Measuring Mobile Starlink Performance: A Comprehensive Look

Recei ed 21 Janua y 2025; accep ed 5 Feb ua y 2025. Da e o publica ion 7 Feb ua y 2025; da e o cu en e sion 24 Feb ua y 2025. Digi al Objec Iden i ie 10.1109/OJCOMS.2025.3539836 Measu ing Mobile S a link Pe o mance: A Comp ehensi e Look DOMINIC LANIEWSKI (G adua e S uden Membe , IEEE), ERIC LANFER , (G adua e S uden Membe , IEEE), AND NILS ASCHENBRUCK , (Membe , IEEE) Ins i u e o Compu e Science, Osnab ück Uni e si y, 49069 Osnab ück, Ge many CORRESPONDING AUTHOR: D. LANIEWSKI (e-mail: [email protected]) This wo k was suppo ed in pa by he Ge man Fede al Minis y o Digi al and T anspo as pa o he “Inno a i e Ne wo k Technologies” Funding P og am unde G an FKZ: 19OI23008C. ABSTRACT Wi h he ecen success o Low Ea h O bi (LEO) sa elli e ne wo ks, such as SpaceX’s S a link, measu ing hei pe o mance has been o g ea in e es o he communi y. While s a iona y S a link pe o mance has been ex ensi ely assessed on a globalized iew, mobile - in-mo ion - usage is ela i ely new. A i s look a i s pe o mance has only been aken by h ee s udies so a . In his pape , we ake a comp ehensi e look a mobile S a link pe o mance om he measu emen pe spec i e by answe ing he ollowing h ee esea ch ques ions: (1) How does mobile S a link pe o mance di e in di e en egions o he wo ld? (2) How does he mobile pe o mance compa e o s a iona y pe o mance? (3) How does obs uc ion impac mobile pe o mance? To answe hese ques ions, we conduc ou own 300 km long es -d i e on he Ge man Au obahn (highway) wi h ca eloci ies up o 140 km/h. We compa e ou esul s o he da ase s o he o he h ee s udies and o s a iona y measu emen s. To he bes o ou knowledge, we a e he i s o deepe analyze he impac o obs uc ions on he pe o mance. Fo his, we map b idges c ossing he highway o ou measu emen s and ind ha hese sho o al obs uc ions cause signi ican bu s packe loss, RTT spikes, and h oughpu d ops. We show ha mobili y-induced ins abili ies can ha e a se e e nega i e impac on he pe o mance o highe -le el applica ions such as HTTP bulk ile ans e . INDEX TERMS Da ase , la high pe o mance, LEO, LEO measu emen , measu emen , mobili y, sa elli e communica ion, S a link, S a link da ase , S a link measu emen , S a link mobili y. I. INTRODUCTION LOW EARTH O bi (LEO) sa elli e ne wo ks a e an eme ging echnology ha p omises o p o ide high- speed b oadband In e ne o emo e and unde se ed a eas wo ld-wide. The e a e mul iple p ojec s o build such ne wo ks. While some p ojec s, such as Amazon’s Kuipe and Telesa ’s Ligh speed, a e in ea ly de elopmen phases and scheduled o be ope a ional in 2025 [16] and 2027 [36], espec i ely, o he p ojec s, such as SpaceX’s S a link, I idium and Eu elsa ’s OneWeb a e al eady a ailable o he public. Wi h 6,218 ope a ional sa elli es as o Janua y 2025 [25],S a link is he la ges LEO-ISP o-da e. S a link was ini ially designed o s a iona y use. Use s connec o he sa elli e ia a special e minal - called he “dish” - ha is ypically moun ed on oo ops o buildings wi h clea iew o he sky. In la e 2022, he Fla High Pe o mance (FHP) dish was in oduced, designed o be moun ed on ehicles such as ca s and boa s and o be ope a ed while in-mo ion. O e he yea s, he esea ch communi y has assessed s a iona y S a link Pe o mance in many measu emen s udies in as ly di e en measu emen scena ios and loca ions om all o e he wo ld [14],[21], [24],[26]. In con as , only h ee measu emen s udies moun ed he FHP dish on a ca and assessed in-mo ion S a link pe o mance in he U.S. [10], in he A c ic egion o no he n Sweden [4], and in a ci y in Wes e n Ge many [20]. In his pape , we aim o ake a comp ehensi e look a mobile S a link pe o mance om he measu emen pe spec i e by answe ing he ollowing esea ch ques ions (RQ): 1) How does mobile S a link pe o mance di e in di e en egions o he wo ld? c 2025 The Au ho s. This wo k is licensed unde a C ea i e Commons A ibu ion 4.0 License. Fo mo e in o ma ion, see h ps://c ea i ecommons.o g/licenses/by/4.0/ 1266 VOLUME 6, 2025 2) How does he mobile pe o mance compa e o s a ion- a y pe o mance? 3) How does obs uc ion impac mobile pe o mance? To answe hese ques ions, we conduc ou own mobile S a link measu emen campaign in an app oxima ely 300 km long es d i e on he Ge man Au obahn (highway) in he no h-wes o Ge many close o he Du ch bo de . We compa e ou esul s o he publicly a ailable da ase s om he U.S. [10], he A c ic [4], and a Ge man ci y [20] o answe RQ1. To answe RQ2, we conduc a second, s a iona y measu emen campaign and compa e he esul s o ou highway measu emen s. The campaign spans one week and consis s o wo s a iona y “S anda d Gen 2” dishes unning di e en se ice plans and an FHP dish placed di ec ly nex o each o he on he oo op o a building wi h an unobs uc ed iew o he sky. In his con ex , we also analyze he impac o di e en S a link dish e sions and se ice plans on he pe o mance. Addi ionally, we map b idges c ossing he highway o ou measu emen esul s o answe RQ3. We hen build an emula ion o ou es d i e ha accu a ely ep oduces he expe ienced link condi ions and use i o analyze he impac o mobili y- induced ins abili ies on eal-wo ld applica ions. Bo h o ou measu emen campaigns con ain all ele an ne wo k pa ame e s such as downlink and uplink h oughpu , he RTT, and packe loss. We elease ou da ase s and ou emula ion as open access da a o he communi y [19]. The es o his pape is s uc u ed as ollows. Fi s , we in oduce ele an backg ound knowledge abou S a link in Sec ion II and discuss ela ed wo k in Sec ion III. Then, we desc ibe ou mobile measu emen campaign in Sec ion IV. A e wa d, we answe he esea ch ques ions in Sec ions V–VII, espec i ely. We hen discuss he impac o he ound S a link ins abili ies unde mobili y on eal-wo ld applica ions in Sec ion VIII. Finally, we discuss limi a ions o ou wo k in Sec ion IX and p esen in e es ing u u e esea ch di ec ions in Sec ion Xbe o e we conclude ou pape in Sec ion XI. II. BACKGROUND S a link opened o he public in 2020, ini ially designed only o s a iona y use. Use s moun a sa elli e an enna ( he “dish”) o places wi h an unobs uc ed iew o he sky, such as oo ops o esiden ial buildings. The an enna connec s o a sa elli e ha is in i s ield o iew (FOV). The sa elli e also connec s o a g ound s a ion (GS). This connec ion (dish - sa elli e - GS) is called he one-hop ben pipe. I is he p edominan a chi ec u e in mos egions o he wo ld. I no GS is in he FOV o he sa elli e, sa elli es o e sion 1.5 and newe can connec o o he sa elli es o ming in e -sa elli e links (ISL) using lase communica ion. This a chi ec u e is called a mul i-hop ben pipe. I is p ima ily used o connec mo e emo e egions o he wo ld. The GS is connec ed o he e es ial ne wo k ia a hie a chical poin o p esence (POP) s uc u e. As o Janua y 2025, S a link ope a es 6,218 sa elli es a o bi s o ou inclina ions: 43◦,53 ◦,70 ◦, and 97.6◦. These o bi s a y in sa elli e densi ies. 2,137 sa elli es a e ope a ional on he 43◦o bi , 3,468 on he 53◦o bi , 384 on he 70◦o bi , and 229 on he 97.6◦o bi [25]. While he 43◦ and 53◦o bi s p o ide se ice o mos egions o he wo ld, he 70◦and 97.6◦o bi s a e p ima ily used o connec he poles. SpaceX o e s di e en dishes o s