[00:00:06.19] uh hi everybody um i am uh ellen zegura um in the school of computer science and uh [00:00:16.12] [00:00:16.12] courtesy appointment and interactive computing and i am delighted to uh have the chance to introduce [00:00:22.15] [00:00:22.15] curtis emerald to you all um as the speaker today um i i personally i have known each other for a [00:00:29.05] [00:00:29.05] while um we we have in common uh networking and an interest in um in connecting uh and doing work [00:00:38.06] [00:00:38.06] in communities i think the last time we saw each other in person was curtis check me here india [00:00:45.05] [00:00:45.22] possible um uh so so that was a long time and a long way away um so curtis is on the faculty uh [00:00:56.15] [00:00:56.15] at the paul allen school of computer science and engineering at the university of washington [00:01:01.03] [00:01:01.03] um and uh he has been there for um a while not too long medium amount of time uh and he [00:01:12.15] [00:01:12.15] came to to uw from uh uc berkeley uh and and i one of the things that i really um [00:01:19.16] [00:01:20.12] uh kind of admire about curtis and his work is that he's part of a kind of [00:01:24.21] [00:01:25.11] first generation i think of of faculty whose work is squarely in information and communication [00:01:32.21] [00:01:32.21] technologies and development so kind of uh um uh you know from day one um centrally concerned with [00:01:41.12] [00:01:41.12] issues of economic political social development and their intersection with technologies for a [00:01:48.19] [00:01:48.19] while the folks working in that area had largely come to it after or in addition to doing other [00:01:55.07] [00:01:55.07] kinds of things and so curse that great example of someone who's doing it um kind of full-on um so [00:02:03.07] [00:02:03.20] uh and i think the kind of work he'll talk about is very much in line with [00:02:11.09] [00:02:13.03] national priorities as reflected in things like smart connected communities [00:02:17.16] [00:02:19.01] projects and solicitation out of the national science foundation as an example [00:02:22.17] [00:02:23.07] so without any more talk from me curtis um i'm really excited to hear your talk and very happy [00:02:30.00] [00:02:30.00] to have you visiting virtually great thank you so much for the introduction and yeah i was in uh [00:02:34.23] [00:02:37.05] i think it might have been my last international trip um uh it seemed so long ago uh so welcome [00:02:43.01] [00:02:43.01] everybody uh as mentioned i'm curtis and i'm here to talk about uh my labs work on reclaiming the [00:02:47.18] [00:02:47.18] internet our journey to community held cellular infrastructure so brief aside i want to talk about [00:02:53.22] [00:02:53.22] the lab i co-direct the ictd lab at the university of washington um as ellen mentioned ict is [00:03:00.08] [00:03:01.16] information communication technologies and development and the way i like to sell that [00:03:05.11] [00:03:05.11] is largely just uh focusing on technology and poverty alleviation so the interaction of those [00:03:11.18] [00:03:11.18] two elements with a specific kind of builder focus uh in my lab at least uh the other uh lead on this [00:03:18.17] [00:03:18.17] is richard anderson he focuses on a wide range of topics primarily health financial services uh and [00:03:24.10] [00:03:24.10] gender uh while i focus on universal internet access which is the body of this talk [00:03:29.03] [00:03:29.03] but there's other work on conservation um in a measurement and a bunch of other fun stuff [00:03:33.20] [00:03:35.05] so there's a high level of what this talk is going to look like just to be clear the goal of my lab [00:03:39.20] [00:03:40.12] and these research efforts is to get everyone everywhere onto the internet [00:03:44.00] [00:03:44.00] and other related critical technologies in an affordable and empowering way [00:03:48.06] [00:03:48.06] um and so we're going to talk through our schemes to get that done [00:03:51.12] [00:03:51.12] we're going to start with a little bit of a background introduction into the space and [00:03:54.13] [00:03:54.13] then talk about three projects two of which i've seen significant publications and then the third [00:03:59.11] [00:03:59.11] which we're still working on these are community cellular manager which has a related deployment [00:04:03.16] [00:04:03.16] with a carrier or telco partner in the philippines uh culty the community lte project with related [00:04:09.07] [00:04:09.07] deployment with community partners in indonesia and our new efforts on cooperative cellular [00:04:13.09] [00:04:13.09] and related deployments with community partners here in seattle uh and down tacoma washington [00:04:17.07] [00:04:18.13] so to start i'm going to lead with just a little bit of a history and this is starting [00:04:22.13] [00:04:22.13] with the history of the internet so as people may know we have out there arpanet