Leveraging deployment models on low-resource devices for cloud services in community networks

Le e aging Low-powe De ices o Cloud Se ices in Communi y Ne wo ks Nuno Apol´ onia, Roshan Seda , Felix F ei ag, Leand o Na a o Uni e si a Poli ` ecnica de Ca alunya, Ba celonaTECH Ba celona, Spain {apolonia, seda , elix, leand o}@ac.upc.edu Abs ac —Communi y ne wo ks a e IP ne wo ks cons an ly being imp o ed ha e ol e in o la ge-scale compu ing pla o ms. This has esul ed om he e o o adap he cloud compu ing model owa ds se ices ha can ope a e and u ilize he esou ces inside he communi y ne wo k. The ne wo k and i s in as uc- u e a e con ibu ed by indi iduals, companies, o ganiza ions and a e main ained by he communi y i sel . Communi y cloud de ices a e o en low compu ing esou ce de ices, such as home ga eways, wi h limi ed capabili ies. Cu en ly, hese de ices a e con igu ed o un communi y se ices only. This has become a d awback o u he adop ion because o con ibu o ’s di icul y o also use he dona ed cloud de ice o p i a e pu poses. We apply con aine -based i ualiza ion o he p oblem o esou ce sha ing in low-capaci y de ices in o de o c ea e a mul i-pu pose execu ion en i onmen in a single de ice. Thus, a single de ice can be con igu ed o deli e o he use and he communi y a mul i-pu pose en i onmen , such as pe sonal and public, isola ed om one ano he , while p ese ing he communi y cloud se ices. Ou compa a i e analysis wi h he cu en in as uc u e in communi y ne wo ks gi es e idence ha he capabili y o he de ices o un concu en se ices is main ained. Keywo ds: i ualiza ion, communi y ne wo ks, cloud se - ices, compu ing cons ained-de ices I. INTRODUCTION Communi y ne wo ks a e la ge-scale, sel -o ganised and decen alised communica ion in as uc u es buil and ope a ed by he communi y i sel . They a e open, ee and neu al IP ne wo ks. As o now, hund eds o communi y ne wo ks a e ope a ed ac oss he wo ld, which a e geog aphically dis ibu ed in di e en pa s o he wo ld wi hou elying on any speci ic social o economic easons. The la ge ne wo ks ha e om 500 o 28,000 nodes, such as Gui i.ne 1, FunkFeue 2, AWMN3, F ei unk4. The in as uc u e is con ibu ed by indi iduals, companies and o ganiza ions in a join e o . Resou ce sha ing wi hin he communi y ne wo ks e e in p ac ice o he sha ing o ne wo k bandwid h om each de ice. This enables a ic om de ices o be ou ed h ough o he s o i s des ina ion. The sha ing o se ices, such as ideo s eaming, s o age, VoIP, which h ough cloud compu ing ha e become common p ac ice in he In e ne , ha dly exis s in communi y ne wo ks. The communi y cloud model could sui o accommoda e se ices and/o esou ce sha ing among communi y membe s. The CONFINE p ojec 5was ini ia ed o ad ance he unde - s anding o he communi y ne wo k model, o explo e he ways 1h p://gui i.ne 2h p://www. unk eue .a 3h p://www.awmn.g 4h p:// ei unk.ne / 5h p://con ine-p ojec .eu o imp o ing he use s quali y o expe ience in a sha ed pla - o m and o measu e he sus ainabili y o he ne wo k. Fo his pu pose, he CONFINE p ojec has de eloped an in as uc u e called Communi y-Lab [1], which p o ides an en i onmen o pe o m communi y ne wo k ela ed expe imen s. The physical de ices in he Communi y-Lab a e mini-PCs and a e low- powe de ices. This complemen s he ene gy-e icien hos ing o se ices and/o sha ing esou ces among o he communi y membe s. Cu en ly, he Communi y-Lab in as uc u e has a limi ed numbe o physical de ices, he e o e he scalabili y o expe - imen a ion is limi ed. Besides, all hese de ices a e con igu ed o pe o m only communi y cloud se ices and his has become a d awback o he de ice owne s. The esul is ha he owne s o he de ices ha e been es ic ed om accessing esou ces o hei p i a e pu poses. We add ess his limi a ion by c ea ing an en i onmen in a single de ice, which can bene i bo h he communi y and he de ice owne . Consequen ly, his p oduces a mul i-pu pose en i onmen in a single de ice, such as one en i onmen o he de ice owne and o he en i onmen s o be sha ed wi h he communi y ne wo k. Ou main con ibu ions include: •The c ea ion o con aine -based esou ce i ualiza ion on op o low-powe de ices, enabling a mul i-pu pose en i onmen isola ed om each o he ; •Use s can sha e a po ion o he de ice esou ces, p ese ing he owne s mul i-se ice alloca ed i ual space in he de ices; •E alua ion o he pe o mance o de ices when un- ning se ices on he i ual en i onmen s, and compa - ison wi h he cu en se ings o he Communi y-Lab de ices. In o de o alida e he p oposed app oach, we c ea e a small scale physical Communi y-Lab in as uc u e using se e al low-powe de ices oge he as compu e and s o age de ices and, sepa a ely, deploy he Communi y-Lab con olle in a i ual machine inside a desk op PC. In his expe imen al sys em we deploy wo applica ions, Tahoe-LAFS [2], [3] and Pee S eame [4], [5], as communi y cloud se ices and mea- su e he pe o mance o hese applica ions in he en i onmen gi en by ou app oach. The es o he pape is o ganised as ollows. Sec ion II explains he Communi y cloud and ela es o he CONFINE p ojec , including he o ma ion o Communi y-Lab es bed wi h i s unc ionali y and he echnologies ha a e being deployed. Sec ion III desc ibes he sys em a chi ec u e and he deploymen app oaches. Sec ion IV e e s o he expe imen al