a iona y and mobile use. Th ee gene a ions (Gen 1 - Gen 3) o he S anda d dish o s a iona y use ha e been eleased. The Gen 2 dish can align i sel o he op imal angle o ack sa elli es due o a buil -in mo o , i has an a e age powe consump ion o 50 - 75 W, and a FOV o 100◦[33]. The Gen 3 dish doesn’ ha e a mo o , i has a highe a e age powe consump ion o 75 - 100 W, and a la ge FOV o 110◦[34]. In con as , he FHP dish is designed o mobile use-cases. I can be moun ed, e.g., on a ca o a boa . I doesn’ ha e a mo o , i has an a e age powe consump ion o 110 - 150 W, and a FOV o 140◦[31]. A se ice plan needs o be booked o ope a e a dish. Cu en ly (as o Janua y 2025), S a link o e s h ee ca e- go ies: Residen ial,Roam, and Mobile [32].TheResiden ial se ice plan is bound o a ixed loca ion. I is ypically used wi h a S anda d dish o p o ide consume -le el In e ne a home. The gene a ed a ic has a medium p io i iza ion in he S a link ne wo k. The Roam se ice plan is p ima ily designed o s a iona y use-cases wi h changing loca ions, such as camping. Howe e , i also suppo s in-mo ion usage. I s da a has he lowes p io i y o all se ice plans. The Mobile se ice plan is designed o in-mo ion use, e.g., on ca s and boa s. I includes 50 GB o 1 TB mon hly p io i y a ic. This a ic enjoys he highes p io i iza ion o all se ice plans. Once he included da a olume is used up, an unlimi ed amoun o Residen ial a ic can be used. While all se ice plans can be booked on he S anda d dishes, he FHP dish equi es he Mobile plan. III. RELATED WORK S a iona y S a link Pe o mance has been ex ensi ely assessed ia eal-wo ld measu emen s in ecen s udies [9], [11],[14],[21],[24],[26],[27]. These ange om small- scale s udies consis ing o a single S a link dish [9],[26] o la ge-scale measu emen campaigns spanning mul iple mon hs and consis ing o mul iple dishes dis ibu ed o e di e en coun ies [11],[14],[21],[24],[27]. In con as , measu emen s udies o mobile S a link pe o mance a e limi ed. In a i s a emp , López e al. [23] moun ed a s anda d Gen-1 e minal, designed o s a iona y esiden ial use, on a ca . They measu ed la ency in a 250 km long es d i e in no he n Denma k wi h eloci ies up o 15 km/h. Ma e al. [24] also moun ed a s anda d Gen- 1 e minal on a an and measu ed uplink and downlink pe o mance in a 30 minu e es d i e in sou h-wes e n Canada a eloci ies o 40-70 km/h. In o al, h ee s udies assessed mobile S a link Pe o mance using an FHP dish VOLUME 6, 2025 1267 LANIEWSKI e al.: MEASURING MOBILE STARLINK PERFORMANCE: A COMPREHENSIVE LOOK FIGURE 1. The FHP dish moun ed on a VW an. moun ed on a ca [4],[10],[20]. All o hem made hei da ase s publicly a ailable. A de ailed compa ison is gi en in Sec ion Vand especially in Tab. 1.Hue al.[10] moun a s a iona y Gen-2 dish and an FHP dish on a ca and conduc a 1,800 km es d i e ac oss he U.S. wi h eloci ies up o 100 km/h and measu e TCP and UDP h oughpu , la ency, and packe loss. They also compa e he pe o mance o he Roam and Mobile se ice plans. Beckman e al. [4] measu e TCP downlink h oughpu and la ency in a 970 km es d i e ac oss he a c ic ci cle in no he n Sweden. Thei eloci y was ypically in he ange o 80-100 km/h. Laniewski e al. [20] build an au onomous measu emen se up ha au oma ically s a s/ends he measu emen s when he ca is u ned on/o . They deploy i on a se ice ca o he local ene gy p o ide in he ci y o Osnab ück, Ge many, o e a span o wo mon hs and measu e UDP h oughpu , la ency, and packe loss. The measu emen s a e mainly conduc ed in he ci y a eloci ies a ound 50 km/h, bu also con ain a sho highway measu emen a 120 km/h. All he men ioned mobile measu emen s ud- ies [4],[10],[20] only ake a i s look a mobile S a link pe o mance by analyzing hei own measu emen s. A globalized iew, which is he cu en end in he s a iona y case, is lacking. Fu he mo e, a de ailed compa ison be ween mobile and s a iona y pe o mance is lacking. A i s look is only p o ided by Laniewski e al. [20]. Las ly, he impac o obs uc ion on S a link pe o mance has no ye been analyzed in nei he he s a iona y no he mobile cases. We aim o ill hese gaps by answe ing ou esea ch ques ions in his pape . IV. THE HIGHWAY MEASUREMENT CAMPAIGN In his sec ion, we desc ibe ou highway measu emen campaign. A. THE PHYSICAL MEASUREMENT SETUP We moun he FHP dish on he oo o a Volkswagen (VW) an (Fig. 1). I is powe ed by a Jacke y Explo e 2000 P o powe s a ion wi h a capaci y o 2 kWh. The FIGURE 2. Ou es d i e ou e om Osnab ück, Ge many, along he Du ch bo de up o Bunde, close o he Ge man No h Sea. The ou e is d i en bo h ways. ne wo k measu emen s ( h oughpu , la ency, and packe loss) a e conduc ed on a lap op (AMD Ryzen 7 PRO 4750 U, 16 GB RAM) ha is connec ed ia Gigabi E he ne o he S a link ou e , which is se in o bypass mode.TheMobile se ice plan wi h 50 GB p io i y a ic is booked o he dish. Ou measu emen s we e conduc ed using only non-p io i y (s anda d) a ic. B. THE MEASUREMENT CAMPAIGN We conduc a 300 km es d i e s a ing in he ci y cen e o Osnab ück, Ge many, o e he A 30 highway and hen u he on he A 31 highway close o he Du ch bo de o Bunde. A e wa d, we d i e he same way back. The selec ed ou e has mul iple unique ea u es, making i an ideal ou e o answe ing ou esea ch ques ions. Fi s , he e a e no speed limi a ions, allowing measu emen s a high ca eloci ies. Second, he gene ally spa se a ic allows keeping high eloci ies o ex ended ime pe iods. Thi d, he e ain ensu es an op imal, unobs uc ed iew o he sky o he S a link dish, as i is la and he e a e no la ge objec s in di ec p oximi y o he oad possibly causing obs uc ion, such as o es s o la ge buildings. Obs uc ion is only caused by b idges c ossing he highway, allowing an isola ed analysis o hei impac on he measu ed pe o mance. Fou h, he a ea a ound he ou e is spa sely popula ed, minimizing a ia ions in S a link pe o mance possibly caused by o he use s p oducing load- luc ua ions wi hin he same se ice cells o he sa elli es. Las ly, he ou e is in be ween he ci ies o Osnab ück and Enschede, gua - an eeing spa ial p oximi y and consequen ly allowing g ea compa abili y o ela ed S a link measu emen s conduc ed om hose ci ies [20],[21]. 