arpanet was [00:04:27.18] [00:04:27.18] the original founding of the internet and the goal of arpanet was really to share mainframe resources [00:04:33.05] [00:04:33.05] among universities while also experimenting with sort of packet designs the idea was that you would [00:04:39.11] [00:04:39.11] put a mainframe in at stanford and through this network people would get access other universities [00:04:43.11] [00:04:43.11] would get access to it now this network grew up it matured it became nsf net which was this big [00:04:49.09] [00:04:49.09] long backbone of uh the research network in the us and eventually became uh privatized in 1990 and so [00:04:56.02] [00:04:56.02] the stewards of the internet at the time needed to figure out what kind of network what kind of [00:04:59.18] [00:04:59.18] private network did they want the internet to be and the system they designed was one with numerous [00:05:05.05] [00:05:05.22] individual actors working together to provide connectivity to communities and individuals and [00:05:10.13] [00:05:10.13] customers and this is reinforcing the protocols in place so these things are known as autonomous [00:05:15.22] [00:05:15.22] systems and they interconnect using the border gateway protocol so bgp actually formalizes this [00:05:21.01] [00:05:21.01] arrangement is a mechanism for these different autonomous systems to share resources and [00:05:25.07] [00:05:25.07] interconnect with each other at places called ixps internet exchange boards the core point is they [00:05:30.06] [00:05:30.06] really imagine this network as a number of small entities working together to provide connectivity [00:05:34.21] [00:05:36.06] so from these beginnings and these uh sort of designs we come to the modern day which [00:05:42.02] [00:05:42.02] is nothing like this um so on the right you see a map of the top internet service providers by state [00:05:48.15] [00:05:48.15] and on the left you see a a table of that and what you see is it's basically [00:05:53.01] [00:05:53.01] a triopoly there are three main isps in the us xfinity which is comcast charter and atnt [00:06:00.12] [00:06:00.12] and that graph on the right actually is a little bit worse than it looks because mid-continent [00:06:04.08] [00:06:04.08] is actually comcast and time warner is actually charter [00:06:07.12] [00:06:08.23] now um you might say um you know okay this is this is you know not what the internet was designed to [00:06:15.12] [00:06:15.12] be but it's not that bad well actually if you take it a little bit farther and look at mobile which [00:06:20.15] [00:06:20.15] is now the dominant paradigm of internet access globally over half of internet traffic or access [00:06:27.16] [00:06:27.16] goes over the mobile network and even in the us most people use their cell phones to access the [00:06:31.11] [00:06:31.11] internet significantly more than they do desktops and clients so if you look at the mobile internet [00:06:36.02] [00:06:36.02] it's actually much much worse so uh here in the us we have three carriers primarily there is a little [00:06:41.14] [00:06:41.14] bit of a long tail notably gci wireless in my home state of alaska but you see verizon 18 teemo being [00:06:48.06] [00:06:48.06] the big three in the philippines where we're going to be talking about a project there are just two [00:06:52.02] [00:06:52.02] in china the largest uh market in the world there are three and they're all owned by the [00:06:56.17] [00:06:56.17] chinese government so really this triopoli duopoly model repeats itself and it's no [00:07:01.12] [00:07:01.12] longer this diverse ecosystem of a number of organizations working together by connectivity [00:07:06.06] [00:07:07.11] okay you might say well it's not what the internet was designed to do but it has connected a lot of [00:07:11.12] [00:07:11.12] people and continues to do so and this is true but it is leaving people out um so a little bit of uh [00:07:18.21] [00:07:18.21] sort of the trajectory of internet growth at the moment while a lot of people are online and you [00:07:22.19] [00:07:22.19] see this this red line here uh is the growth of internet adoption up to 4.1 uh sorry 4.13 [00:07:30.00] [00:07:30.00] billion people at the moment that blue line on the bottom is the rate of change of this and what you [00:07:36.04] [00:07:36.04] see is that the rate of change of internet growth is actually shrinking pretty rapidly going from [00:07:40.19] [00:07:40.19] 7.5 growth in 2016 to 3.5 percent in 2021 um and so this triopoly duopoly model has connected a [00:07:49.14] [00:07:49.14] lot of people but is diminishing its returns and connecting people and so okay uh now what uh why [00:07:56.00] [00:07:56.00] might this be the case so the gsma this is the gsm association it is the trade organization for the [00:08:03.14] [00:08:03.14] cellular networks provided this wonderful report in 2016 with following quote in most countries [00:08:09.11] [00:08:09.11] even