Pub­lic In­ter­net

3:20PMMay 2 2018Daniel Tompkins

Archive

In­tro­duc­tion

Nearly a decade ago, Face­book pi­o­neered the con­cept of so­cial media. How­ever, it wasn’t until re­cently— as the plat­form boasts 2.1 bil­lion users— that the full con­se­quences of such an ubiq­ui­tous net­work have burst into the public eye.1 Now, Zucker­berg’s famed motto, “move fast and break things,” may have in fact con­tributed to a broken democ­racy.

In his re­cent tes­ti­mony be­fore Con­gress, it also be­came ev­i­dent that our elected rep­re­sen­ta­tives' lack of dig­ital lit­eracy only serves to ex­ac­er­bate the sit­u­a­tion.2

Tech­nology and Pol­i­tics

Tech­nology has had an ever more in­ti­mate re­la­tion­ship with pol­i­tics— and I'm ap­pro­pri­ating the term tech­nop­o­l­i­tics to de­scribe this en­tan­gle­ment. It's a pur­pose­fully broad term, a hy­per­ob­ject (to borrow a term from Tim­othy Morton) for rec­on­ciling dis­parate processes into dis­crete events and be­hav­iors.3

Tech­nop­o­l­i­tics could be i.e., using the In­ternet to in­flu­ence po­lit­ical cam­paigns. It could be psy­cho­graph­i­cally cu­rated in­for­ma­tion on so­cial media, policy and reg­u­la­tion changes (GDPR), net neu­trality, and much more.4 My op­ti­mism is that "tech­nop­o­l­i­tics" will also pro­mote de­lib­er­a­tion, es­pe­cially in an­swering the fol­lowing ques­tions:

How have wire­less in­for­ma­tion net­works sup­ported democ­racy and shared ex­pe­ri­ences? How does net­work topology af­fect its per­for­mance in this ca­pacity?

Dig­ital Public Spaces

Face­book's con­tin­uous and re­lent­less im­mer­sion in tech­nop­o­l­i­tics de­serves at­ten­tion. In some ways, the ser­vice has be­come a dig­ital pri­vately-owned public space (POPS). In the past, de­ter­minedly public spaces have been legally pro­tected as arenas of de­lib­er­a­tive prac­tice and free speech. The growing pri­va­ti­za­tion of these spaces, how­ever, wicks away some al­lowance of free ex­pres­sion— bol­stering the legal de­fense of any im­posed reg­u­la­tions by sit­u­ating it­self as bor­rowed, rather than public.

Own­er­ship and au­thority are counter-in­tu­itive to public de­lib­er­a­tion. His­tor­i­cally, the com­mons— or vil­lage green— was shared among mul­tiple owners and would func­tion­ally tran­si­tion to suit each of the user's unique needs. De­spite the chal­lenges of POPS, they have often served as pop­ular lo­ca­tions for shared ex­pe­ri­ence and the ex­change of in­for­ma­tion or news.

For ex­ample, the public house (a "pub", or tavern) was once a prime hub; as was the modern mall— cap­i­tal­ism's ideal in­ter­sec­tion with de­mo­c­ratic de­lib­er­a­tion. Malls were once vi­brant cen­ters of shared ex­pe­ri­ence among a di­verse de­mo­graphic.

Since 2002, how­ever, the so-called "re­tail apoc­a­lypse" has erased 480,000 jobs at these so­cial mar­ket­places.5 With Amazon, the com­mer­cial func­tion has be­come ob­so­lete, but what about its so­cial func­tion? If people are adapting so­cial media to sim­u­late these in­ter­ac­tions, how does Face­book differ from the mall, the park, or the plaza?

Psy­cho­graphic Mod­eling

Al­though Face­book might pro­vide an ad­e­quate plat­form for or­ga­ni­za­tion and de­lib­er­a­tion, it is not the prop­erty of its users. Ad­di­tion­ally, Face­book in some ways re­sists ap­pro­pri­a­tion by pro­viding a stan­dard for com­mu­ni­ca­tion in ways that phys­ical space re­mains am­biguous. Fur­ther­more, Face­book's com­plex al­go­rithms or­ga­nizing in­for­ma­tion have been found to avoid serving any­thing "dis­agree­able" to the tastes and opin­ions of you, or your psy­cho­graphic model.6 The lack of ide­o­log­i­cally op­po­si­tional con­tent has re-en­forced, and po­lar­ized, fil­tered “echo cham­bers”— fa­cil­i­tating par­tisan con­flict and dis­trust.