1268 VOLUME 6, 2025 TABLE 1. Compa ison o he di e en mobile S a link measu emen campaigns. We d o e he ca a eloci ies o 80, 100, 120, and 140 km/h o app oxima ely 10 minu es s aigh . We aimed o d i e app oxima ely he same eloci ies a he same highway sec ions on bo h ways. As he an didn’ ha e c uise con ol and as we had o conside he a ic si ua ion, he eloci ies wi hin hose sec ions a ied sligh ly. The ou e and ca eloci ies a e isualized in Fig. 2. The es d i e ook place on he e ening o he 19 h o Ap il 2024, app oxima ely om 18 h o 23 h. This a oids he e ening ush hou on he highway. Ea lie s udies [18],[21],[24] ound ha wea he condi- ions can signi ican ly impac S a link pe o mance. Hea y ain wi h p ecipi a ion le els >4mm/h and cloudiness can educe he UDP downlink h oughpu by up o 30-45 % and 5 %, espec i ely [18],[21],[24]. O he wea he condi ions, such as wind speed, empe a u e, humidi y, and sola adia ion, we e ound o ha e no impac [21]. The uplink h oughpu , RTT, and packe loss a e gene ally una ec ed by he wea he [21]. We use h ee wea he s a ions o he Ge man Me eo ological Se ice (DWD) close o ou d i ing ou e o moni o he wea he : Osnab ück-Belm (s a ion ID: 342), Lingen (s a ion ID: 15813), and Dö pen (s a ion ID: 6159). The app oxima e loca ions a e also ma ked in Fig. 2. O e all, i was a ainy and cloudy day wi h o al p ecipi a ions be ween 0 h and 18 h o 17.2 mm, 12.9 mm, and 10.6 mm in Osnab ück-Belm, Lingen, and Dö pen, espec i ely. Du ing he es d i e, he e was 0.37 mm p ecipi a ion be ween 18:10 h and 18:40 h in Osnab ück- Belm, 0.09 mm be ween 18:00 h and 18:50 h in Lingen, and no p ecipi a ion in Dö pen. E en hough hese a e low p ecipi a ion le els du ing he es d i e, he hea ie p ecipi a ion ea lie ha day led o a we oad causing whi led wa e by p eceding ca s, po en ially impac ing he S a link pe o mance. C. MEASUREMENT DETAILS On he way o Bunde, we con inuously measu ed UDP h oughpu using ipe 3 (Ve sion 3.9) agains a se e loca ed in he ne wo k ope a ing cen e o he local uni e si y. The se e is connec ed o he 5 Gbi /s ibe uni e si y ne wo k. We measu ed downlink and uplink h oughpu in pa allel using wo sepa a e ipe ins ances o a oid CPU limi a ions o ipe ’s bidi ec ional mode [21]. Ta ge da a a es o 500 Mbi /s o he downlink and 100 Mbi /s o he uplink we e used o sa u a e he links. Since ipe was unning cons an ly, he da a was collec ed a one-second g anula i y. On he way back o Osnab ück, we measu ed he RTT and packe loss using he ping u ili y wi h he S a link ga eway as des ina ion and in e -packe spacings o 10 ms. P ecisely, we used he ollowing command: ping − D−i0.01 100.64.0.1. Du ing he comple e es d i e, we measu ed he GPS loca ions and ehicle eloci ies using he app GPS T acks (Ve sion 4.4.9) unning on an iPhone 13 wi h iOS 17.4.1. V. MOBILE PERFORMANCE IN DIFFERENT WORLD REGIONS In his sec ion, we answe RQ1: How does mobile S a link pe o mance di e in di e en egions o he wo ld? We do his by compa ing he esul s o ou highway es d i e o exis ing da ase s om he U.S. [10], he A c ic [4], and S a link on he oad om Osnab ück ci y, Ge many [20].Fo his, we i s gi e an o e iew o he measu emen campaigns and explain ele an di e ences. A e wa d, we analyze he h oughpu , packe loss, and he impac o he ehicle speed on he pe o mance. Las ly, we summa ize ou indings. A. MEASUREMENT CAMPAIGN COMPARISON Tab. 1summa izes he de ails o he di e en measu emen campaigns. Fo simplici y and be e compa abili y, we e e o ou measu emen campaign as GER-Highway, o he campaign o Laniewski e al. [20] as GER-Ci y, o he campaign o Hu e al. [10] as U.S., and o he campaign o Beckman e al. [4] as A c ic. Ou GER-Highway campaign is highligh ed. We see ha he campaigns di e as ly. They we e conduc ed o e a ime span o one yea in di e en egions VOLUME 6, 2025 1269 LANIEWSKI e al.: MEASURING MOBILE STARLINK PERFORMANCE: A COMPREHENSIVE LOOK FIGURE 3. Th oughpu dis ibu ions o he di e en da ase s. “n” deno es he numbe o samples pe da ase . FIGURE 4. Bu s leng hs. o he wo ld, in di e en e ains, and unde di e en wea he condi ions. Mo eo e , he measu emen se ups and measu emen p ocesses we e signi ican ly di e en . Fo example, he measu emen de ice was connec ed o he dish ia di e en echnologies (E he ne s. WLAN), he la ency measu emen s we e conduc ed agains di e en a ge s (ben - pipe s. ull-pa h la ency) and a di e en g anula i ies, and di e en anspo p o ocols we e used o he h oughpu measu emen s (TCP s. UDP). All hese ac o s can possibly impac he measu emen esul s, limi ing di ec compa abili y o he da ase s. Because o he as di e ences in he la ency measu emen s, we ocus ou analysis on he h oughpu and packe loss. B. THROUGHPUT The dis ibu ion o he downlink h oughpu is isualized in Fig. 3(a) as iolin plo s and box plo s. The GER-Ci y da ase has he highes median h oughpu , ollowed GER-Highway, he A c ic, and he U.S. wi h medians o 263, 218, 186, and 134 Mbi /s, espec i ely. Thei 95 % con idence in e als (CIs) do no o e lap, indica ing s a is ical signi icance. Thei s anda d de ia ions (s d) a e compa able wi h 94, 104, 77, and 90 Mbi /s, espec i ely. The iolin plo s e eal ha he U.S. da ase has a la ge clus e o samples a 0 Mbi /s, accoun ing o app oxima ely 15 % o he o al samples. This beha io is unique and can ha e di e en causes such as ins abili ies in he measu emen se up, bad sa elli e co e age, o equen obs uc ion. The emaining samples o he U.S. da ase a e nea ly e enly dis ibu ed be ween app oxima ely 50 Mbi /s and 250 Mbi /s wi h a small clus e a ound 160 Mbi /s. The GER-Ci y and A c ic da ase s show a compa able sample dis ibu- ion. Bo h show a small clus e a ound he hi d qua ile, which is app oxima ely loca ed a 310 Mbi /s and 220 Mbi s, espec i ely. In con as , he dis ibu ion o he samples o ou GER-Highway da ase app oxima ely ollows a no mal dis ibu ion be ween 450 Mbi /s and 0 Mbi /s. The uplink h oughpu dis ibu ion is depic ed by Fig. 3(b).1The median h oughpu is 14.9 Mbi /s, 15.4 Mbi /s, and 18 Mbi /s wi h s ds o 10.2 Mbi /s, 8.6 Mbi /s, and 14.4 Mbi /s o ou GER-Highway da ase , he GER-ci y da ase , and he U.S. da ase , espec i ely. The 95 % CIs o he medians o e lap o he GER-Highway and GER-Ci y da ase s. While his indica es a simila uplink pe o mance in ci ies and on he highway in Cen al Eu ope, i also shows a be e pe o mance in he U.S. The iolin plo s e eal he eason o his obse a ion. The samples o ou GER-Highway da ase , as well as he GER-ci y da ase , ollow app oxima ely a no mal dis ibu ion be ween 0 Mbi /s and 40 Mbi /s wi h clus e s a ound 15-18 Mbi /s. In con as , he samples o he U.S. da ase a e clus e ed a ound 9 Mbi /s, 28 Mbi /s, and 48 Mbi /s. The da ase does no con ain samples in he ange o 35-40 Mbi /s and a ound 25 Mbi /s. In e p e a ion: The obse ed beha io s ha e di e en oo causes. The A c ic is se ed only by he 97.6◦o bi [4], which is popula ed by ewe sa elli es. This, in combina ion wi h he di e en measu emen se up using mul iple pa allel TCP connec ions o sa u a e he link, likely leads o he wo s downlink h oughpu compa ed o he measu emen s om GER-Ci y and ou GER-Highway es d i e. The U.S. and Eu ope a e bo h se ed by he 53◦,70 ◦, and 97.6◦ o bi s, esul ing in a highe sa elli e densi y. The signi ican ly wo se downlink h oughpu in he U.S. can be caused by a highe load caused by mo e cus ome s o by wo se co e age o g ound s a ions. The di e ences in he downlink h oughpu be ween he GER-Ci y da ase and ou GER- Highway measu emen s a e s iking, since bo h use a simila measu emen se up and simila measu emen ools. This is likely caused by he ainy wea he condi ions. In p e ious, s a iona y measu emen s [21], i was ound ha ain causes 1The A c ic da ase is no isualized, because i does no con ain uplink h oughpu measu emen s, as indica ed by Tab. 1. 