se up used. The e alua ion and esul s a e p esen ed in sec ion © 2016 Else ie . This manusc ip e sion is made a ailable unde he CC-BY-NC-ND 4.0 license h p://c ea i ecommons.o g/licenses/by-nc-nd/4.0/ V. The ela ed wo k is desc ibed in sec ion VI. Conclusion and u u e wo k is explained in sec ion VII. II. COMMUNITY CLOUDS Communi y clouds a e o med by a collabo a i e e o o c ea e a compu ing pla o m whe e he in as uc u e is sha ed be ween a numbe o o ganiza ions in o de o p o ide a pla o m o common compu ing conce ns. One pa icula objec i e is o p o ide a scalable compu ing pla o m o conduc communi y ne wo k ela ed esea ch. In his sec ion, we explain he sys em o e iew o he CONFINE Communi y- Lab in as uc u e. Fig. 1. Sys em O e iew o he Communi y-Lab Tes bed A. CONFINE P ojec : Communi y Ne wo ks Tes bed o he Fu u e In e ne The CONFINE p ojec goal is o augmen he capabili ies o communi y ne wo ks by p o iding a pla o m o he exis - ing communi y ne wo k in which use s can use se ices wi h ease [6], as i hey we e deploying in any commodi y cloud pla o m. Membe s o he CONFINE communi y ne wo k es bed a e p i ileged o ge a se o IP add esses (IP 4) in o de o hem o be able o un mul iple se ices, as hey may equi e. This enables membe s o un mul iple se ices on op o he exis ing ne wo k, while sha ing he esou ces wi h he communi y. Communi y-Lab, shown in Fig. 1, is an in as uc u e ha p o ides a se o ools allowing esea che s o easily deploy, un, moni o and expe imen communi y cloud se - ices, p o ocols and applica ions in a eal communi y IP ne wo k (Gui i.ne , FunkFeue , AWMN and F ei unk) ins ead o simula ed en i onmen s. The pla o m is moni o ed by a single en i y named Communi y-Lab con olle , which allows use s o lease he e- sou ces om he ne wo k, and deploy hei expe imen s on he selec ed nodes. Pa icula ly, use s can choose geog aphically dis ibu ed compu ing esou ces h ough he con olle and a e able o cus omize he deploymen acco ding o hei speci ic equi emen s. In addi ion, use s a e allowed o choose he app op ia e con igu a ions o he compu ing esou ces, i.e., o ha e public IP 4 add esses, which can be used o communica e wi hin he communi y ne wo k. Each Communi y-Lab node can con ain se e al sli e s, shown in Fig. 1, which a e g ouped a a highe le el in slices. As such, a slice is de ined as a se o esou ces sp ead ac oss se e al physical de ices in he es bed which allows use s o un expe imen s o e i . A sli e is de ined as he pa i ion o he esou ces (o i ual machine) o a communi y node assigned o a speci ic slice. The pu pose o he con olle is o manage and con ol he es bed h ough simple ope a ions such as managing use s, nodes, slices and sli e s. This con olle p o ides an agg e- ga ion poin whe e membe s can egis e hei de ices as Communi y-Lab nodes. In he web in e ace esea che s can choose geog aphically dispe sed nodes o c ea e slices o hei expe imen s. The nodes e ie e he gi en in o ma ion o deploy local sli e s ac ing as con aine s in he de ices. Whene e use s make a eques o deploy a new sli e he con olle c ea es a Linux con aine on he node by alloca ing he esou ces equi ed o un he new sli e . The e o e each sli e uns on he node isola ed om one ano he . Linux con aine s gua an ee isola ion in e ms o secu i y and esou ces howe e he hos ke nel sys em is sha ed be- ween all con aine s. In his way, use s can deploy many sli e s in a single node o un many se ices concu en ly. To be pa o he Communi y-Lab in as uc u e he de ices equi e o ope a e a speci ic ope a ing sys em, based on Open- W 6con igu ed o p o ide au oma ically an open ne wo k connec ion wi h he Communi y-Lab con olle , becoming pa o he es bed o he expe imen s. These de ices se e as he in as uc u e laye o he Communi y-Lab and mos a e low- esou ce de ices which can be a o dable o ha e a he edges o hese ypes o ne wo ks. B. CLOUDY: Communi y Cloud Dis ibu ion Cloudy Dis ibu ion7(Cloudy OS) has been c ea ed unde he CLOMMUNITY p ojec 8 o p o ide communi y ne wo ks an easy way o manage and deploy cloud in as uc u es and in e aces o se ice disco e y and deploymen . The esul is ha any use can enjoy he bene i s o cloud se ices which a e eely a ailable in he communi y wi hou elying on any speci ic cloud in as uc u e. The Cloudy OS is a ee and open sou ce so wa e and is a cus omized e sion o Debian Linux. By de aul , i comes wi h an ins alla ion o inc9VPN daemon which c ea es a secu ed p i a e o e lay ne wo k be ween hos s on he In e ne . Wi h he help o inc VPN, ne wo ked nodes can communica e secu ely wi h each o he . In addi ion, Cloudy OS uses A ahi10 (o Se 11 in la es e sions), which is a ze o-con igu a ion ne wo king implemen- a ion, o publish and disco e he se ices in he communi y. Fo he simplici y o se ice disco e y, he Cloudy OS p o ides an in e ace ha e ches he se ices in he o e lay and lis s he se ices in o de o he use s o easily connec hem. C. Vi ualiza ion Sys ems Mos o he Communi y ne wo k de ices ha a e used in he Communi y-Lab in as uc u e a e low-powe de ices such as Home ga eways, se -up boxes, esea ch de ices. Howe e hese de ices a e capable o unning mul iple communi y cloud se ices simul aneously. Fo ins ance, he Cloudy OS comes wi h a ew p e-ins alled se ices such as Tahoe-LAFS 12 dis ibu ed ile s o age sys em and