As of 2017, some 67% of Amer­i­cans "get at least some of their news" from so­cial media. There­fore, an in-depth un­der­standing of how these net­works be­have is crit­ical. From 1990 to 2016, the number of people working in jour­nalism has fallen by about 250,000.7 Many of these jobs have simply dis­ap­peared— dis­abled by the hy­brid media en­vi­ron­ment brought about by the In­ternet.8

The Death of Local News

In a talk at Har­vard's HKS school, The Wash­ington Post media re­porter Mar­garet Sul­livan cited the death of local news as the greatest threat to democ­racy.9 Local news has largely been sub­sumed by the Face­book “feed”, Twitter and other so­cial media sources— at­tracting users with in­for­ma­tion ag­gre­ga­tion and con­stant, rapid avail­ability and con­ve­nience.

The trend that I’d like to ad­dress in this paper is the adop­tion of small-scale, dis­trib­uted "in­tranets". Sev­eral ini­tia­tives, usu­ally in larger cities, are al­ready un­derway (the Red­hook Ini­tia­tive, De­troit Com­mu­nity Tech­nology Pro­ject, NYC Mesh, Hy­per­boria, Pro­ject Byzan­tium, and Li­breMesh among others in Eu­rope) to bring af­ford­able— even free— In­ternet-ac­cess, usu­ally to ne­glected and poor com­mu­ni­ties within the US.

Es­sen­tially, these net­works dis­tribute wire­less con­nec­tions from a se­ries of inter-con­nected hubs, breaking up the di­rect In­ternet ser­vice-provider (ISP) to con­sumer re­la­tion­ship in favor of a cheaper com­mu­nity-driven al­ter­na­tive. At pre­sent, these net­works are merely re-dis­trib­uting a full-fledged In­ternet con­nec­tion; how­ever, Cuba's his­toric un­der­ground SNET, or “street net­work”, pre­sents an in­triguing ex­ample for how these mesh net­works could evolve with in­ternal grass­roots sup­port.

A case study of the Cuban SNET— its net­work pro­to­cols, in­fra­struc­tural as­sembly, and grass­roots cul­ture— could lend valu­able in­sight into how com­mu­ni­ties in the United States might re­struc­ture their own in­for­ma­tion pipelines in sup­port of po­lit­ical and so­cial ag­o­nism and democ­racy. Ad­di­tion­ally, I hy­poth­e­size that such a net­work ap­plied en masse— but cen­tered around local in­for­ma­tion ecolo­gies— could re­vi­talize news (pro­viding ac­count­ability against mis­in­for­ma­tion, or “fake news”), draw power away from big tech, and en­courage vig­orous cul­tural growth on­line and off.

Dig­ital Net­works

Re­cently, I've seen re­newed in­terest in re­vising the models for how we ex­change data and in­for­ma­tion across the Web. Juan Benet, most no­tably, in­tro­duced the idea of an In­ter­plan­e­tary File System (IPFS) that would dy­nam­i­cally record its "state"— an en­crypted snap­shot of its col­lec­tive data.10

This "soft" net­work has al­ready begun to en­list or­di­nary com­puters for main­taining this cryp­to­graphic "ledger" of the In­ternet. Through this model, it cre­ates a trans­forming dig­ital image of the In­ternet. Ad­di­tion­ally, this system could al­to­gether pre­vent the loss of data by archiving the Web into dis­crete, re­triev­able packets.

"Mesh" Net­works

An­other in­ter­esting de­vel­op­ment— this time in the con­sumer market— are com­pa­nies like goTenna.11 Their mesh net­work don­gles are es­sen­tially In­ternet-age walkie-talkies. In­stead of sound, goTenna's de­vices con­nect to phones or lap­tops— ex­panding the types of media and in­for­ma­tion that can be shared without a cen­tral routing au­thority. Ad­di­tion­ally, apps— like Dr. Gardner-Stephen's Serval— are starting to turn phones into ad hoc "nodes" in self-as­sem­bling net­works.12

In 2015, an­other app— FireChat— was tested along with goTenna's de­vices at Burning Man. Sim­ilar to Serval, FireChat gen­er­ated a self-as­sem­bling, peer-to-peer net­work that al­lowed at­ten­dees to con­nect with one an­other— without needing to proxy through a data-center or com­mer­cial car­rier.13

Now that the In­ternet fan­tasy has over­whelmed the skep­tics and made its way into 80% of Amer­ican house­holds, people are be­gin­ning to re-eval­uate it's his­tor­ical func­tion and how it can better adapt to modern media, cul­ture, and pol­i­tics.14 The con­cept of a wire­less In­ternet— as well as the ethe­real "Cloud"— pro­motes some­what of a farce: that the In­ternet does not take up phys­ical space.