1270 VOLUME 6, 2025 FIGURE 5. Th oughpu ca ego ized in o speed bucke s. Sample sizes pe speed bucke ( om le o igh , nsamples in b aces): Downlink: <70: [722;2920;18028;12733], 70 −90: [754;103;4544;13148], 90 −100: [1309;46;5441;16909], 110 −130: [1535;30;1280;136], ≥130: [541]; Uplink: <70: [722;2920;4623], 70 −90: [754;103;333], 90 −100: [1309;46;268], 110 −130: [1535;30], ≥130: [541]. he UDP downlink h oughpu o d op up o 50 %, depending on he p ecipi a ion le el, bu ha he uplink h oughpu emains una ec ed. Ou esul s show exac ly his beha io , wi h he median downlink h oughpu o ou GER-Highway measu emen s being 17 % lowe han he one o GER-Ci y. Fo he uplink h oughpu , he clus e ed dis ibu ion o he samples in he U.S. da ase is likely caused by measu emen se up-speci ic elemen s, leading o no aluable compa abili y o ou GER-Highway measu emen s and he GER-Ci y da ase . C. PACKET LOSS The packe loss beha io o he di e en da ase s can be ha dly compa ed because o di e en measu emen me hod- ologies. In ou GER-Highway campaign, i is measu ed using ICMP on he ben -pipe and on he packe -le el based on sequence numbe s. The GER-Ci y da ase measu es he ull-pa h loss using ICMP wi h a epo ing in e al as pe cen ages o e 250 packe s. The U.S. da ase measu es he ull-pa h loss using TCP e ansmissions and epo s i using pe cen ages o e measu emen in e als o a ying leng hs. Because o hese di e ences, we only p esen ou measu emen esul s. O e all, ou measu emen esul s show a packe loss a e (PLR) o 2.84 % and a e age bu s leng h o 11 packe s. The dis ibu ion o he bu s leng hs is isualized by Fig. 4.We de ine a bu s as he numbe o consecu i ely los packe s. App oxima ely 50 % o he packe loss e en s a e isola ed losses wi h a bu s leng h o 1. A u he app oxima ely 10 % possess a leng h o 2 and app oxima ely 35 % leng hs be ween 3 and 40. In addi ion o he shown beha io , we obse e bu s s wi h leng hs a ound 500, 1,100 and 2,000 packe s, bu omi hem in ou isualiza ion o be e isibili y. In e p e a ion: Wi h a PLR o app oxima ely 3 %, ou measu emen shows signi ican ly mo e packe loss han epo ed by he U.S. da ase (up o 1.3 % e ansmissions) and by he GER-Ci y da ase (1 % wi h spikes up o 10 % in ci y-a eas wi h la ge buildings). La e , in Sec ions VI-C and VII, we show ha he packe loss, especially he bu s s, is pa ially caused by pe iodic S a link econ igu a ions and obs uc ion. Toge he , hese accoun o app oxima ely 32 % o he o e all packe loss. Fu he analysis is needed as pa o u u e wo k o de e mine he cause o he emaining 68 % o he packe loss. D. THE IMPACT OF VEHICLE VELOCITY The impac o he ehicle eloci y on he h oughpu is isualized in Fig. 5.2Conside ing he downlink h oughpu , we see a decline om a median o 234 Mbi /s o eloci ies o 70-90 km/h o 180 Mbi /s o eloci ies ≥130 km/h. This ep esen s a 23.1% dec ease. This beha io is con a y o he ones in he U.S. and he a c ic, which a e la gely cons an o he di e en eloci ies, as well as he one om GER- Ci y, which sees a no able inc ease only o high eloci ies o 110-130 km/h. The uplink h oughpu in Fig. 5(b) shows a di e en beha io . The U.S. da a sees an inc ease om a median o 15.4 Mbi /s a <70 km/h eloci ies o 27.1 Mbi /s o eloci ies be ween 90 and 110 km/h, ep esen ing a 76 % inc ease. The GER-Ci y da ase shows an inc ease o highe ca eloci ies be ween 110 and 130 km/h. In con as , ou highway d i e has median uplink h oughpu s o 13.2, 16.1, 16.3, 12.9, and 13.1 Mbi /s o inc easing ehicle eloci ies. We no e ha ou measu emen s a eloci ies <70 km/h a e no ep esen a i e, as hey we e conduc ed in cons uc ion a eas on he highway. A hose places, he highway is na ow wi h a single lane and ypically he e a e mo e objec s causing obs uc ions. In e p e a ion: Based on he di e ence o 3 Mbi /s be ween he as es and slowes h oughpu , we a gue ha he uplink h oughpu can be seen as independen o he ehicle eloci ies and he changes a e likely noise caused, e.g., by di e ing sample sizes. The au ho s o he U.S. da ase [10] a gue ha he ehicle eloci y is negligible compa ed o he sa elli e eloci ies o app oxima ely 28,000 km/h. While hei downlink h oughpu , as well as he A c ic downlink h oughpu , a e in line wi h his a gumen a ion, hei uplink h oughpu inc eases signi ican ly wi h inc easing eloci ies. Fu he mo e, he au ho s o he GER-ci y da ase [20] no e ha a decline o downlink h oughpu can be obse ed 2We no e ha no all da ase s con ain samples a highe ca eloci ies, as indica ed by Tab. 1. This leads o missing plo s in he a ec ed eloci y bucke s. VOLUME 6, 2025 1271 LANIEWSKI e al.: MEASURING MOBILE STARLINK PERFORMANCE: A COMPREHENSIVE LOOK be ween a s anding ehicle and a d i ing ehicle, men ioning ha he ehicle speed while d i ing has no signi ican impac . Thei obse ed inc ease o high eloci ies in bo h, downlink and uplink h oughpu , is likely caused by less obs uc ion because he da a is om one sho highway d i e whe e he S a link dish had an unobs uc ed iew o he sky. In addi ion, he speed bucke o 110-130 km/h only con ains 30 samples, leading o non- ep esen a i e esul s. In con as , ou da a shows dec easing pe o mance o he downlink h oughpu and cons an pe o mance o uplink h oughpu . To u he analyze possible easons o he han he ehicle eloci y o he obse ed deg ada ion o he downlink h oughpu , we isualized he h oughpu measu emen s on a map and analyzed he su ounding e ain using sa elli e pic u es and Google S ee View. The e ain is la on he comple e ou e o ou measu emen campaign, wi h no moun ains o e en hills in p oximi y, and h oughpu d ops do no co ela e wi h ees o o he objec s nex o he highway. Fu he mo e, he e is no co ela ion o he numbe o b idges c ossing he highway a he di e en speed bucke s and he e was no p ecipi a ion du ing he measu emen segmen s wi h ca eloci ies (>110 km/h). Consequen ly, we belie e ha he ca eloci y is likely causing he obse ed beha io . Howe e , he eason o he con a y beha io obse ed by he di e en measu emen s udies emains unclea and subjec o u u e wo k. E. TAKEAWAY Based on he p esen ed esul s, RQ1 (How does mobile S a link pe o mance di e in di e en egions o he wo ld?) can be answe ed he ollowing way. In e ms o he downlink h oughpu , he bes pe o mance is achie ed in Cen al Eu ope, ollowed by he A c ic and he U.S. No conclusion can be d awn abou he uplink h oughpu because i was no measu ed in he A c ic and he da a o he U.S. is challenging o in e p e , as he obse ed beha io may be caused by speci ics o he measu emen campaign. Simila ly, no conclusion can be d awn abou he la ency because o as ly di e en measu emen se ups. VI. MOBILE VS. STATIONARY PERFORMANCE In his sec ion, we answe RQ2: How does he mobile pe o mance compa e o s a iona y pe o mance? Fo his, we conduc a sepa a e s a iona y measu emen campaign, o which we compa e ou GER-Highway measu emen esul s. Fi s , we desc ibe ou measu emen campaign. Nex , we compa e he esul s o ou GER-Highway es d i e. In his con ex , we also analyze whe he ha dwa e di e ences in he used dish e sions and he used se ice plans cause pe o mance di e ences. Then, we analyze he impac o S a link’s 15 second econ igu a ion in e al on he mobile and s a iona y pe o mance. Las ly, we summa ize ou indings. A. THE STATIONARY MEASUREMENT CAMPAIGN We conduc a i e-day long s a iona y e e ence measu emen om 19 h o 23 d May 2024 wi h h ee S a link dishes placed di ec ly nex o each o he on he oo op o a uni e si y building wi h unobs uc ed iews o he sky. I consis s o h ee S a link dishes wi h di e en se ice plans: (1) A S anda d Gen-2 dish wi h he Residen ial se ice plan, which we will i le as S anda d in his sec ion, (2) A S anda d Gen-2 dish wi h he Roam se ice plan, which we call