Pee S eame 13: P2P ideo 6h p://openw .o g 7h p://cloudy.communi y 8h p://clommuni y-p ojec .eu 9h p://www. inc- pn.o g 10h p://www.a ahi.o g 11h ps://www.se dom.io 12h p:// ahoe-la s.o g 13h p://pee s eame .o g s eaming amewo k. As we discussed p e iously, hese de- ices a e con igu ed o se e only communi y cloud se ices. We can say ha , as an en y-poin , i ualiza ion can gi e us he means o c ea e mul i-pu pose en i onmen s in a single de ice. The e o e, we can s udy wo main i ualiza ion echniques and measu e he sys em beha iou o each case in e ms o pe o mance and complexi y. Use s can choose one o hese mechanisms conside ing he complexi y o he pla o m con igu a ions, sys em pe o mance and he ha dwa e suppo ha he de ices ha e. One ype o i ualiza ion sys em is called Vi ual machines (o machine emula ion) such as QEMU which is an open sou ce machine emula o , used o un i ual machines on op o an ope a ing sys em such as Linux. I is also capable o di ec i ualiza ion when using he KVM (Ke nel-based Vi ual Machine) ke nel module in Linux and ha ing ha dwa e compa ible wi h i ualiza ion echnology. O he wise, i can only emula e machines, and hus he i ual machines c ea ed canno di ec ly access some o he ha dwa e which can p o ide a be e gues pe o mance. Ano he i ualiza ion echnology, also a ailable wi h mos Linux ke nels, is called Linux Con aine s (LXC), and i is compa able o o he i ualiza ion echnologies. Howe e i may lack some o he secu i y and isola ion me hods ha o he i ualiza ion echnologies ha e, such as OpenVZ.14 Also, i can be mo e ligh weigh since i uses he al eady in place ea u es o he Linux ke nels ha adop ed his ype o i ualiza ion. I sepa a es he use con ex o each con aine and main ains a sha ed link o he hos ke nel in o de o un mul iple sys ems in an OS-Le el i ualiza ion me hod. III. SYSTEM ARCHITECTURES The a chi ec u e o ou p oposed sys em includes wo app oaches o i ualiza ion. In he i s app oach we used a i ual machine (QEMU) in o de o sepa a e he con ex in which a se ice can be deployed. In he second app oach we used a LXC based app oach in which he i ualiza ion is ligh weigh in o de o ha e a be e pe o mance while unning se ices. A. Se ices Deploymen The p oposed app oaches, le e aging i ualiza ion e- sou ces, a e enhanced wi h di e en con ex s whe e owne s can sha e (sha ed con ex ) only pa o he de ices esou ces while also accessing hei own (p i a e) con ex , whe e hey can un hei own independen se ices, as demons a ed in Fig. 2. B. QEMU deploymen app oach Ou ini ial app oach includes a deploymen o i ual machines wi h he use o QEMU, ins alled on op o he Cloudy OS. This allows us o concu en ly un se ices while s ill isola ing he de ice o be used by i s owne . In a physical de ice we ins all Cloudy OS whe e we can execu e se e al ins ances o QEMU c ea ing i ual en i onmen s in o de o un mul iple se ices. As an example, Fig. 3 shows he deploymen o his scena io using wo Communi y-Lab nodes and i s sli e s as independen se ices on op o he Cloudy OS. This is done in o de o gain con ol o e he de ice o enable mul iple se ices unning om di e en use con ex s while main aining he owne s’ abili y o execu e his own se ices. 14h p://open z.o g Fig. 2. Example o wo de ice deploymen . De ices a e di ided in wo con ex s o owne and sha ed o he CN. In sha ed con ex se e al sli e s un as nes ed con aine s belonging o each slice deployed. Fig. 3. Example o QEMU deploymen app oach. Two se ices unning (se ing as Communi y-Lab nodes) on i ual machines. LXC uns om wi hin he Communi y-Lab node o c ea e sli e s. The pe o mance o QEMU is g ea ly inc eased when un- ning i wi h ke nel in eg a ion (KVM) o i ualiza ion suppo om he ha dwa e. This gua an ees ha each i ual machine has highe pe o mance and some o he physical esou ces can be di ec ly u ilized by he i ual machines. Howe e in low-powe de ices such op ion may no be a ailable, ins ead QEMU has o emula e he whole p ocess o he i ual machine which hinde s he pe o mance o he i ual machine. Unde his scena io wi hin ce ain condi ions (such as KVM enabled o i ualiza ion suppo ) we can achie e a sepa a ion o se ices and u ilize a physical de ice o be sha ed be ween he owne and he communi y ne wo k. This ype o deploymen is a quick and easy way o os e ing mul iple se ices in a single de ice. Fu he mo e, his app oach can be ai ly enough when se ices need highe isola ion om he hos sys em by enhancing i s secu i y o when he se ice pe o mance is no a ec ed. C. Linux Con aine s deploymen app oach Ou p oposed app oach includes a deploymen o Linux con aine s (LXC). This allows unning concu en se ices wi h a low o e head o he i ual esou ces and p ocesses. LXC c ea es di e en use con ex s in he hos machine o deploy sepa a e sys ems, bu sha es he same ke nel (OS-Le el i ualiza ion) In his app oach we conside he isola ion o he se ices while s ill gua an eeing a highe pe o mance o se ices since he e is no emula ion in ol ed. The i ualiza ion laye conside ed is in he OS-Le el, whe e he hos ke nel is sha ed be ween con aine s and he hos sys em. In Fig. 4 we show an example o he deploymen o his app oach using he Cloudy OS as he hos sys em and deploying a Communi y- Lab node in a LXC con aine . I is wo h men ioning ha he Communi y-Lab nodes deploy hei own sli e s as LXC