob ained om he analysis
has been di ided in o he h ee subsec ions desc ibed below.
A. E h2 Ne wo k Analysis
In blockchain p o ocols, he ne wo k’s secu i y elies on a
high numbe o nodes wo king oge he . The ne wo k has a
ma gin o pee s ha can go o line wi hou necessa ily leading
o he p o ocol c ash. E h2 needs a leas 66% o he alida o s
ac i ely a es ing blocks o inalize [23].
The i s pa o he analysis consis s o ha ing he c awle
ge a lis o pee s and connec o hem. F om he agg ega ion
o 17516 pee s disco e ed by he d 5 p o ocol o e he se en
days o c awl, he c awle could connec wi h only 8980 o
hem. We we e expec ing a highe connec ion a e. Howe e ,
empo a y disconnec ions o alida o s mig a ing om one
clien o ano he could no be he only eason o he non-
connec ed pee a io. The cu en implemen a ion o he E h2
main clien s does no en o keep he p e ious public keys
o he beacon node whene e his one is eboo ed. This
choice inc eases he p i acy o he alida o hos ed behind
he node bu also inc eases he chances o coun ing no longe
ac i e pee s. I is possible hen ha he p e iously men ioned
8980 connec ed pee s measu emen includes double-coun ed
o eboo ed pee s. The obse ed dis ibu ion could also be
a ec ed by a non-po o wa ding con igu a ion by some o
he non-connec ed pee s. Those pee s s ay unde he p o ec ion
o he ISP i ewalls, p e en ing us om ini ia ing he pee ing
handshake.
A he momen , we es ima e ha a ound 5000 unique ac i e
nodes a e pa icipa ing in he ne wo k. Howe e , om he
pee s we managed o connec wi h, only 4415 o he success ul
connec ions we e ac i ely s a ed by he c awle , while he
o he 4565 o he pee ings we e incoming ones, as we can
app ecia e in Figu e 2. As p e ious wo ks [8] [10] poin s ou ,
he base o some p2p ne wo k i ’s likely o ely on a small
po ion o pee s om he en i e ne wo k. O en called co e-
pee s o supe -pee s, hese pee s compound he nucleus o
he ne wo k. These a e pee s ha i is easy o connec wi h
and ansmi in o ma ion eliably. The 4415 pee s we managed
o connec and exchange da a easily could ep esen a good
po ion o ha nucleus o he ne wo k. P e ious wo k [24]
shows ha in a ne wo k such as he E h2 (based on PoS), pee s
mus show s abili y. This is expec ed since he disconnec ion
o alida o s leads o economic slashing. F om now on, we
will e e o hose pee s ha show s abili y as s eady-pee s.
Taking a close look, in Figu es 3, we can obse e ha om
all he pee s we ied o connec , he ones ha ailed epo ed
di e en e o s as a esul o he dialing a emp . In he igu e,
we app ecia e ha 1081 o he pee s ese he connec ion, 4132
o hem epo ed an unce ain e o , while 7812 o he pee s
epo ed a imeou e o in he a emp o dial a connec ion.
We can ca ego ize wo kinds o pee s in his g oup: he
ac i e, eachable, and s able ones, and he ac i e bu no
eachable pee s. The i s ones p o ided mos o he messages
ha we saw. Meanwhile, he second ype could each he
c awle ia oppo unis ic eques s and d op he connec ions
once a eply is ecei ed. Fu he mo e, we no e ha he c awle
does no ul ill all he unc ionali ies o a ull node. The e o e,
eques s o chain in o ma ion o blocks a e di ec ly d opped
om ou side. Despi e his appa en di e ence be ween he
pee ypes, we also see in Figu e 4 di e ences in he ime
we ha e been connec ed o he pee s. These a ia ions a e
ela ed o he p uning p ocess o he GossipSub p o ocol. This
p ocess aims o keep a p ede ined numbe o pee connec ions.
Se e al clien implemen a ions educe he maximum numbe
o pee s he node can connec o keep he compu ing powe
and ne wo k bandwid h low.