Local-first Net­works

In re­ality, data are stored on servers, and on your own per­sonal com­puters. The In­ternet de­scribes these fa­cil­i­ties, but be­yond that it points to the in­fra­struc­ture that makes the ex­change of data pos­sible be­tween those stores. There is also a dis­tinc­tion here from the World-Wide Web (WWW)— which de­scribes the dig­ital or­ga­ni­za­tion of the data the In­ternet car­ries and dis­trib­utes.

In Al­bert-László Barabási and Réka Al­bert's Emer­gence and Scaling in Random Net­works, the topog­raphy of a scale-free net­work— like the WWW— fol­lows a unique set of ob­serv­able laws, or be­hav­iors.15 One such be­havior re­calls the "rich-get-richer" phe­nomena, or pref­er­en­tial at­tach­ment.

This prop­erty models the way in which new nodes as­semble into a net­work, showing pref­er­ence to­ward nodes with high de­gree (con­nec­tivity). As a re­sult, well-es­tab­lished node be­come in­creas­ingly ver­tical. Franklin Foer, in his book World Without Mind, is crit­ical of the Frightful Five (Google, Amazon, Apple, Mi­crosoft, and Face­book) and the way in which they've cap­i­tal­ized on this con­cen­tra­tion of ac­tivity and power.16

Re-Imag­ining the In­ternet

In a study con­ducted in Changsha, China— De­gree and con­nec­tivity of the In­ter­net's scale-free topology— the re­searchers de­manded:

In order to better un­der­stand the dy­namic be­hav­iors of the In­ternet, to ac­cu­rately de­pict the topo­log­ical struc­ture of the In­ternet and to avail­ably con­trol the ef­fi­cient op­er­a­tion of the In­ternet there should be fur­ther re­search on the sta­tis­tical char­ac­ter­is­tics of In­ternet topology.17

In the same study, the three re­searchers built a math­e­mat­ical model, not of the WWW, but of the In­ternet— sim­ilar to that of Barabási and Réka, but "at the au­tonomous sys­tems (AS) level."18 As­serting that the av­erage de­gree— the con­nec­tivity of a node (e.g., a server or wire­less base-sta­tion)— is one of the most im­por­tant vari­ables at in­di­cating other global net­work be­hav­iors and prop­er­ties.

The con­nec­tivity of nodes on the In­ternet has been found to follow a sim­ilar power-law dis­tri­b­u­tion, with the most con­nected nodes being very few, but reaching a ma­jority of the re­mainder. Like­wise, the least con­nected nodes make up the ma­jority, but have a smaller av­erage de­gree.

Net­work Topology

A high-de­gree node in iso­la­tion might be said to have more of a star topology— one origin serving and con­necting many ad­ja­cent nodes— which car­ries risk by con­cen­trating data and traffic over a single node. Hun­dreds of thou­sands of com­puters might rely on a single hub. Be­cause of its high de­gree, it will also act as an in­ter­me­diary for ex­changes be­tween other nodes. As a re­sult, if the hub were dis­abled, those com­puters might also lose con­tact with one an­other.

Speaking to po­lar­iza­tion, highly con­nected nodes that be­come com­pro­mised— per­haps pro­moting a single ide­ology, or cen­soring op­posing in­for­ma­tion— might also con­tribute to a rad­i­cal­iza­tion of opinion. In an ideal dis­trib­uted net­work, each node is able to reach every other node in the net­work. The dis­tri­b­u­tion of the net­work— as well as data re­dun­dan­cies— could sig­nif­i­cantly mit­i­gate this risk.