Roam, and (3) An FHP dish wi h he Mobile se ice plan, whichwecallFHP. Since he Mobile se ice plan includes 50 GB p io i y a ic, we ensu ed ha hey we e consumed be o e s a ing ou measu emen s. Thus, all measu emen s we e conduc ed using s anda d a ic p io i iza ion. All h ee dishes a e ope a ed in bypass mode and connec ed ia 1 Gbi /s E he ne o he measu emen compu e , which is equipped wi h a ou -po PCIe E he ne adap e . This se up ensu es equal measu emen condi ions o all dishes. Ou measu emen ools and pa ame e iza ion a e iden ical o he ones we used o ou GER-Highway es d i e, leading o di ec compa abili y o bo h campaigns. We u ilize he ping u ili y o measu e he RTT and packe loss. We use he same con igu a ion as o ou GER-Highway es d i e (c . Sec ion IV-C) wi h he S a link ga eway as a ge and in e packe spacings o 0.01 s. We measu e UDP downlink and uplink h oughpu in pa allel using sepa a e ipe 3 e sion 3.9 ins ances o a oid possible CPU limi a ions o ipe ’s bidi ec ional mode. All h ee dishes measu e agains hei own ins ances unning on a se e loca ed in he ne wo k ope a ing cen e o he local uni e si y, connec ed o he 5 Gbi /s ibe uni e si y ne wo k. We se a ge da a a es o 800 Mbi /s o he downlink and 100 Mbi /s o he uplink o sa u a e he links. Using hese se up and ools, we measu e UDP down- link and uplink h oughpu , RTT, and packe loss in a sequen ial p ocess. Fi s , we measu e he RTT and packe loss o one minu e, ollowed by he pa allel downlink and uplink h oughpu measu emen o wo minu es. We lea e app oxima ely se en minu es space be ween measu emen s in o de o minimize in e e ence wi h o he S a link use s in he same se ice cell, leading o one measu emen un app oxima ely e e y 10 minu es. Ou speci ic measu emen se up allows s a ing he measu emen s exac ly a he same ime o all h ee dishes. Du ing ou measu emen s, he e was a 8 hou ain pe iod s a ing on 21 h May a 21 h wi h low p ecipi a ion o a maximum o 1.5 mm pe hou . We moni o ed he wea he wi h a F oggi DP2000 wea he s a ion placed di ec ly nex o ou measu emen se up. B. MOBILE VS. STATIONARY PERFORMANCE COMPARISON We now compa e he esul s o ou s a iona y measu emen campaign o he ones o ou GER-Highway es d i e. The compa ison is isualized by Fig. 6. 1272 VOLUME 6, 2025 FIGURE 6. Compa ison o he mobile and s a iona y S a link pe o mance. 1) THROUGHPUT The downlink h oughpu dis ibu ions a e shown in Fig. 6(a). Median h oughpu s o 292 Mbi /s, 243 Mbi /s, 307 Mbi s, and 218 Mbi /s can be obse ed o he S anda d, Roam, FHP, and GER-Highway se ups, espec i ely. Thei 95 % CIs do no o e lap. This means ha he mobile pe o mance is app oxima ely 90 Mbi /s o 29 % lowe han he s a iona y (FHP) one. We can also see ha he FHP dish has he bes s a iona y pe o mance, being app oxima ely 5 % and 26 % as e han he S anda d and Roam se ups. The sample dis ibu ions o he S anda d and FHP dish a e simila wi h one clus e a ound he hi d qua ile, ano he one app oxima ely 50 Mbi /s abo e he i s one, and a long ail o lowe h oughpu s. The Roam dish has app oxima ely equally dis ibu ed samples be ween he i s and hi d qua ile, esul ing in o e all wo se median pe o mance. In con as , he samples o he GER-Highway measu emen s ollow app oxima ely a no mal dis ibu ion cen e ed a ound he median. A di e en beha io can be obse ed o he uplink h ough- pu depic ed in Fig. 6(b). The S anda d, Roam, FHP, and GER-Highway se ups ha e median h oughpu s o 21 Mbi /s, 17 Mbi /s, 18 Mbi /s, and 15 Mbi /s, espec i ely. Thei 95 % CIs do no o e lap. Consequen ly, he mobile pe o mance is 17 % lowe han he s a iona y one. The s a iona y FHP dish also ou pe o ms he Roam se up by app oxima ely 6 %, bu ge s ou pe o med by he S anda d se up by 17 %. Fo all ou se ups, he samples a e app oxima ely no mally dis ibu ed be ween he i s and hi d qua ile, wi h he S anda d and FHP dish ha ing ano he clus e o samples app oxima ely 10 Mbi /s abo e he hi d qua ile. In e p e a ion: The 29 % gap be ween mobile and s a- iona y pe o mance in he downlink h oughpu can be pa ially explained by he we wea he condi ions du ing ou GER-Highway es d i e (c . Sec ion IV-B). Since ain and clouds do no impac uplink h oughpu [21], i is likely ha he 17 % gap o he uplink h oughpu can also be oughly expec ed o he downlink h oughpu unde ideal wea he condi ions. Fu u e esea ch is needed o de e mine he oo cause o he 17 % gap. 2) LATENCY The RTT dis ibu ions a e shown in Fig. 6(c). The ou lie s a e omi ed o cla i y. The S anda d, Roam, FHP, and GER-Highway se ups possess median RTTs o 21.3 ms, 20.5 ms, 19 ms, and 27.6 ms, espec i ely. In esul , he RTT o he mobile se up is 45 % highe han he one o he s a iona y FHP se up. While he 95 % CIs do no o e lap, he pe o mance o he s a iona y se ups is oughly compa able in p ac ice, wi h di e ences o 1 ms in he median. All s a iona y dishes ha e simila ly dis ibu ed samples, wi h clus e s a ound he i s and hi d qua iles. In con as , he samples o he GER-Highway campaign a e clus e ed a ound he i s qua ile, wi h a long ail owa ds highe RTTs. In e p e a ion: In Sec ions VI-C and VII we will discuss ha he 15 second econ igu a ion in e al o S a link, and sho o al obs uc ion pe iods cause sho RTT spikes. While hese ac o s con ibu e o he signi ican ly highe RTTs o ou GER-Highway measu emen s compa ed o he s a iona y FHP se up, hei occu ing equency is oo low o make a meaning ul impac . Elimina ing hese spikes om ou da a s ill yields a median RTT o 27.5 ms, ende ing hose ac o s negligible in he global con ex . The causing ac o s need o be ound in u u e wo k. 3) PACKET LOSS Fo he packe loss, a simila beha io can be obse ed as o he La ency. The dis ibu ion o he bu s leng hs is isualized by Fig. 6(d). The S anda d, Roam, FHP, and GER-Highway se ups ha e PLRs o 0.36 %, 0.41 %, 0.35 %, and 2.85 % wi h a e age bu s leng hs o 2, 2, 2.1, and 11.3 packe s, espec i ely. The e is mo e han eigh imes mo e packe loss in he mobile scena io han in he s a iona y scena io. Fu he mo e, he packe loss is mo e han i e imes mo e bu s y. The packe loss and bu s loss beha io is la gely simila o all s a iona y dishes, wi h mo e han 60 % o he loss e en s being isola ed packe losses o jus one packe . In e p e a ion: Two ac o s con ibu e o he high packe loss o ou GER-Highway es d i e: he 15 sec- ond econ igu a ion in e al o S a link and objec s ha obs uc he communica ion beam. The econ igu a ion in e al accoun s o app oxima ely 9 %, and he obs uc ion induced packe loss o app oxima ely 23 % o he o al packe loss. A de ailed analysis o bo h ac o s ollows in Sec ions VI-C and VII. I is subjec o u u e wo k o de e mine he causes o he emaining 68 % o he packe loss. VOLUME 6, 2025 1273 LANIEWSKI e al.: MEASURING MOBILE STARLINK PERFORMANCE: A COMPREHENSIVE LOOK FIGURE 7. RTT. 4) THE IMPACT OF THE DISH VERSION AND SERVICE PLAN As a side p oduc o ou analysis, we can also de e mine he impac o he dish e sion and he used se ice plan on he s a iona y S a link pe o mance. Ou esul s show ha he FHP se up ou pe o ms he S anda d se up in he downlink h oughpu by app oxima ely 5 %, in he RTT by app oxima ely 10 %, and by he packe loss o 2 %. A he same ime, i s uplink h oughpu is 17 % wo se han he one o he S anda d se up. These