Blockchain ne wo ks a e geog aphically dis ibu ed all
a ound he wo ld. Ha ing he pee s dis ibu ed means ha
he pee s a e less likely o ge a ec ed by he same ype
o egional p oblems o e en s (e.g., Na ional censo ship).
In Figu e 5, we can app ecia e ha almos hal o he pee
dis ibu ion concen a es in wo coun ies, he US (32,65%)
and Ge many (10,37%). No e ha o isualiza ion pu poses
he igu e only displays coun ies wi h mo e han 100 pee s
4
Fig. 5: Numbe o pee s classi ied by
coun y o o igin.
Fig. 6: Numbe o pee s connec ed om
each clien .
Fig. 7: Ex apola ion o he Clien ype
dis ibu ion.
o e he 97 coun ies compiling he o al lis .
In a decen alized ne wo k such as E h2, low la ency
connec ions a e essen ial o acking e e y occu ing e en
as as as possible. In his case, on he E h2 Beacon Chain
p o ocol, he e is a new Slo e en e e y 12 seconds, which
means ha e e y 12 seconds, a andomly chosen alida o
will ha e he chance o p opose a Beacon Block and he e o e
ecei e he oken ewa d. In Figu e 8, we can obse e he
Round T ip Time (RTT) o he la es connec ions ha he
c awle eco ded while being connec ed o he pee s. The RTT
measu es he ime du a ion ha i akes since we send a eques
o he des ina ion pee un il we ecei e a esponse. The la ency
be ween he des ina ion pee and us is hal o he RTT o lowe .
We can obse e ha he majo i y o pee s s ay wi h an RTT
below h ee seconds1, ha ing an ou lie wi h a 67.049 seconds
RTT. No e ha he c awling p ocess was done om a node
loca ed in F ank u and ha he RTT includes he ime he
eques e ook o p ocess he eques . Thus, pee s si ua ed on
he o he side o he wo ld, and pee s wi h a highe wo k load
would usually epo a highe la ency, e.g., he ou lie pee
wi h an IP om he D¨
usseldo (See Sec ion V-C).
B. E h2 Clien s In e ac ion
Cu en ly, he e a e i e E h2 clien s a ailable o pa icipa e
on he ne wo k. Wi h ou ool, we we e able o de ec which
clien is used by each connec ed pee , and i is e en possible
o ge he e sion hey a e using, as shown in Figu e 6. No e
ha unknown pee s could pa icipa e on he ne wo k as boo
nodes, o he c awle s, possible a acke s, o jus pee s ha did
no epo hei clien in he me ada a. Also, we could see
how many e sions o each clien a e ou he e in he wild, as
shown in Table I. This is use ul because we can obse e o e
a long pe iod how clien eams deploy new e sions and how
as use s upda e hei clien s.
As we can see in Figu e 6, P ysm is he mos connec ed
clien wi h 82.9% o he connec ed pee s (7446 pee s). In
con as , in he Pee s o e, we obse e ha 62.84% o he pee s
use he TCP po used by P ysm clien s (13000), showing ha
mos nodes in he E h2 ne wo k a e using P ysm Clien s, bu
in a lowe pe cen age ha he one obse ed by he c awle . 2
1No e ha some pee s do no ha e any measu ed RTT assigned because
he c awle didno ecei e any answe om hem.
2Al hough i is possible o change he dialing po o he hos in all he
clien s, he pe cen age o use s ha change i o his one is low.
The e a e wo main easons o his dispa i y. Fi s , he igu e
does no show all he pee s he c awle disco e ed, bu only
hose i managed o exchange pee me ada a wi h.
Finally, issues ela ed o po - o wa ding also played a ole
in p e en ing some connec ions. To be e unde s and how he
clien dis ibu ion ep esen s he ne wo k, we calcula ed an
es ima ion based on he ad e ised po and he obse ed dis-
ibu ion om hose pee s wi h whom we exchanged me ada a.
As shown in Figu e 7, we can see ha P ysm is he mos
dominan clien in he ne wo k, ollowed by Ligh house and
Teku. This la ge dispa i y and lack o so wa e decen aliza ion
is somehow conce ning (See Sec ion VI).
TABLE I: Numbe o e sions obse ed o each clien .