While the star model— a single hub with high con­nec­tivity— deals with heavy traffic by "widening the pipes" (amassing more storage, elec­tricity, and com­puting power), the modern "mesh" net­work han­dles traffic in a dif­ferent way. In a sense, they are "self-as­sem­bling"— i.e. dy­namic— using met­rics like phys­ical prox­imity, signal strength, and shortest walk (fewest "hops") to de­ter­mine which nodes should be ac­ti­vated.

How our de­vices ef­fi­ciently nav­i­gate this net­work is a dif­fi­cult problem, and it has proven to be a sig­nif­i­cant ob­stacle for pop­ular adop­tion. An ad hoc net­work or routing pro­tocol— rather than re­lying on static con­nec­tions— re­acts to changing con­di­tions and per­sis­tent con­nec­tions de­spite mo­bile nodes. In this way, a user on the net­work might walk from source A to source B, and as they cross the threshold from one signal to the other the signal switches without dis­con­necting.

Sev­eral other open-source pro­jects have de­vel­oped al­ter­na­tive pro­to­cols to suit a new dis­trib­uted in­fra­struc­ture. Like the ham radio craze be­fore it, hun­dreds of people are moving to these net­works and dis­cussing their im­ple­men­ta­tion. I'd also like to note that across dif­ferent fo­rums, I've often seen the term "mesh" net­works. For the most part, I think they are thinking de­cen­tral­ized.

Mesh net­works seem to de­scribe a more per­fect grid of routing de­vices and servers. I don't think this is what they're hoping for, and it's not the way net­works will want to nat­u­rally be­have. A de­cen­tral­ized net­work, how­ever, does calls for a flat­tening of de­grees— more ho­mo­geneity across local nodes and re­peaters.

Cuban SNET

The Cuban SNET, or "street net­work", was first con­ceived by the on­line gaming com­mu­nity in Ha­vana when teenagers started stringing LAN ca­bles across al­leys to play World of War­craft, Call of Duty and other ti­tles. This also al­lowed for IRC (an early form of In­ternet cha­t­room) and file-sharing, which drew more pop­ular sup­port. A more or­ga­nized push brought these small-scale LAN net­works to­gether, and cre­ated for the first time a dig­i­tally con­nected Ha­vana.

For a long time, gov­ern­ment workers and tourists were the only ones given ac­cess to the World Wide Web. Now, Cubans can pur­chase scratch-cards for timed con­nec­tions at public hotspots sit­u­ated in Ha­vana, and other cities. They are ex­pen­sive, and the con­nec­tion is crowded and slow.

It would be il­legal for anyone to get In­ternet ac­cess without per­mis­sion from ETECSA, the state comms com­pany. How­ever, they've largely turned a blind eye to­ward the SNET, and it has been al­lowed to grow (and sim­ilar net­works of smaller size have popped up else­where on the is­land)— yet its mod­er­a­tors strictly dis­allow re­li­gious or po­lit­ical speech, sex­u­ally ex­plicit ma­te­rial, or com­mer­cial pro­mo­tion in order to en­sure its con­ti­nuity.

As a re­sult of a com­pli­cated his­tory of iso­la­tion— mostly as a re­sult of the ag­gres­sive diplo­macy of the United States— Cuba de­vel­oped without a dig­ital link to the world. The in­for­ma­tion quar­an­tine, as well as the thou­sands of job­less pro­fes­sionals, cre­ated a cul­tural pre­con­di­tion for the in­ternal growth of these net­works.

In Ha­vana, the SNET has es­tab­lished it­self as a func­tioning al­ter­na­tive to the global Web— in a country with only 5.6% house­hold In­ternet pen­e­tra­tion.19 The only reason a global con­nec­tion is ever used is gen­er­ally to check Face­book or face­time a family member abroad— an ac­tual ge­o­graphic sep­a­ra­tion to which the SNET re­mains dis­mem­bered.

The gov­ern­ment hasn't made it an easy task. Pur­chasing routers or other wire­less tech in the 2.4GHz or 5GHz range also re­quires per­mis­sion from ETECSA, and can't be brought into the country. As a re­sult, these so­cial and tech­nical en­tre­pre­neurs must be careful not to over­step the law— or, not to get caught doing so...