con aine s and wi h ou app oach his does no change. This is p o ided by uning he con igu a ions o LXC o deploy nes ed con aine s, while adding he AppA mo 15 secu i y policies o he secu i y conce ns. Fig. 4. Example o LXC deploymen app oach. One se ice unning (se ing as Communi y-Lab node) on a con aine i ualiza ion occu ing in OS-le el. Nes ed LXC uns om Communi y-Lab node only con igu a ion is needed o un nes ed con aine s. Wi h his app oach, we can use he de ice as a sha ed en i onmen be ween he owne and he communi y cloud en i onmen main aining he isola ion om each use con ex . The secu i y issues ha a ise om such usage a e espec ully handled by LXC o he Linux secu i y policies. The e o e he con aine s ha e access o he hos ke nel which may p o e o be a mino isola ion o some se ices, equi ing mo e isola ion should be handled wi h he i ual machines app oach. Unde his scena io we can achie e a ligh e isola ion and a concu en access o he physical de ices. This makes he se ices unning wi h highe pe o mance han wi h he QEMU app oach. I is ele an o men ion ha in his app oach he sys em unning in he con aine s only ha e access o he hos ke nel as such he hos ke nel equi es o be compa ible (o wi h he equi ed ke nel modules loaded) in o de o co ec ly un he sys em inside he con aine , i.e. he Communi y-Lab nodes equi e ha he o e layFS ile sys em should be na i ely, enabling i o deploy he eques ed sli e s on he de ices. A main ea u e o ou p oposed deploymen scena io is he in oduc ion o a i ual en i onmen in which se ices a e able o un wi h di e en con ex s and main aining an isola ed pa o he de ice o be used by he owne . As a esul when using he Communi y-Lab node as a se ice each node egis e ed in he Communi y-Lab con olle can deploy i s own sli e s in he physical de ice wi hou in e e ing wi h o he se ices o he owne usage. This enables sha ing o esou ces o a communi y cloud en i onmen and main aining he owne s exclusi e access o he de ice. IV. EXPERIMENTAL SETUP Fo ou expe imen al se up, we used ou physical esea ch de ices wi h di e en con igu a ions. These de ices a e buil wi h In el powe ed A om N2600 CPU p ocesso s. Two o hem ha e 2 GB o RAM, 60 GB s o age and 2 GB o RAM, 120 GB s o age, and he o he de ice has 4 GB o RAM wi h 500 GB o s o age disk and linked o he communi y ne wo k (Gui i.ne ). A desk op compu e se up wi h he P oxmox sys em16 and buil wi h an In el(R) Co e(TM) i7-3770 CPU @ 3.40GHz (8 co es), wi h 1TB o s o age disk and 16 GB o RAM was used o 15h p://wiki.appa mo .ne 16h ps://www.p oxmox.com deploy a local Communi y-Lab con olle in a i ual machine en i onmen . I was necessa y o use such deploymen in o de o no a ec he pe o mance o he cu en Communi y-Lab in as uc u e, which is in a p oduc ion s a e. Fo ou expe imen s, we se up a eplica o he Communi y- Lab es bed a chi ec u e, such as, one local con olle which con ols he o e all sys em and a se o compu ing nodes (Communi y-Lab nodes) ha can be used o deploy sli e s om he con olle . In his case, he local con olle can be deployed ei he in a con aine o in a sepa a e i ual machine ins ance. Howe e , his depends on he use s’ equi emen s and he esou ces which a e a ailable. The eason o he laye ed i ualiza ion in he physical de ices is ha we in end o augmen he cu en se ices o he Cloudy OS such ha he use s can ha e a se ice ha deploys Communi y-Lab nodes in an au oma ic manne . This allows aking ad an age o he i ualiza ion en i onmen in o de o use s o sha e hei esou ces h ough he Communi y- Lab pla o m and ex ending he numbe o nodes p esen in he Communi y-Lab. Fu he mo e we can examine he applicabili y and measu e he e iciency o deploy communi y cloud se ices in o de o unde s and he easibili y o sha ing de ices in communi y ne wo ks while main aining he owne s’ space in he de ice. Challenges: One o he main challenges when c ea ing ou expe imen al en i onmen is o se up he Communi y-Lab con olle in a local en i onmen using a i ual machine, and using Cloudy OS as he base o hos ope a ing sys em. The p ocess o deploymen o he Communi y-Lab con olle in a local en i onmen can be a challenging ask o he common communi y use . The CONFINE P ojec wiki17, howe e , con ains u o ials on how a use can manage o ins all he con olle wi h ease. I is also no ed ha some o he s eps a e dependen on he ope a ing sys em used, and ha i can b eak he communica ion o e lay i no co ec ly con igu ed (i.e. Tinc miscon igu a ion). Ano he challenge aced is he ac ha he cu en Communi y-Lab node sys em equi es speci ic Linux ke nel modules. These mus be enabled in he hos sys em, and wi hou i , he sli e s may no wo k as expec ed and he e o e no unc ion co ec ly when deploying/ unning he sli e s. V. EVALUATION In he e alua ion o he p oposed app oach (LXC deploy- men app oach) we used physical esea ch de ices connec ed o Gui i.ne . These de ices a e low-powe , ha e less esou ces compa ed o desk op PCs, a e mos ly he same de ices used wi hin he Communi y-Lab in as uc u e and a e gene ally simila o he sha ed de ices om communi y ne wo ks. These de ices a e deployed wi h scena io LXC, p e iously desc ibed. A. Expe imen s Ou expe imen s we e pe o med in o de o e alua e he p oposed deploymen , summa ized in Table I, by u ilizing di e en se ices wi h di e en pu poses. In he i