Ligh house Teku Nimbus P ysm Lodes a Unknown
17 17 1 29 1 1
On he beha io compa ison o he di e en clien s wi h he
c awle , we can obse e se e al di e ences. Fi s , we compa e
he a e age numbe o connec ions and disconnec ions o each
one o he clien s, as shown in Figu es 9 and 10. Please no e
ha he ool was no launched, p io i izing he quali y o he
connec ions. I was p io i izing a la ge numbe o connec ions
om he Pee s o e nodes. We can app ecia e how he clien
dis ibu ions ge mi o ed in he numbe o dialed connec ions
and disconnec ions in he igu es. P ysm is he clien ype wi h
a highe a e age in he pe cei ed dialing e en s om hose
iden i ied pee s o he same clien , ollowed by Ligh house,
Teku, Nimbus, and Lodes a in ha o de .
Howe e , he numbe o connec ions does no equa e o con-
nec ed ime because clien s could a emp mul iple connec ions
while s ill disconnec ing sho ly a e he connec ion has been
es ablished. The e o e, we keep ack o bo h connec ions and
disconnec ions, allowing us o ha e accu a e o al and a e age
connec ion imes. Thus, we plo in Figu e 11 he a e age
amoun o ime (in minu es) ha each ype o clien was
connec ed o ou ne wo k moni o ing ool.
We can obse e ha he esul s ha Figu es 9 and 10 do
no ma ch di ec ly wi h he one obse ed in Figu e 11. While
P ysm was showing a high le el o ac i i y on he dialing
e en s, he a e age ime spen connec ed o ou c awle ge s
educed o he second posi ion. A simila ela ion can be ob-
se ed in he o he clien s wi h he clea excep ion o Lodes a .
5

Fig. 8: RTT dis ibu ion wi h pee s. Fig. 9: A e age connec ions o pee s
classi ied by clien ypes.
Fig. 10: A e age disconnec ions o pee s
classi ied by clien .
The su p ising di e ence be ween he pe cei ed dialing e en s
and he a e age connec ed ime o Lodes a s ands up. Despi e
ha ing a low connec ion a io, Lodes a pee s showed a high
le el o s abili y on he pe o med connec ions. Se e al aspec s
could in luence he obse ed phenomenon.
Fi s , we should poin ou he di e en pee ing s a egies
adop ed by he clien s while implemen ing he GossipSub
p o ocol. While some clien s can adop mo e s ic app oaches
o make su e hei pee s a e con ibu ing o pass in o ma ion,
o he s like Lodes a seem o ha e mo e lexible app oaches,
allowing o long connec ion pe iods.
In his case, we can see ha he pee ing s a egy we selec ed
o he ool also added a bias o he a e ages. By p io i izing
he numbe o connec ions o e hei quali y, he ool d ops he
chance o connec o a s able pee i mo e uniden i ied pee s
a e in he Pee s o e. Thus, p io i izing o new pee s ins ead
o s able ones, plus he numbe o incoming connec ions,
in luences he connec ed ime obse ed o each clien .
On he o he hand, compu ing he a e age o he me ics
could also a ec he obse ed esul s. The connec ions, dis-
connec ions, and he connec ed ime ge calcula ed om he
agg ega ed connec ions and disconnec ions o all he nodes
om a clien ype, di ided by he numbe o nodes om he
same ype. The me hod used o ob ain he a e ages leads
o dispe se any ou s anding pee om he mos connec ed
clien ypes. Meanwhile, o he lowes dialed clien ypes,
his a e age can ge biased as i happens wi h Lodes a .
The combina ion o he low numbe o connec ion e en s
and he low connec ed ime on a e age o he P ysm nodes
leads us o hink ha many P ysm nodes a e behind non-
o wa ded po s o ISP i ewalls. A he same ime, i also
leads us o belie e ha incoming connec ions end o be mo e
oppo unis ic han s able connec ions in he gene al e m.
The ime connec ed o a pee does p o ide some insigh
in o he s abili y o he clien . Some clien s migh be mo e
agg essi e han o he s p uning pee s o a oid bad ac o s.