esul s a e su p ising since he FHP dish has a wide FOV han he S anda d dish wi h 140◦ o 100◦(c . Sec ion II) and a signi ican ly highe powe consump ion. They indica e ha he be e ha dwa e is p ima ily needed o keep he connec ion while d i ing, likely because o equen changes o he su ounding en i onmen causing obs uc ion and o ien a ion changes. Fu he mo e, he Roam se up ails he S anda d se up by 17 %, 18 %, and 19 % in e ms o downlink h oughpu , uplink h oughpu , and PLR, espec i ely. Howe e , i ou pe o ms he S anda d se up by 4 %. S a link s a es ha a ic o he Roam se ice plan has he lowes p io i iza ion. Ou esul s indica e ha his does no only ha e a nega i e impac on he pe o mance du ing imes o ne wo k conges ion, as ou se up o h ee dishes is highly unlikely o sa u a e he cell capaci y. C. THE IMPACT OF THE 15 S RECONFIGURATION INTERVAL S a link ope a es a global ne wo k con olle ha econ ig- u es he dish-sa elli e-g ound s a ion pa h e e y 15 seconds a he 12 h, 27 h, 42 nd, and 57 h seconds o e e y minu e [27],[28],[35]. A hese econ igu a ions, all ac i e clien s a e econnec ed o he sa elli es [11],[27],[35], sa el- li es econnec o he g ound s a ions [28], and equencies a e ealloca ed [5]. The dish-sa elli e econnec ion is espe- cially in e es ing because he global ne wo k con olle assigns sa elli es o dishes based on di e en ac o s such as load and geospa ial condi ions [35]. In consequence, a dish migh connec o a new sa elli e (hando e ) o econnec o he old se ing sa elli e. Ea lie s udies ound ha hese econ igu a ions cause sho , sub-second, RTT spikes and h oughpu d ops [27]. Fu he mo e, i was ound ha bu s packe loss occu s a some econ igu a ions [14],[28].This FIGURE 8. Loss. p oblem, howe e , has been mi iga ed by S a link in ea ly 2024 [28]. In his sec ion, we compa e he impac o he econ igu a- ions on he mobile and s a iona y pe o mance. We ocus ou analysis on he RTT and packe loss, since he measu emen in e al o 10 ms allows an accu a e de ec ion o he econ igu a ion imes. We don’ conside he h oughpu , because he 1 second g anula i y o ou measu emen s is insu icien o analyze sub-second beha io . We de ec he econ igu a ions in ou da a by u ilizing imes amp ma ching. Since ou clocks a e no p ecisely synch onized o S a links sys em clock, he econ igu a ions likely do no occu exac ly a he scheduled 12 h, 27 h, 42 nd, and 57 h seconds o e e y minu e. Fo each scheduled econ igu a ion, we de ine a ±1s in e al in which he econ igu a ion occu s. We u ilize he inding ha sho RTT spikes occu a each econ igu a ion [27] o de ine he imes amp o he packe wi h he maximum RTT in ou ±1s in e al o be he ime a which he econ igu a ion happens. We analyze he dis ibu ion o hese spikes and he packe loss ha occu s a ound hem. Fo his, we ca ego ize ou GER-Highway da a in o eloci y bucke s and compa e he beha io o he FHP se up o ou s a iona y measu emen s. In he ollowing analysis, econ igu a ions occu ing unde obs uc ion a e elimina ed, since obs uc ion adds addi ional ins abili ies, as we will discuss in Sec ion VII. 1) LATENCY Fig. 7 isualizes he dis ibu ions o he RTT spikes occu - ing a he econ igu a ions. All ca eloci y bucke s o ou GER-Highway da ase show compa able beha io s wi h median RTT spikes o app oxima ely 100 ms, o e lapping con idence in e als, and app oxima ely no mally dis ibu ed samples cen e ed a ound he medians. In compa ison, he RTT spikes o he s a iona y use-case a e app oxima ely 30 % lowe , wi h a median o 72 ms. The CIs do no o e lap, indica ing s a is ical signi icance. This indica es ha he RTT spike is gene ally s onge unde mo ion, bu he eloci y o he mo ion has no impac . Recon igu a ion-induced RTT spikes do no explain he signi ican ly highe RTTs o he mobile scena io shown in Fig. 6(c) because hei occu ing equency is negligible in compa ison o he o e all sample size. Elimina ing all econ igu a ion spikes om he da a leads 1274 VOLUME 6, 2025 FIGURE 16. Exce p o HTTP download h oughpu o e ime unde he GER-Highway link condi ions. Di e en measu emen se ups also a ec ou analysis o answe RQ2, limi ing ou da ase compa ison o ou GER- Highway es d i e and ou own s a iona y measu emen s. Finally, he analysis o he impac o b idges on S a link pe o mance conduc ed in Sec ion VII o answe RQ3 is limi ed by he measu emen accu acy discussed in Sec ion VII-A. This includes limi ed GPS accu acy, unce - ain ies caused by clock synch oniza ion, unce ain ies o he b idge loca ions, and unknown o ien a ions o he beam. X. FUTURE RESEARCH DIRECTIONS Ou wo k opens up in e es ing u u e esea ch di ec ions, which we ou line in his sec ion. Finding mobili y-speci ic ins abili y ac o s: In ou anal- ysis in Sec ions VI and VII, we ound ha he pe iodic 15 s S a link econ igu a ion in e al and obs uc ion caused by b idges cause sho RTT spikes and bu s packe loss. Bo h ac o s oge he accoun o 32 % o he app oxima ely 3 % o e all packe loss, bu a e negligible con ibu ing ac o s o he gene ally highe RTTs unde mobili y shown in Fig. 6(c). Finding he ac o s causing he emaining 68 % packe loss and he highe RTTs is essen ial o gain a deep unde s anding o he sys em. We see sudden o ien a ion changes o he dish, caused, e.g., by sudden changes in he d i ing di ec ion o simple ins abili ies due o une en oads, as a likely addi ional ac o as i possibly leads o subop imal beam- o ming beha io . Howe e , addi ional measu emen s udies a e needed o alida e his. De eloping mi iga ion s a egies: In Sec ion VIII,we ha e demons a ed he nega i e impac o mobili y-induced ins abili ies on HTTP bulk ile ans e as an exempla y TCP applica ion. This highligh s he necessi y o de elop app op ia e coping mechanisms. In he pas , conges ion con ol algo i hms o handle ins abili ies caused by he 15 s econ igu a ion in e al ha e been de eloped o TCP [7] and QUIC [13]. Bo h app oaches u ilize he ecu ing na u e o S a link’s econ igu a ions happening a globally de ined poin s in ime (12 h, 27 h, 42nd, and 57 h second o e e y minu e). A compa able app oach migh be applicable o o he ac o s causing ins abili ies unde mobili y, such as obs uc- ion. Especially b idges migh be i ing candida es since hei posi ions a e ixed and known. Howe e , he dynamic en i onmen changes make such an adap a ion signi ican ly mo e challenging compa ed o coping wi h he s a ic S a link econ igu a ions. A b idge p edic ion would depend on an accu a e measu emen o es ima ion o he posi ion and eloci y o he ca , making conges ion con ol managemen highly complex and si ua ion-speci ic. The complexi y can be u he inc eased by aking addi ional obs uc ion ac o s in o accoun , such as la ge buildings in ci ies. Hyb id app oaches, consis ing o a ke nel-based conges ion con ol algo i hm ha can be dynamically pa ame e ized by he applica ion laye based on he ca posi ion and eloci y, migh be easible. As an al e na i e o a new conges ion con ol algo i hm, applica ion-speci ic modi ica ions migh be used. This could be, o example, new bu e ing s a egies o ideo s eaming. Gene al impac o obs uc ion: The ou e o ou GER- Highway es d i e p o ided almos op imal, unobs uc ed condi ions o he S a link dish, as discussed in Sec ion IV. Only b idges c ossing he highway caused sho