A re­cent (2017) study shows that the SNET is cen­tered around sev­eral hubs, or pil­lars:

While an "in­di­vidual node may con­nect up to 200 users... a pillar may con­nect tens of re­gional nodes." Each pillar "has a dis­tinc­tive name, some­time re­flecting ge­og­raphy or the SNET's origin in gamer net­works. The pil­lars' re­gions some­times overlap, but to­gether they cover a sig­nif­i­cant por­tion of the Ha­vana met­ro­pol­itan area."20

El Pa­quete Se­m­anal

Many of the movies, TV shows, and music pop­ular in the United States are phys­i­cally dis­trib­uted via hard drive— what's known as El Pa­quete Se­m­anal, the weekly packet. The re­port notes that heavy-band­width ac­tivity is lim­ited out­side of the hours of 3AM to noon, during which time El Pa­quete can be beamed across the pil­lars.

The paper marks 2011 as the be­gin­ning of the SNET, but per­haps that is the ac­tual de­ploy­ment of the ".snet" top level do­main (e.g., .com, .org, .net). I've read in dif­ferent places that the gamer net­works started as early as 1998.

The study also men­tions that the SNET uses an open shortest path first (OSPF) routing pro­tocol— a variant of Dijk­stra's shortest walk al­go­rithm used to search for the quickest path across an in­ter­con­nected web of nodes. An­other in­ter­esting note in the re­port added that, "each pillar is con­nected to at least two others."

As with the SNET, local net­work ini­tia­tives in the US are using a pow­erful set of ac­cess points and re­peaters to stretch sig­nals from more pow­erful hubs. In re­ality, it is ge­o­graph­i­cally des­ig­nating in­vis­ible zones of in­tranet cov­erage with unique top-level do­mains.

Con­clu­sion

The re­la­tion­ship of ISP-to-con­sumer gives only a single op­tion: the In­ternet. I wonder if the de­vel­oping de­cen­tral­ized net­works might begin to shape an all-wire­less, com­mu­nity-as­sem­bled Public In­ternet— a par­allel sur­face to the com­mer­cial In­ternet that can be switched on or off.

Imag­ining, for ex­ample, Brooklyn as a con­tainer for a se­ries of com­mu­nity in­tranets— dis­con­nected from other com­mu­nity net­works. Not com­pletely dis­con­nected, but having the ability to bridge, filter, or query the con­tents of this Public In­ternet.

By map­ping this kind of dig­ital com­mons to phys­ical space and lim­iting each clus­ter's ex­po­sure to high-de­gree nodes, this sep­a­ra­tion might in­crease the av­erage de­gree of nodes within each in­tranet. In this way, it might re­semble a se­ries of dig­ital arch­i­pel­agos.

The way dis­tinct cul­tures, or in­sti­tu­tions, de­velop around these dig­ital spaces, and the way they could be ap­pro­pri­ated for de­lib­er­a­tive democ­racy is also in­triguing con­sid­ering the pos­si­bility of higher av­erage de­gree within each dig­ital com­mons and how that pro­ceeds to con­nect these dis­tricts.

Lastly, it seems that pop­ular par­tic­i­pa­tion in the as­sembly of these net­works might be vital in de­ter­mining their suc­cess. In­ternal main­te­nance, de­vel­op­ment, apps and ser­vices that make this dig­ital public space unique should be an im­por­tant value that these com­mu­ni­ties build. Al­to­gether, I be­lieve it could also foster con­tinued dig­ital lit­eracy, learning, in­for­ma­tion equality and com­mu­nity con­nec­tivity.

Footnotes

  1. Face­book, Inc. "Face­book Q4 2017 Re­sults." News re­lease. In­vestor.fb.com. Ac­cessed May 4, 2018. https://​s21.q4cdn.com/​399680738/​files/​doc_­fi­nan­cials/​2017/​Q4/​Q4-2017-Earn­ings-Pre­sen­ta­tion.pdf .

  2. The Guardian. US News: Opinion. "Mark Zucker­berg's Face­book Hearing Was an Utter Sham." News re­lease, April 11, 2018. Www.the­guardian.com . Ac­cessed May 4, 2018. https://​www.the­guardian.com/​com­men­t­is­free/​2018/​apr/​11/​mark-zucker­bergs-face­book-hearing-sham .

  3. Morton, Tim­othy. Hy­per­ob­jects: Phi­los­ophy and Ecology after the End of the World. Min­neapolis: Uni­ver­sity of Min­nesota Press, 2014.

  4. "EU GDPR In­for­ma­tion Portal." EU GDPR Portal. Ac­cessed May 04, 2018. https://​www.eu­gdpr.org/ .