Howe e , his does no say much abou he di e en clien s’
pa icipa ion in he ne wo k. Fo ha , we measu e he o al
numbe o Beacon Block, and Agg ega e and P oo messages
ecei ed om each pee , g ouping hem by ype o clien .
As obse ed in Figu es 12 and 13 and as i was expec ed, he
clien s ha in a e age ha e mo e s able connec ions ends o
sha e mo e messages. In an a emp o unde s and a li le be e
why Lodes a , P ysm, and Ligh house a e he mos cha y o
he clien s, we s udied he a e age RTT o each clien . In
Figu e 14 we can app ecia e ha Nimbus go o e x4 RTT
and o e x18 RTT o Lodes a wi h espec o he es o
he clien s. As p e iously men ioned, pee s like he Lodes a
ou lie exposed in Table II can signi ican ly impac he a e age
RTT o he clien s, in pa icula when a clien has only a ew
pee s connec ed o he c awle .
Lea ing he ype o clien aside, we also wan ed o see i
all pee s communica e in ela i ely simila amoun s o i any
node is communica ing signi ican ly mo e han o he s. Thus,
we plo in Figu e 15 he numbe o beacon Block messages
ecei ed om each pee .
One o he modi ica ions ha make his analysis unique is
he modi ica ion ha A mia ma implemen s on he GossipSub
p2p p o ocol. The o iginal p o ocol sa es ne wo k bandwid h
consump ion by only sending he me ada a o a message wi h
nodes ou side he mesh be o e sending he whole message. I
he ecei e has no seen he message be o e, hen i asks o
he en i e message o be sen , i i has seen i be o e, hen i
does no ask o i . In addi ion o ha , e en o pee s inside
he mesh, i he message has al eady been seen, GossipSub
does no o wa d his o he clien applica ion. Thus, u u e
messages ha ha e been seen om di e en pee s will no be
ecei ed. Ou modi ica ion on he GossipSub p o ocol allows
o wa ding he message o he applica ion e en i he message
was seen be o e, allowing he c awle o keep ack o all he
ne wo k a ic wi hou inc easing i .
In he esul s we can app ecia e which nodes ake he ole o
s eady-pee s, si ua ing hemsel es in he cen e o he o e lay.
As we can see, wo pee s communica ed o e 2000 Beacon
blocks, while he la ge majo i y o he pee s did no sha e
mo e han 100. Mo e p ecisely, 90% o he BeaconBlocks
we e ecei ed only by 0.31% o he pee s, while 99.69% o
he pee s ansmi ed less han 15 Beacon Blocks. The high
numbe o ne wo k a ic pe cei ed om jus he 0.31% o
he ne wo k highligh s how he E h2 p2p ne wo k s ill sha es
he same ne wo k opology as he ones p e iously men ioned
on Sec ion II. This opology also explains why some nodes
a e mo e cha y and sha e as e han o he s.
In addi ion, we analyzed how many messages pee s send
conce ning he ime hey keep connec ed. A node ansmi ing
oo many messages in a sho connec ion ime could signal a
deny o se ice (DoS) a ack. In con as , a node wi h a e y
6
Fig. 11: A e age ime connec ed o pee s
om each clien .
Fig. 12: A e age o messages ecei ed
on he BeaconBlock opic om pee s
il e ed by clien s.
Fig. 13: A e age o messages ecei ed
on he BeaconAgg ega eAndP oo opic
om pee s il e ed by clien s.
Fig. 14: A e age Round T ip Time (RTT)
classi ied by clien ypes.
Fig. 15: Numbe o Beacon Blocks Re-
cei ed om each Pee .
Fig. 16: To al messages o Connec ed
Time.
long connec ion bu e y ew o no messages could signal
an a emp o an eclipse a ack. The dis ibu ion is shown in
Figu e 16. No e ha in his igu e, we coun Beacon Blocks as
well as P oo Agg ega ions messages. The dis ibu ion shows
a dense cloud o pee s ha s ayed connec ed be ween 0.01
and 1 minu es and ansmi ed be ween 0 and 1.000 messages.