pe iods o o al obs uc ion. While his enabled an isola ed analysis o hei impac on S a link pe o mance, o he ypes o obs uc ion causing signal deg ada ion emain an open esea ch ques ion. Especially in e es ing would be o analyze he impac o la ge buildings nex o he oad, and ees co e ing he oad. Fo he o me , a es d i e h ough a ci y wi h la ge skysc ape s would be help ul, and o he la e , a es d i e h ough o es s would be necessa y. Such indings could, o example, help o assess S a link’s po en ial o open up new economic oppo uni ies in u al a eas, such as connec ing ag icul u al machines. Wea he impac unde mobili y: As discussed in Sec ion IV, he impac o basic wea he condi ions, such as p ecipi a ion, cloudiness, empe a u e, wind speed, and humidi y, on S a link pe o mance has been analyzed o he s a iona y use-case [17],[18],[21]. I was shown ha p ecipi a ion and cloudiness deg ade he downlink h ough- pu . I emains an open esea ch ques ion whe he addi ional mobili y- ela ed ac o s, such as inc easing pa icle densi ies a inc easing eloci ies o whi led wa e by p eceding ca s, in ensi y he deg ada ions caused by p ecipi a ion. Such an analysis ideally also includes o he ex eme wea he condi ions, especially hea y snow and og. Howe e , we see p ac ical challenges measu ing hese unde mobili y due o dec easing d i e s sa e y a inc easing eloci ies. VOLUME 6, 2025 1281 LANIEWSKI e al.: MEASURING MOBILE STARLINK PERFORMANCE: A COMPREHENSIVE LOOK Simula ion Models: Ex ensi e eal-wo ld measu emen s o mobile S a link pe o mance, like we did in his pape , a e qui e cos -in ensi e. Con ibu ing ac o s a e he acquisi ion cos s o he FHP dish and o a powe s a ion o powe supply, ma e ials o ins alling he dish on he oo o he ca (e.g., a oo ack), he se ice plan, uel, and, mos impo an ly, he ime o he d i e . Consequen ly, i is p ac ically in easible o always es newly designed algo i hms o applica ions ia eal-wo ld measu emen s. An accu a e simula ion model would no only sol e his issue bu would also allow esea che s ac oss he wo ld o benchma k hei de elopmen s unde de ined condi ions, ul ima ely leading o be e compa abili y o di e en solu ions. In heo y, such a simula ion model con aining all b idges c ossing Ge man highways can be buil . Howe e , in o de o make i as accu a e o he eal-wo ld as possible, addi ional ac o s which a e no ye su icien ly unde s ood, such as wea he and neighbo ing all buildings, need o be included. Channel Models: Channel models a e used o desc ibe he signal cha ac e is ics o wi eless sys ems. These cha ac e is- ics can include ac o s such as gene al signal p opaga ion, signal a enua ion (e.g., by oposphe ic e ec s), sa elli e cons ella ion cha ac e is ics (e.g., ele a ion), and use equip- men mobili y. In elease 15, he 3 d Gene a ion Pa ne ship P ojec (3GPP) has de ined guidelines o building chan- nel models o non-geos a iona y o bi (NGSO) sa elli e ne wo ks [1]. Since hen, many NGSO channel models ha e been de eloped. A comp ehensi e o e iew is gi en by Baeza e al. [3]. Howe e , none o hese models is gene ally applicable as hey all base on di e en assump ions, a e buil o di e en use-cases, o add ess di e en aspec s o he signal. Cu en ly, no channel model exis s o S a link. While such a model would be immensely help ul o he unde s anding o he sys em and he design o imp o emen s, i s de elopmen is se e ely hinde ed by he closed na u e o he sys em. XI. CONCLUSION In his pape , we ook a comp ehensi e iew on mobile S a link pe o mance om he measu emen pe spec i e. In pa icula , we answe ed he ollowing h ee esea ch ques ions. RQ1: How does mobile S a link pe o mance di e in di e en egions o he wo ld? We ound ha limi a ions o he conside ed measu emen campaigns, di e ences in measu emen se ups, and di e en esul epo ing me hods made a compa ison o uplink h oughpu and RTT di icul . In e ms o he downlink h oughpu , we ound ha Cen al Eu ope ou pe o ms he A c ic by app oxima ely 15-30 % and he U.S. by app oxima ely 40-50 %. In e es ingly, ana- lyzing he impac o he ehicle eloci y on he h oughpu s yielded pa ly con a y esul s be ween he da ase s. I is subjec o u u e wo k o shed ligh on he oo cause. RQ2: How does he mobile pe o mance compa e o s a iona y pe o mance? Th ough a compa ison o ou GER- Highway es d i e o ou own s a iona y measu emen s, we ound ha he mobile pe o mance is signi ican ly wo se han he s a iona y pe o mance. In e ms o downlink h oughpu , uplink h oughpu , RTT, PLR, and a e age bu s leng hs, he gap is 29 %, 17 %, 45 %, 814,%, and 538 %, espec i ely. In his con ex , we also ound ha he 15 s S a link econ igu a ion in e al causes signi ican ly mo e ins abili ies unde mobili y. The RTT spikes a e 42 % highe , he ac ions o econ igu a ions causing packe loss inc ease by 30 pe cen age poin s, and he a e age bu s leng hs a e up o 388 % la ge unde mobili y. The econ igu a ion- induced ins abili ies oge he wi h obs uc ion o he dish pa ially, bu no en i ely, explain he lowe pe o mance unde mobili y. Addi ional ac o s nega i ely impac ing he mobile pe o mance need o be ound in u u e wo k. RQ3: How does obs uc ion impac mobile pe o mance? We analyzed he obs uc ion caused by 98 b idges c ossing he ou e o ou es d i e. We ound ha hey cause bu s loss, RTT spikes, and d ops o he h oughpu . These pe o mance deg ada ions can occu when he ca is be o e, unde , o behind he b idge, depending on he di ec ion o he beam. These deg ada ions can se e ely deg ade he pe o mance o TCP-based applica ions such as HTTP bulk ile ans e . In u u e wo k, mechanisms o mi iga e hese e ec s should be de eloped. ETHICAL STATEMENT This wo k does no use any sensi i e da a. Following commu- ni y bes -p ac ices, we ca e ully conduc ed he measu emen s in line wi h he ai -use policy o he ne wo k p o ide . As a as we we e able o asce ain, ou measu emen s did no in e e e wi h se ice o o he use s. ACKNOWLEDGMENT The au ho s like o hank he SWO Ne z GmbH o le ing hem build hei measu emen se up in o hei an and allowing hem o ga he he da a. REFERENCES [1] “S udy on new adio (NR) o suppo non- e es ial ne wo ks ( elease 15),” 3GPP, Sophia An ipolis, F ance, Rep. 38.811, 2020. [2] M. Amend. “MP-DCCP o enabling ans e o UDP o IP a ic o e mul iple da a pa hs in mul i-connec i i y ne wo ks.” Accessed: Feb. 14, 2025. [Online]. A ailable: h ps://da a acke .ie .o g/doc/slides-104- s wg-sessb-43-ma kus-amend-mul ipa h-dccp/ [3] V. M. Baeza, E. Lagunas, H. Al-H aishawi, and S. Cha zino as, “An o e iew o channel models o NGSO sa elli es,” in P oc. 96 h IEEE Veh. Technol. Con ., 2022, pp. 1–6. [4] C. Beckman, J. Ga cia, H. Mikkelsen, and P. Pe sson, “S a link and cellula connec i i y unde mobili y: D i e es ing ac oss he a c ic ci cle,” in P oc. Wi eless Telecommun. Symp. (WTS), 2024, pp. 1–9. [5] R. Blázquez-Ga cía, D. C is allini, M. Ummenho e , V. Seidel, J. Heckenbach, and D. O’Hagan, “Expe imen al compa ison o S a link and OneWeb signals o passi e ada ,” in P oc. IEEE Rada Con ., 2023, pp. 1–6. [6] “Wo ldwide b oadband speed league 2024.” 2024. [Online]. A ailable: h ps://www.cable.co.uk/b oadband/speed/wo ldwide-speed-league/ [7] X. Cao and X. Zhang, “SaTCP: Link-laye in o med TCP adap a ion o highly dynamic LEO sa elli e ne wo ks,” in P oc. IEEE In . Con . Compu . Commun. (INFOCOM), 2023, pp. 1–10. [8] B. Fü S aßenwesen. “Fokus: B ücken.” 