s e alu- a ion scena io we used Tahoe-LAFS dis ibu ed s o age, and a s o age benchma k applica ion o e alua e he impac o he p oposed deploymen on he physical de ices. In he second e alua ion scena io we used Pee S eame , a pee - o-pee ideo s eaming applica ion, in o de o e alua e he impac on se ices ha ha e ime sensi i e da a p ocessing. The hi d 17h p://wiki.con ine-p ojec .eu TABLE I. SUMMARY OF OUR SCENARIOS AND SETTINGS Scena io 1 2 3 Numbe o local 8 8 8 Communi y-Lab nodes Numbe o sli e s 8 8 16 Se ices deployed Tahoe-LAFS Pee s eame Tahoe-LAFS and Pee s eame Pe o mance S o age Chunks Recei ed S o age benchma k Me ics benchma k and Played-ou Chunks Recei ed and Played-ou e alua ion scena io combines bo h se ices and allows an e alua ion o he concu ency o se ices wi hin he same physical de ices wi h ou p oposed deploymen . Fo each scena io we collec ed esul s and plo ed hem agains he baseline alues. The baseline alues we e ob ained by unning he same se o expe imen s on he Communi y-Lab in as uc u e ( om ou g oups’ ea lie wo ks [7]) unde he same se o con igu a ions. This gi es us he beha iou o he p oposed sys em in e ms o pe o mance and use expe ience. In he Tahoe-LAFS expe imen s we measu ed he pe o - mance o ead and w i e ope a ions in o de o unde s and he impac hese ype o ope a ions ha e on he p oposed deploymen . In he Pee S eame expe imen s we measu ed he a e age chunk a es (da a ha is ecei ed in he pee s side) and a e age chunks played ou on pee s (da a ha is sen o be wa ched in he pee s side) in o de o measu e he quali y o he ideo s eam. In bo h cases we measu ed he CPU u iliza ion o demons a e ha hese low-powe de ices can deli e he mul i-pu pose execu ion en i onmen while main aining he mul i-se ice communi y cloud model. Fi s Scena io: In ou i s e alua ion scena io, we c ea ed one sli e o each a ailable Commmuni y-Lab node (eigh sli e s in o al) and each sli e unning he Cloudy OS. Fo each o he sli e s we an a Tahoe ins ance, Tahoe-LAFS is a se ice ha comes bundled wi h he Cloudy OS and is used o dis ibu ed s o age. The scena io was es ed in se e al uns using he same con igu a ion o each, ex ending o an amoun o 3 hou s each. We used se en sli e s as Tahoe-LAFS s o age ins ances ( hese ins ances se e as s o age o he iles w i en by any clien ). One o hese sli e s an Tahoe’s In oduce (a publish- subsc ibe hub esponsible o no i y clien s and s o age nodes abou each o he ) and he eigh h sli e ope a es he Tahoe- LAFS clien ins ance which can w i e o ead iles om a moun ed olde ha accesses di ec ly he Tahoe-LAFS sys em. This is done h ough he use o ssh s and use ke nel module18 allowing a olde on he Tahoe-LAFS sys em o become a ailable on any compu e sys em seamlessly. Fu he mo e we an a well-es ablished disk benchma k applica ion, i.e. IOZone [8], o measu e he s o age ope a ions o each node in o de o e alua e he p oposed deploymen and compa e i wi h he pe o mance o he same se ices when using he cu en Communi y-Lab, a p esen ime. Resul s: Fig. 5 shows he measu emen s o he baseline sys em co esponding o he cu en Communi y-Lab en i on- men , and he same se o ope a ions on he p oposed de- ploymen (named se 1). No e ha all ope a ions a e comple ed when he ansac ions be ween ins ances a e inished he e o e he esul s accoun wi h he pe o mance o all Communi y- 18h p:// use.sou ce o ge.ne /ssh s.h ml Fig. 5. Pe o mance o Tahoe-LAFS se ice, baseline as he cu en deploymen and se 1 as wi hin he p oposed deploymen on he i s e alua ion scena io (ope a ions shown as a e age All, w i e, e-w i e, ead, e- ead, among o he s). Rep esen ing a e age, and de ia ions o each ope a ion. 20 40 60 80 100 120 140 160 180 0 10 20 Time (min) CPU U iliza ion(%) Fig. 6. CPU u iliza ion on Tahoe-LAFS clien node in he i s e alua ion scena io Lab nodes used. In he p oposed deploymen he e a e wo se ices unning on he same physical de ices he e o e each Tahoe-LAFS sys em has concu en access o he esou ces which is, as expec ed, e lec ed by a gene al lowe ope a ion speed o all he ope a ions measu ed. Howe e his s ill accomplishes he ope a ions wi h he app oxima ed speeds as he baseline e alua ion. The benchma k applica ion s esses he de ice esou ces in o de o ind he maximum speed o ope a ions such as eading/w i ing o iles. I is impo an o no ice ha while ne wo k pa is an impo an p ocess in dis ibu ing iles h oughou he ins ances, ou e alua ion accoun s mo e o he esou ce usage locally. Thus, CPU u iliza ion is impo an o unde s and he sys em beha iou in he p oposed deploymen . Fig. 6 shows he a e age CPU u iliza ion du ing h ee hou s o unning he IOZone benchma king applica ion in he Tahoe- LAFS clien ins ance. I has consumed a ound 20% o CPU ime on a e age du ing he comple e es . The clien node pe o ms enc yp ion/dec yp ion and e asu e code compu a ion on each ile block in he phase o w i ing and/o eading o/ om he disk. Fu he mo e he Tahoe-LAFS clien ins ance consumes mo e CPU ime as opposed o Tahoe-LAFS s o ages. Second Scena io: In ou second e alua ion scena io, we used he same sli e deploymen (one pe node), unning concu en ly a Cloudy OS empla e. In each o hese sli e s, we an a Pee S eame expe imen . Fo his scena io, we an ou es s in di e en ime pe iods, wi h 1 hou each un, in o de o accoun o di e en ne wo k ac i i y. We