Pa icula ly in e es ing is he case o he do in he op igh
o he igu e. This pee ansmi ed almos 230,000 messages
and s ayed connec ed o a ound 1.5 hou s. We could call his
as eady-pee because i shows an ex emely s able connec ion
and pa icipa ion on he beacon da a ansmission.
TABLE II: Example o Ga he ed In o ma ion om a Pee
Da a Type In o ma ion
Clien Type Lodes a
Clien Ve sion 0.30.2
Coun y Ge many
Ci y D¨
usseldo
RTT 67.049
Connec ions 1
Disconnec ions 1
Connec ed Time(min) 86
Beacon Blocks 430
Beacon Agg ega ions 42508
To al Messages 42938
C. Indi idual Pee Analysis
The p e ious analysis pe o med on he E h2 ne wo k shows
a global iew o he ne wo k. The ga he ed da a can be used o
analyze pee s indi idually. We could ex ac a la ge amoun o
in o ma ion ega ding speci ic pee s. To showcase an example,
we show on Table II he da a ex ac ed om one o he pee s
in he ne wo k. We ha e omi ed sensi i e in o ma ion o
secu i y and p i acy easons as pee Id, node Id, he public
key, and IP ela ed o he pee . This pee showed an unusually
high RTT, which caugh ou a en ion. Combining he ee
analysis ha we p esen ed can help us moni o and keep ack
o nodes suspec ed o malicious beha io . This demons a es
ha he used echniques and he ool could be used o secu i y
pu poses and he ne wo k s a us s udy.
VI. DISCUSSION
This pape p esen s he obse ed geog aphical and clien
dis ibu ion in he ecen ly launched E h2 Beacon Chain. We
ound ha al hough he pla o m incen i izes decen aliza ion
o secu i y pu poses, decen aliza ion is no en i ely achie ed
ye . The ne wo k c awle has iden i ied ha almos 43% o he
17,516 c awled nodes in he ne wo k a e loca ed in he US
and Ge many, ep esen ing he opposi e o a geog aphically
decen alized ne wo k. P e ious wo ks [25] ha e demons a ed
ha he ne wo king pe o mance comple ely elies on he
loca ion o he node. This means ha nodes in egions a om
he men ioned wo coun ies could be in disad an age, which
could in u n lead o some le el o geog aphical cen aliza ion.
In addi ion o he geog aphical dis ibu ion issue, he p e-
sen ed wo k displays ha he ne wo k could be acing a c i ical
mono-clien dependency. The pe o med es ima ions show ha
abou 62% o he ne wo k elies on he same single clien
ype, which could mean a inaliza ion c isis i a ulne abili y
o bug is ound in his pa icula clien . This is one o he mos
c i ical poin s his c awle analysis highligh s, because a bug
impac ed ha speci ic clien in Ap il 24 h causing inaliza ion
issues du ing mul iple hou s [26].
7
We also analyzed he numbe o alida o s in ela ion o
he numbe o Beacon nodes, and in ela ion o he numbe
o IP add esses obse ed in he ne wo k. Wi h 179,825 ac i e
alida o s and 17,516 c awled pee s in E h2 a he momen ,
we can es ima e ha he e a e a ound 10 alida o s on a e age
hos ed behind each o he Beacon nodes. We also obse e
ha om he 6635 success ully iden i ied pee s (37% o he
pee s o e), 530 sha e he same IP add ess wi h a leas one
mo e node. In ac , we obse ed ha 100 o he iden i ied
IPs hos mo e han 4 Beacon nodes a he same ime. So a ,
he analysis ep esen s ha 1.5% o he iden i ied IPs hos
9.8% o he iden i ied alida o s. This again, poin s o a le el
o decen aliza ion lowe han he ideal one a ge ed wi h he
ansi ion om PoW o PoS.
O e all, his c awle analysis highligh s se e al conce ning
poin s o he ecen ly launched E h2 Beacon Chain. F om
se e al pe spec i es, geog aphical-wise, implemen a ion-wise
as well as he physical hos ing o he E h2 alida o s, we ha e
obse ed a no ideal le el o decen aliza ion. The a ge goal
was ha anyone, anywhe e wi h a low-powe de ice, such as
Raspbe y Pi, could pa icipa e in he E h2 ne wo k. This is
echnically easible oday, howe e in p ac ice we obse e a
comple ely di e en pic u e. One eason o his could be ha
he ne wo k is s ill e y young and E h2 has no been ully
deployed ye . Howe e , we belie e i is impo an o boos he
e o s o gua an ee a much highe le el o decen aliza ion.