2024. [Online]. A ailable: h ps://www.bas .de/DE/Ingenieu bau/Fach hemen/b ueckens a is ik/ b uecken_hidden_node.h ml 1282 VOLUME 6, 2025 [9] J. Ga cia, S. Sundbe g, G. Caso, and A. B uns om, “Mul i- imescale e alua ion o S a link h oughpu ,” in P oc. 1s ACM Wo kshop LEO Ne w. Commun., 2023, pp. 31–36. [10] B. Hu e al., “LEO sa elli e s. cellula ne wo ks: Explo ing he po en ial o syne gis ic in eg a ion,” in P oc. 19 h In . Con . Eme g. Ne w. Exp. Technol. (CoNEXT), 2023, pp. 45–51. [11] L. Izhike ich, M. T an, K. Izhike ich, G. Akiwa e, and Z. Du ume ic, “Democ a izing LEO sa elli e ne wo k measu emen ,” P oc. ACM Meas. Anal. Compu . Sys ., ol. 8, no. 1, pp. 1–26, 2024. [12] J. Dee e. “John Dee e announces s a egic pa ne ship wi h SpaceX o expand u al connec i i y o a me s h ough sa elli e commu- nica ions.” 2024. [Online]. A ailable: h ps://www.dee e.com/en/ou - company/s a ic/john-dee e-pa ne ship-wi h-spacex/ [13] V. Kamel, J. Zhao, D. Li, and J. Pan, “S a QUIC: Tuning conges ion con ol algo i hms o QUIC o e LEO sa elli e ne wo ks,” in P oc. 2nd In . Wo kshop LEO Ne w. Commun. (LEO-NET), 2024, pp. 43–48. [14] M. M. Kassem, A. Raman, D. Pe ino, and N. Sas y, “A b owse -side iew o S a link connec i i y,” in P oc. 22nd ACM In e ne Meas. Con . (IMC), 2022, pp. 151–158. [15] D. Ka abi, M. Handley, and C. Roh s, “Conges ion con ol o high bandwid h-delay p oduc ne wo ks,” in P oc. ACM SIGCOMM Con . (SIGCOMM), 2002, pp. 89–102. [16] T. Kohns amm, “E e y hing you need o know abou p ojec Kuipe , Amazon’s sa elli e b oadband ne wo k.” 2024. [Online]. A ailable: h ps://www.abou amazon.com/news/inno a ion-a -amazon/ wha -is-amazon-p ojec -kuipe [17] E. Lan e , D. Laniewski, D. O en, and N. Aschenb uck, “Wea he - based link p edic ion o LEO-sa elli e ne wo ks using he We Links da ase ,” in P oc. IFIP Ne w. Con ., 2024, pp. 586–588. [18] E. Lan e , D. Laniewski, M. Wehmeie , and N. Aschenb uck, “Obse ing he skies—G ound-based cloud de ec ion o e alua ing he impac o clouds on LEO communica ions,” in P oc. 2nd In . Wo kshop LEO Ne w. Commun. (LEO-NET), 2024, pp. 19–24. [19] D. Laniewski, E. Lan e , and N. Aschenb uck, “S a link-on- he- au obahn da ase .” 2025. [Online]. A ailable: h ps://gi hub.com/sys- uos/S a link-on- he-Au obahn [20] D. Laniewski, E. Lan e , S. Beginn, J. Dunke , M. Dücke s, and N. Aschenb uck, “S a link on he oad: A i s look a mobile S a link pe o mance in cen al Eu ope,” in P oc. 8 h Ne w. T a . Meas. Anal. Con . (TMA), 2024, pp. 1–8. [21] D. Laniewski, E. Lan e , B. Meije ink, R. an Rijswijk-Deij, and N. Aschenb uck, “We Links: A la ge-scale longi udinal S a link da ase wi h con iguous wea he da a,” in P oc. 8 h Ne w. T a . Meas. Anal. Con . (TMA), 2024, pp. 1–9. [22] D. Laniewski e al., “Demo: The impac o LEO sa elli e ne wo k ins abili ies on he pe o mance o ne wo king applica ions,” in P oc. 49 h IEEE Con . Local Compu . Ne w. (LCN), 2024, pp. 1–4. [23] M. López, S. B. Damsgaa d, I. Rod íguez, and P. Mogensen, “Connec ing u al a eas: An empi ical assessmen o 5G e es ial- LEO sa elli e mul i-connec i i y,” in P oc. 97 h IEEE Veh. Technol. Con ., 2023, pp. 1–5. [24] S. Ma, Y. C. Chou, H. Zhao, L. Chen, X. Ma, and J. Liu, “Ne wo k cha ac e is ics o LEO sa elli e cons ella ions: A S a link- based measu emen om end use s,” in P oc. IEEE In . Con . Compu . Commun. (INFOCOM), 2023, pp. 1–10. [25] J. McDowell. “Eno mous (‘mega’) sa elli e cons ella ions.” 2024. [Online]. A ailable: h ps://plane 4589.o g/space/con/conlis .h ml [26] F. Michel, M. T e isan, D. Gio dano, and O. Bona en u e, “A i s look a S a link pe o mance,” in P oc. 22nd ACM In e ne Meas. Con . (IMC), 2022, pp. 130–136. [27] N. Mohan e al., “A mul i ace ed look a S a link pe o mance,” in P oc. ACM Web Con . (WWW), 2024, pp. 2723–2734. [28] J. Pan, J. Zhao, and L. Cai, “Measu ing he sa elli e links o a LEO ne wo k,” in P oc. IEEE In . Con . Commun. (ICC), 2024, pp. 4439–4444. [29] “NTP FAQ.” 2022. [Online]. A ailable: h p://www.n p.o g/n p aq/ NTP-s-algo/#5131-how-accu a e-will-my-clock-be, [30] Guidelines o he Design o Mo o ways, Road T ansp. Res. Assoc., Cologne, Ge many, 2008. [31] “S a link la high pe o mance ki speci ica ions.” SpaceX. 2024. [Online]. A ailable: h ps://api.s a link.com/public- iles/ speci ica ion_shee _ la _high_pe o mance.pd [32] “S a link se ice plan speci ica ions.” SpaceX. 2024. [Online]. A ailable: h ps://www.s a link.com/legal/documen s/DOC-1400- 28829-70 [33] “S a link s anda d ki speci ica ions.” SpaceX. 2024. [Online]. A ailable: h ps://api.s a link.com/public- iles/ S a li%20P odu%20Speci ica ions_S anda d.pd [34] “S a link s anda d speci ica ions.” SpaceX. 2024. [Online]. A ailable: h ps://api.s a link.com/public- iles/speci ica ion_shee _s anda d.pd [35] H. B. Tan ee , M. Puchol, R. Singh, A. Bianchi, and R. Ni hyanand, “Making sense o cons ella ions: Me hodologies o unde s anding S a link’s scheduling algo i hms,” in P oc. 19 h In . Con . Eme g. Ne w. Exp. Technol., 2023, pp. 37–43. [36] “ThinKom and Telesa expand ag eemen o low ea h o bi ope a ions,” Telesa . 2024. [Online]. A ailable: h ps://www. elesa . com/p ess/p ess- eleases/ hinkom-and- elesa -expand-ag eemen - o - low-ea h-o bi -ope a ions/ [37] “The ne em De elope s: Ne em—Ne wo k emula o .” Accessed: Feb. 14, 2025. [Online]. A ailable: h ps://www.man7.o g/linux/man- pages/man8/ c-ne em.8.h ml [38] I. Tsa e a, T. V. Doan, and V. Bajpai, “A decade long iew o in e ne a ic composi ion in Japan,” in P oc. IFIP Ne w. Con . (IFIP Ne w.), 2023, pp. 1–9. DOMINIC LANIEWSKI (G adua e S uden Membe , IEEE) ecei ed he B.S. deg ee in in o ma ion sys ems om he Uni e si y o Müns e in 2017, and he M.S. deg ee in compu e science om Osnab ück Uni e si y in 2019, whe e he is cu en ly pu suing he Ph.D. deg ee wi h he Dis ibu ed Sys ems G oup. His esea ch in e es s include poin cloud and ideo s eaming, In e ne measu emen s, machine lea ning o ne wo ks, and obo communica ions. ERIC LANFER (G adua e S uden Membe , IEEE) ecei ed he mas e ’s deg ee in compu e sci- ence om Osnab ück Uni e si y, Ge many. He is cu en ly pu suing he Ph.D. deg ee wi h he Dis ibu ed Sys ems G oup, Osnab ück Uni e si y, whe e his esea ch p ima ily cen- e s on ne wo k secu i y, pa icula ly in he a ea o ne wo k in usion de ec ion. His aca- demic jou ney includes in e na ional expe iences wi h S ockholm Uni e si y and he Uni e si y o Twen e. Addi ionally, he con ibu es o eaching and esea ch in he ield o In e ne measu emen s and wo ks on p ojec s, such as explo ing he pe o mance o sa elli e ne wo ks. NILS ASCHENBRUCK (Membe , IEEE) ecei ed he G adua e Diploma and Ph.D. deg ees in compu e science om Bonn Uni e si y, Ge many, in 2003 and 2008, espec i ely. He was a Senio Resea che and he Head o he Resea ch A ea “Tac ical Wi eless Mul i-Hop Ne wo ks” wi h he Communica ion Sys ems G oup, Bonn Uni e si y, whe e he has been holding a Tenu ed P o esso ship o Dis ibu ed Sys ems since 2012. His esea ch ocus is on dependable and obus ne wo ked sys ems including scena io modeling, a ic enginee ing, and ne wo k secu i y. VOLUME 6, 2025 1283