used se en sli e s as pee s (each pee e ie es da a om he ne wo k in o de o play ou he s eaming ideo locally). Pee S eame uses an o e lay ne wo k o exchange da a (known as chunks) be ween i s pee s. We also se up one sli e o se e as he sou ce pee which dissemina es he sou ce ideo pa i ioned in chunks (as de aul , one ame o he ideo is one chunk) o be played ou by he o he pee s. The Pee S eame sou ce pee ge s a li e came a s eam and sends he chunks o he o e lay ne wo k be ween each pee ha wa ches ha s eam. This esul s in each pee ying o e ch chunks om o he pee s o play ou he con inuous s eam and display i locally o he use s. Fig. 7. A e age chunks ecei ed a e a pee s om Pee S eame execu ion in he second e alua ion scena io. Baseline as he cu en deploymen in Communi y-Lab. Fig. 8. A e age play-ou a io a pee s om Pee S eame execu ion in he second e alua ion scena io. Baseline as he cu en deploymen in Communi y- Lab. Resul s: Fig. 7 and Fig. 8 depic he measu emen s o he a e age o chunks ha a pee can ecei e and he chunks pe cen age ha we e sen o be played ou a e aged by all pee s. As a baseline we ha e he cu en deploymen om p e ious expe imen s wi h he Communi y-Lab es bed in which he p oposed deploymen is able o each wi hou losing oo much da a on a e age. I is also no ed ha because o he sha ed esou ces he e is a minimum amoun o chunks ha a e no ecei ed in he p oposed deploymen . These ime sensi i e applica ions equi e ha da a should a i e on ime o be displayed/p ocessed, i he da a does no appea in he ime allo ed i is disca ded. The e o e in he p oposed deploymen his amoun does no a y much om he baseline. Fig. 9 shows he a e age measu emen s o CPU u iliza ion du ing an hou , when he se ice uns con inuously. The al e a ions we see in he CPU u iliza ion is in ac because he Pee S eame neighbou hood size changes o e ime ( he o e lay ne wo k is cons an ly upda ed e en when he e a e no new pee s) and he e o e he u iliza ion o he esou ces change when he ope a ions o edoing he opology o he ne wo k is pe o med. Mo eo e he CPU u iliza ion on he sou ce node is highe han on he pee nodes since i anscodes he ideo s eam and pa i ions i in o chunks ha will be sen o he 10 20 30 40 50 60 5 10 15 20 Time (min) CPU U iliza ion(%) Fig. 9. CPU u iliza ion on Pee S eame (PS) sou ce node on second e alua ion scena io pee s. Thus he CPU u iliza ion on he pee s is conside ably lowe and do no in e e e as much wi h o he se ices unning concu en ly. This is a esul o he pee s unning a mo e ligh weigh p ocess such as ga he ing and decoding o chunks. Thi d Scena io: In ou hi d e alua ion scena io, we an bo h se ices (Tahoe-LAFS and Pee S eame ) in sepa a e slices, unning concu en ly in a Cloudy OS empla e on he same local Communi y-Lab nodes. We measu ed he pe o mance o ou p oposed app oach while unning di e en se ices in a concu en way. In his scena io we an he same numbe o expe imen s as be o e, while cu ing down he Tahoe-LAFS es o 1 hou . I also uses he same deploymen o sli e s as be o e, howe e , each physical de ice uns concu en ly ou sli e s each wi h i s own se ice and g oups o wo om he same slice. Fig. 10. Pe o mance o Tahoe-LAFS se ice, baseline as he cu en deploymen and se 1 as wi hin he p oposed deploymen in hi d e alua ion scena io (ope a ions shown as A e age All, w i e, e-w i e, ead, e- ead, among o he s). Rep esen ing a e age, and de ia ions o each ope a ion. Resul s: Fig. 10 shows he esul s om he cu en de- ploymen in Communi y-Lab as baseline, agains he p oposed deploymen in Se 1. I also shows ha when di e en concu - en se ices a e unning on he same de ice, he impac on he se ices a e no iceable, when he e is a highe load o he sys em. Howe e , in a no mal usage o se ices he impac on he ope a ions is a enua ed wi h ime, o wi h scheduling. Fig. 11 and Fig. 12 show he esul s o he cu en de- ploymen on he Communi y-Lab as a baseline agains he p oposed deploymen esul s. We can see a no iceable, o a ce ain deg ee, a ia ion o he chunks played ou which a ec s he ideo quali y pe cei ed. This is due o he CPU ime being sha ed among mo e p ocesses. Howe e , while concu en se ices may di e , o ou esul s we can say ha he loss is minimal when using he p oposed deploymen agains he cu en Communi y-Lab deploymen . The a e age CPU u iliza ion o he Tahoe-LAFS clien ins ance is shown in Fig. 13. These measu emen s we e aken Fig. 11. A e age chunks ecei ed a e a pee s when Tahoe-La s is also unning ( hi d e alua ion scena io). Baseline as he cu en deploymen in Communi y-Lab. Fig. 12. A e age chunk playou a pee s when Tahoe-LAFS is also unning ( hi d e alua ion scena io). Baseline as he cu en deploymen in Communi y- Lab. 10 20 30 40 50 60 100 200 Time (min) CPU U iliza ion(%) PS Tahoe Fig. 13. A e age CPU u iliza ion o Tahoe-LAFS and Pee S eame (PS) in hi d e alua ion scena io o e one hou pe iod in he hi d e alua ion scena io. The igu e shows he CPU u iliza ion o he Tahoe-LAFS clien ins ance is a ound 35% - 40% on a e age o e he cou se o expe imen , meaning ha i uses a mos hal o one co e o he de ice. This is a esul o he added p ocessing in he clien ins ances while he s o age ins ances ha e a lowe p ocess u iliza ion, pe o ming only ead and w i e ope a ions. On he o he hand, he a e age CPU consump ion o Pee S eame is nea ly i