VII. CONCLUSION AND FUTURE WORK
In his pape , we ha e p esen ed some o he esul s we
ex ac ed om he E h2 p2p ne wo k analysis. The collec ed
in o ma ion, anspa en ly ga he ed wi h ou ne wo k moni o -
ing ool A mia ma, allowed us o ge a solid o e iew o he
ne wo k s a e. Ou esul s show he heal hy beha io in he
ne wo k bu also some conce ns ega ding decen aliza ion.
The in oduced s udy iden i ied ce ain geog aphical, clien ,
and possible IP/En i ies cen aliza ion deg ees ha could mean
a po en ial haza d a any poin o he ne wo ks’ pe o mance.
This wo k in ends o epo he inconsis encies ha he ne -
wo k is acing on i s pa h owa ds his new E h2.
E en hough he p esen ed analysis p o ided meaning ul
ne wo k insigh s abou he E he eum2 ne wo ks’ opology, we
a e awa e o he cu en limi a ions ha he A mia ma ool
has. Ou u u e plan is o ex end he analysis wi h a deepe
ack o he ecei ed messages, op imize he cu en ly used
me hods o achie e a la ge numbe o iden i ied pee s, and
upg ade he ool owa ds a dis ibu ed c awling s a egy ha
would allow us o pe cei e mo e accu a e pee in o ma ion.
ACKNOWLEDGMENT
This wo k has been suppo ed by he E he eum Founda-
ion unde G an FY20-0198. We would like o hank he
esea che s o he E he eum Founda ion o hei eedback and
sugges ions. In pa icula , we would like o hank @P o o-
lambda as he main Rumo main aine o his help se ing up
ou expe imen s and his cons uc i e eedback on his s udy.
REFERENCES
[1] “E he eum whi epape ,” h ps://e he eum.o g/en/whi epape /.
[2] E. Hildenb and , M. Saxena, N. Rod igues, X. Zhu, P. Daian, D. Gu h,
B. Moo e, D. Pa k, Y. Zhang, A. S e anescu e al., “Ke m: A comple e
o mal seman ics o he e he eum i ual machine,” in 2018 IEEE 31s
Compu e Secu i y Founda ions Symposium (CSF). IEEE, 2018, pp.
204–217.
[3] M. Woh e and U. Zdun, “Sma con ac s: secu i y pa e ns in he
e he eum ecosys em and solidi y,” in 2018 In e na ional Wo kshop on
Blockchain O ien ed So wa e Enginee ing (IWBOSE). IEEE, 2018,
pp. 2–8.
[4] “Phase-0,” h ps://no es.e he eum.o g/@dj wo/Bkn3zpwxB.
[5] D. Vyzo i is and Y. Psa as, “Gossipsub: A secu e pubsub p o ocol o
uns uc u ed, decen alised p2p o e lays,” 2019.
[6] M. Ripeanu, “Pee - o-pee a chi ec u e case s udy: Gnu ella ne wo k,”
in P oceedings i s in e na ional con e ence on pee - o-pee compu ing.
IEEE, 2001, pp. 99–100.
[7] R. Ma ei, A. Iamni chi, and P. Fos e , “Mapping he gnu ella ne wo k,”
IEEE In e ne Compu ing, ol. 6, no. 1, pp. 50–57, 2002.
[8] S. Sen and J. Wang, “Analyzing pee - o-pee a ic ac oss la ge ne -
wo ks,” in P oceedings o he 2nd ACM SIGCOMM Wo kshop on
In e ne measu men , 2002, pp. 137–150.
[9] M. Jo ano i´
c, F. Annexs ein, and K. Be man, “Modeling pee - o-pee
ne wo k opologies h ough small-wo ld models and powe laws,” in IX
Telecommunica ions Fo um, TELFOR. Ci esee , 2001, pp. 1–4.