  5. San­burn, Josh. "Why the Death of Malls Is About More Than Shop­ping." TIME Mag­a­zine, July 20, 2017. Ac­cessed May 4, 2018. https://​time.com/​4865957/​death-and-life-shop­ping-mall/ .

  6. Aue, Mary Von. "Cam­bridge An­a­lyt­i­ca's ‘Psy­cho­graphic Mod­eling Tech­nique’: What to Know." In­verse (blog), March 18, 2018. Ac­cessed May 4, 2018. https://​www.in­verse.com/​ar­ticle/​42459-psy­cho­graphic-mod­eling-tech­nique-face­book-data .

  7. Bu­reau of Labor Sta­tis­tics, U.S. De­part­ment of Labor, The Eco­nomics Daily, Em­ploy­ment trends in news­paper pub­lishing and other media, 1990–2016 on the In­ternet. Ac­cessed May 4, 2018. https://​www.bls.gov/​opub/​ted/​2016/​em­ploy­ment-trends-in-news­paper-pub­lishing-and-other-media-1990-2016.htm .

  8. Here, I'm refering to the hy­brid media en­vi­ron­ment posited by David Karpf in An­a­lytic Ac­tivism | Karpf. An­a­lytic Ac­tivism. Place of Pub­li­ca­tion Not Iden­ti­fied: Ox­ford Uni­ver­sity Press, 2016.

  9. Sul­livan, Mar­garet. "Mar­garet Sul­livan: The State of the Media in 2018." In­ter­view by Nicco Mele. Www.hks.har­vard.edu . Feb­ruary 6, 2018. Ac­cessed May 4, 2018. https://​www.hks.har­vard.edu/​events/​mar­garet-sul­livan-state-media-2018 .

  10. Labs, Pro­tocol. "IPFS Is the Dis­trib­uted Web." Lit­erary Merit. Ac­cessed May 04, 2018. https://​ipfs.io/ .

  11. "About GoTenna." GoTenna. Ac­cessed May 04, 2018. https://​www.gotenna.com/​pages/​about .

  12. Gardner-Stephen, Paul. "The Serval Pro­ject: Re­claiming Your Phone." Huff­post, Sep­tember 29, 2013. Ac­cessed May 4, 2018. https://​www.huff­in­g­ton­post.com/​paul-gard­ner­stephen/​the-serval-pro­ject-re­clai_b_3677089.html .

  13. Yu, Alan. "How One App Might Be A Step To­ward In­ternet Every­where." Www.npr.org , April 7, 2014. Ac­cessed May 4, 2018. https://​www.npr.org/​sec­tions/​all­tech­con­sid­ered/​2014/​04/​07/​298925565/​how-one-app-might-be-a-step-to­ward-in­ternet-every­where .

  14. Rainie, Lee, and D'Vera Cohn. "Census: Com­puter Own­er­ship, In­ternet Con­nec­tion Varies Widely across U.S." Pew Re­search Center, Sep­tember 19, 2014. Ac­cessed May 4, 2018. http://​www.pewre­search.org/​fact-tank/​2014/​09/​19/​census-com­puter-own­er­ship-in­ternet-con­nec­tion-varies-widely-across-u-s/# .

  15. Barabasi, Al­bert-Laszlo, and Reka Al­bert. "Emer­gence of Scaling in Random Net­works." Sci­ence, Oc­tober 15, 1999.

  16. Foer, Franklin. World Without Mind: The Ex­is­ten­tial Threat of Big Tech. Pen­guin Press, 2017.

  17. Zhang, Lian-Ming, Xiao-Heng Deng, Jian-Ping Yu, and Xiang-Sheng Wu. "De­gree and Con­nec­tivity of the In­ter­nets Scale-free Topology." Chi­nese Physics B 20, no. 4 (2011): 048902. doi:10.1088/​1674-1056/​20/​4/​048902.

  18. Ibid. 17

  19. Pujol, Ed­uardo E. P., Will Scott, Eric Wus­trow, and J. Alex Hal­derman. "Ini­tial Mea­sure­ments of the Cuban Street Net­work." Pro­ceed­ings of the 2017 In­ternet Mea­sure­ment Con­fer­ence on - IMC 17, 2017. doi:10.1145/​3131365.3131395.

  20. Ibid. 19