e imes highe han wha Tahoe- LAFS uses. This is because while unning he las scena io we also eco ded he ideo o he disk (in all pee s) in luencing mo e he esou ce u iliza ion o he se ice in o de o ge he mos u iliza ion o each se ice. The CPU u iliza ion on he pee s emains lowe han in he sou ce pee and s ill easible o be unning concu en ly wi h o he se ices. B. Discussion We could obse e ha he p oposed deploymen has an impac on he se ice pe o mance. Howe e , we can say ha i is minimal while achie ing mo e se ices wi hin one physical de ice. We can also s a e ha bo h scena ios, LXC and QEMU, can be used acco ding o he esou ces a ailable. QEMU is sui able when se ices equi e highe secu i y ea u es, highe isola ion o he ke nel sys em and when esou ces a e no cons ained (high-powe de ices wi h ha dwa e-enabled i ualiza ion). LXC a ge s low-capaci y de ices o achie e concu ency be ween se ices. The quali y impac ha he se ices ha e is minimum, when using se ices ha do no equi e high demand o he esou ces. In ou expe imen s he esou ce u iliza ion was key o unde s and he impac and how we can op imize he sha ed esou ces. The esul s o ou e alua ion indica e a iabili y on he execu ion o concu en se ices. One o he ac o s o his is he use con ex has o be changed o each p ocess. This can esul in mino delays un il he se ice can ully u ilize he CPU. Mo eo e we can say ha using such deploymen s may gua an ee isola ion o se ices and main ain mos ly he same quali y o se ice. The e o e, ou app oach showed o be easible o he cu en deploymen o de ices on communi y ne wo ks wi h minimal pe o mance loss in he se ices. Also achie ing use con ex isola ion on he de ices o mul iple pu poses and gua an eeing he owne s isola ion and u iliza ion o he de ices. Fu he s udy o achie e and op imize he maximum amoun o con ex s o se ices wi hin a de ice will be deal in u u e wo k. VI. RELATED WORK In his sec ion we e iew some wo ks which add ess simila issues as his pape . In [9] he au ho s epo on how he deploymen o he cloud model on op o IP-based communi y ne wo ks in he case o Gui i.ne was unde aken. They elabo a e a sys em o he p oposi ion ha he use s can bene i om cloud-based se ices inside o he ne wo k wi hou ha ing o consume hem om he In e ne . Ins ead hey can u ilize he esou ces ha al eady exis in hei communi y ne wo k, while also g an ing access o cloud-based se ices. The a ea o Fog compu ing [10] is ela ed o ou wo k in he concep o in eg a ing edge de ices. To his end, Fog compu ing aims o ex end da a cen e -based cloud compu ing by in eg a ing hos s a he edges o he ne wo ks in o he cloud p ocess. Edges a e seen as being p oac i e componen s o se ices, da a usage and s o age. In he Pe sonal clouds p oposal [11], he au ho s enhance he capabili ies o mobile de ices, seen as low-powe ed de- ices, by using ei he emo e o nea by cloud esou ces, ins ead o ha ing o p ocess da a locally and hus educing consump- ion wi hin he mobile de ices. Fu he mo e, his wo k explains how ne wo k esou ces a e o be in eg a ed in a he e ogeneous en i onmen , while enhancing he use expe ience. In his way, each de ice can be seen as a single de ice cloud and ac i e pa icipan s can un he se ices which a e o in e es o he end use s. The wo k in [12] shows how clouds ha ha e unde -u ilized esou ces can be enhanced by sha ing hese esou ces wi h o he communi ies, while s ill main aining he same aspec s ha he cloud owne s ha e ag eed upon. Wo k epo ed in [13] desc ibes echniques ha may sa is y he o load compu a ion o mobile de ices. A compa ison is p o ided o be e unde s and and succeed when doing p ocessing on low-powe ed de ices. This sys em only accoun s o se ice concu ency and no use independen , hus only conside ing he i ualiza ion laye o se ice o un concu - en ly in a low-powe ed de ice. In he Pa ad op sys em [14] he au ho s ha e c ea ed a pla o m in which low powe ed esou ces a e used, such as home ga eways, in o de o deploy di e en se ices unning and p ocessing concu en ly and wi h di e en da a. T ying o ge he mos use ou o such de ices, while also aking ad an age o he pa allelism ha de ices can c ea e be ween compu a ions. This sys em only accoun s o se ice concu - ency and no use independency, hus i only conside s he i ualiza ion laye o se ices o un concu en ly in a low- powe ed de ice. The wo k in [15] demons a es ha con aine -based sys em i ualiza ion is pe o med well o e hype iso s by gi ing he oppo uni y o isola ion in e ms o secu i y and esou ces. Thei esul s show ha con aine -based sys em i ualiza ion p o ides up o 2x he sys em pe o mance o hype iso s o se e - ype wo kloads and scale u he while main aining he sys em pe o mance a a highe le el. VII. CONCLUSION AND OUTLOOK Communi y ne wo ks a e IP-based ne wo ks ha can ha e many nodes ha a e connec ed by wi eless links h ough a e i o y. These ne wo ks we e ini ially s a ed o gi e hei membe s access o he In e ne . We a gue ha hese ne wo ks a e unde u ilized i only In e ne access is a ge ed. The esou ces in he ne wo k can ac ually suppo mo e se ices, which can gi e communi y membe s ad an ages and bene i s when pa icipa ing in sha ed se ices. The con aine -based ligh weigh i ualiza ion app oach p oposed in his pape o con ibu ed low-capaci y de ices, main ains he bene i s o unning concu