[10] M. Cas o, P. D uschel, Y. C. Hu, and A. Rows on, “Exploi ing ne wo k
p oximi y in dis ibu ed hash ables,” in In e na ional Wo kshop on
Fu u e Di ec ions in Dis ibu ed Compu ing (FuDiCo). Ci esee , 2002,
pp. 52–55.
[11] D. S u zbach, R. Rejaie, and S. Sen, “Cha ac e izing uns uc u ed
o e lay opologies in mode n p2p ile-sha ing sys ems,” IEEE/ACM
T ansac ions on Ne wo king, ol. 16, no. 2, pp. 267–280, 2008.
[12] B. K. Mohan a, D. Jena, S. S. Panda, and S. Sobhanayak, “Blockchain
echnology: A su ey on applica ions and secu i y p i acy challenges,”
In e ne o Things, ol. 8, p. 100107, 2019.
[13] S. Delgado-Segu a, C. P´
e ez-Sol`
a, J. He e a-Joancoma ´
ı, G. Na a o-
A ibas, and J. Bo ell, “C yp ocu ency ne wo ks: A new p2p
pa adigm,” Mobile In o ma ion Sys ems, ol. 2018, 2018.
[14] V. Deshpande, H. Badis, and L. Geo ge, “B cmap: Mapping bi coin
pee - o-pee ne wo k opology,” in 2018 IFIP/IEEE In e na ional Con-
e ence on Pe o mance E alua ion and Modeling in Wi ed and Wi eless
Ne wo ks (PEMWN). IEEE, 2018, pp. 1–6.
[15] T. Wang, C. Zhao, Q. Yang, and S. Zhang, “E hna: Analyzing he
unde lying pee - o-pee ne wo k o he e he eum blockchain,” a Xi
p ep in a Xi :2010.01373, 2020.
[16] S. Ben Ma iem, P. Casas, and B. Donne , “Vi isec ing blockchain p2p
ne wo ks: Un eiling he bi coin ip ne wo k,” in ACM CoNEXT s uden
wo kshop, 2018.
[17] S. K. Kim, Z. Ma, S. Mu ali, J. Mason, A. Mille , and M. Bailey,
“Measu ing e he eum ne wo k pee s,” in P oceedings o he In e ne
Measu emen Con e ence 2018, 2018, pp. 91–104.
[18] A. Bi yuko , D. Kho a o ich, and I. Pus oga o , “Deanonymisa ion
o clien s in bi coin p2p ne wo k,” in P oceedings o he 2014 ACM
SIGSAC Con e ence on Compu e and Communica ions Secu i y, 2014,
pp. 15–29.
[19] “Gene al specs o he d 5 p o ocol,”
h ps://gi hub.com/e he eum/de p2p/blob/mas e /disc 5/disc 5.md.
[20] “A mia ma-c awle eposi o y:,” h ps://gi hub.com/Co ze/a mia ma.
[21] “E h2 main ne in o,” h ps:h ps://gi hub.com/e h2-clien s/e h2-mainne .
[22] “Libp2p pubsub-gossipsub o k:,” h ps://gi hub.com/Co ze/go-libp2p-
pubsub/ ee/a mia ma.
[23] V. Bu e in and V. G i i h, “Caspe he iendly inali y gadge ,” a Xi
p ep in a Xi :1710.09437, 2017.
[24] A. Kiayias, A. Russell, B. Da id, and R. Oliynyko , “Ou obo os: A
p o ably secu e p oo -o -s ake blockchain p o ocol,” in Annual In e na-
ional C yp ology Con e ence. Sp inge , 2017, pp. 357–388.
[25] L. Ki e , A. Salman, D. Le in, A. Mislo e, and C. Ni a-Ro a u, “Unde
he hood o he e he eum gossip p o ocol,” in P oceedings o he 2021
In e na ional Con e ence on Financial C yp og aphy and Da a Secu i y
(FC’21), 2021.
[26] “E h2 mainne inciden e ospec i e,” h ps://medium.com/p ysma ic-
labs/e h2-mainne -inciden - e ospec i e- 0338814340c.
8