en se ices on hese low-powe de ices and deli e s a mul i-pu pose sys em. Also i ea u es isola ion o bo h esou ces and secu i y whe eby gua an eeing a be e o e all sys em pe o mance e en on esou ce-cons ained de ices. This pape also add essed he cu en limi a ions o communi y cloud in as uc u es such as he scalabili y o he exis ing in as uc u e, and he pe o - mance o low-powe de ices. We demons a ed ha h ough i ualiza ion we can deli e a mul i-pu pose en i onmen . This leads us o un mul iple se - ices in low-powe de ices (which a e simila o he esou ces sha ed in communi y ne wo ks) o gua an ee he communi y cloud en i onmen emains easible while gi ing he owne s a space o hei own usage. We eplica ed he Communi y- Lab in as uc u e o deploy di e en se ices in he de ices a ailable in o de o unde s and he pe o mance issues ha ou p oposed app oach can ha e. F om he e alua ion pe o med, we showed ha he impac on he pe o mance o unning concu en se ices, while in di e en use con ex s is minimum compa ed o he cu en s a e o he de ices, deployed h oughou he communi y ne wo k. In ac , in he p oposed app oached he e alua ed se ices pe o m much he same way as in he cu en Communi y-Lab app oach, sugges ing ha he p oposed ap- p oach is easible o deploy on such in as uc u es. Mo eo e he p oposed app oach can augmen he es ab- lished ne wo k in as uc u e o communi y ne wo ks and he abili y o use s o ha e access o mo e compu ing powe , explo ing he communi y cloud en i onmen s and achie ing a mul i-pu pose sys em. This way we can add alue o he communi y ne wo ks, wi hou deploying mo e physical de- ices, and lowe ing he impac on he se ices when unning concu en ly wi hin he communi y ne wo k in as uc u e. As a nex s ep, we conside o deploy and moni o esou ces using he p oposed app oach in he Communi y-Lab es bed, such ha we can iden i y con igu a ion o secu i y issues ha may hinde he pe o mance o he se ices. Also, we will conside di e en ypes o se ices unning concu en ly o speci ically imp o e and op imize hei pe o mance applying he p oposed app oach. ACKNOWLEDGEMENT This wo k was suppo ed by he Eu opean F amewo k P og amme 7 FIRE Ini ia i e p ojec s CONFINE (FP7- 288535), CLOMMUNITY (FP7- 317879), Uni e si a Poli cnica de Ca alunya-Ba celonaTECH and by he Spanish go e nmen unde con ac TIN2013-47245-C2-1-R. REFERENCES [1] B. B aem e al., “A case o esea ch wi h and on communi y ne wo ks,” SIGCOMM Compu . Commun. Re ., ol. 43, no. 3, pp. 68–73, Jul. 2013. [2] Z. Wilcox-O’Hea n and B. Wa ne , “Tahoe: The leas -au ho i y ilesys- em,” in P oceedings o he 4 h ACM In . Wo kshop on S o age Secu i y and Su i abili y, se . S o ageSS ’08. New Yo k, NY, USA: ACM, 2008, pp. 21–26. [3] M. Selimi and F. F ei ag, “Tahoe-LAFS Dis ibu ed S o age Se ice in Communi y Ne wo k Clouds,” in 4 h IEEE In . Con e ence on Big Da a and Cloud Compu ing 2014. Sydney, Aus alia: IEEE, Dec. 2014. [4] L. Baldesi, L. Macca i, and R. Lo Cigno, “Imp o ing P2P S eam- ing in Communi y-Lab Th ough Local S a egies,” in 10 h IEEE In . Con e ence on Wi eless and Mobile Compu ing, Ne wo king and Com- munica ions, La naca, Cyp us, Oc obe 2014, pp. 33–39. [5] R. Bi ke e al., “A chi ec u e o a ne wo k-awa e p2p- applica ion: The napa-wine app oach,” IEEE Communica ions Magazine, ol. 49, pp. 154–163, 06/2011 2011. [6] M. Selimi e al., “Cloud-based ex ension o communi y-lab,” in 22nd In . Symposium on Modelling, Analysis Simula ion o Compu e and Telecommunica ion Sys ems, IEEE, Sep 2014, pp. 502–505. [7] M. Selimi, F. F ei ag, R. Cen elles, and A. Moll, “Dis ibu ed S o - age and Se ice Disco e y o He e ogeneous Communi y Ne wo k Clouds,” in 7 h IEEE/ACM In . Con e ence on U ili y and Cloud Compu ing (UCC’14). London, UK: IEEE/ACM, Dec. 2014. [8] W. D. No co and D. Capps, “Iozone ilesys em benchma k,” URL: www. iozone. o g, ol. 55, 2003. [9] J. Jimenez e al., “Suppo ing Cloud Deploymen in he Gui i.ne Communi y Ne wo k,” in 5 h Global In o ma ion In as uc u e and Ne wo king Symposium (GIIS 2013), T en o, I aly, Oc . 2013. [10] F. Bonomi e al., “Fog compu ing: A pla o m o in e ne o hings and analy ics,” in Big Da a and In e ne o Things: A Roadmap o Sma En i onmen s, se . S udies in Compu a ional In elligence, N. Bessis and C. Dob e, Eds. Sp inge In . Publishing, 2014, ol. 546, pp. 169–186. [11] M. Jang e al., “Pe sonal clouds: Sha ing and in eg a ing ne wo ked esou ces o enhance end use expe iences,” in INFOCOM, 2014 P oceedings IEEE, Ap il 2014, pp. 2220–2228. [12] B. Sao apakhi an and M. De e sikio is, “Enhancing Compu ing Powe by Exploi ing Unde u ilized Resou ces in he Communi y Cloud,” in IEEE In . Con e ence on Communica ions (ICC 2011), Kyo o, Japan, Jun. 2011. [13] K. Kuma e al., “A su ey o compu a ion o loading o mobile sys ems,” Mob. Ne w. Appl., ol. 18, no. 1, pp. 129–140, Feb. 2013. [14] D. F. Willis, A. Dasgup a, and S. Bane jee, “Pa ad op: A mul i- enan pla o m o dynamically ins alled hi d pa y se ices on home ga eways,” in P oceedings o he 2014 ACM SIGCOMM Wo kshop on Dis ibu ed Cloud Compu ing, se . DCC ’14. New Yo k, NY, USA: ACM, 2014, pp. 43–44. [15] S. Sol esz e al., “Con aine -based Ope a ing Sys em Vi ualiza ion: A Scalable, High-pe o mance Al e na i e o Hype iso s,” in Eu oSys’07, he Eu opean Chap e o SIGOPS. Po ugal: ACM SIGOPS, 2007.