in africa mobile operators have already rolled out 2g and 3g network coverage as far as [00:08:14.19] [00:08:14.19] possible within the envelope of a commercially sustainable business model that commercially [00:08:19.09] [00:08:19.09] sustainable business model to be clear is that duopoly tripoli model we've been talking about [00:08:23.11] [00:08:24.23] so um what does this mean well what this means is if you're one of the 800 million people without [00:08:31.05] [00:08:31.05] any access at all this being um uh 2g networks 3g networks internet networks any of that that growth [00:08:37.12] [00:08:37.12] has stopped the telecoms are not rolling out new equipment to you uh it's uh it's going somewhere [00:08:42.19] [00:08:42.19] else for those 3.5 billion people without broadband access that growth is slowing this kind [00:08:48.08] [00:08:48.08] of makes sense if you think about uh how a mobile network works uh and how telecoms work right they [00:08:54.06] [00:08:54.06] roll out 4g as they as the markets are are ready for them um and connect people through that but [00:09:00.17] [00:09:00.17] at some point they start to reach semi-rural or sorry semi-urban semi-rural markets where [00:09:05.14] [00:09:05.14] the revenues are lower and it makes more sense to instead go to the next generation of equipment [00:09:09.22] [00:09:09.22] and deploy maybe 5g equipment in the city where that revenue is going to be higher [00:09:13.11] [00:09:14.00] and so essentially these people are just not going to be getting internet access and are going [00:09:18.12] [00:09:18.12] to be left out of what is increasingly crucial infrastructure now we as well as a large pool of [00:09:24.15] [00:09:24.15] other researchers have been working on solving this space through attacking the business model [00:09:28.23] [00:09:28.23] angle specifically we've been working on a set of solutions known as community networks [00:09:33.14] [00:09:33.14] now community networks are networks built owned and operated by users in a participatory and open [00:09:39.16] [00:09:39.16] manner by being built by local people they're able to be optimized for local needs leverage [00:09:44.13] [00:09:44.13] local resources making it less expensive and provide local services to meet local needs at [00:09:49.12] [00:09:49.12] the end of the day community networks are really designed to retake that distributed nature of the [00:09:54.23] [00:09:54.23] internet and provide better more affordable more local networks um a large pool of small community [00:10:00.21] [00:10:00.21] networks working together connecting the xps is really the vision of what's going on there and um [00:10:06.23] [00:10:06.23] there are numerous numerous examples of successful community networks throughout the world uh the big [00:10:11.05] [00:10:11.05] one is wifi.net this is in catalonia and has 35 000 nodes by far the gorilla of community networks [00:10:18.02] [00:10:18.02] uh but there are other examples all over the place nyc mesh is the most notable one in the u.s [00:10:22.10] [00:10:22.10] um now that's in uh new york uh mostly brooklyn and uh manhattan uh fry funk in germany ultimately [00:10:29.03] [00:10:29.03] in argentina so on and so forth so lots of people working in the space lots of people [00:10:33.20] [00:10:34.10] building fun stuff our particular chunk of this is an area known as community cellular networks [00:10:40.19] [00:10:41.09] now these are taking those community network ideas and saying that what we really want to use [00:10:46.02] [00:10:46.02] is cellular protocols for those because they're well optimized specifically for rural connectivity [00:10:50.15] [00:10:52.06] cell phone networks provide wide area connectivity and they support handsets like mobile phones as a [00:10:56.17] [00:10:56.17] tier one client instead of expecting people to have laptops which as mentioned is the growth [00:11:00.23] [00:11:00.23] area in the future so this is a rough background and kind of the area and space that we're working [00:11:07.07] [00:11:07.07] the first project i want to talk about is community seller manager [00:11:10.15] [00:11:11.07] so community seller manager is an attempt to bridge these two worlds what we really want to do [00:11:16.21] [00:11:16.21] is combine the scale and operations of telcos and their ability to get connectivity out to everyone [00:11:23.07] [00:11:23.07] with the reach and customization and localization of community networks um so how do we do this well [00:11:31.09] [00:11:31.09] first to slow down a little bit we have to talk about and i know this isn't a networking talk but [00:11:35.18] [00:11:35.18] we're gonna have a little bit in here of the network the the structure of a mobile network [00:11:41.20] [00:11:42.17] so this is our diagram of a mobile network uh you can see a bunch of radios and cell phones [00:11:48.02] [00:11:48.02] around them talking to what i have labeled here as a mobile network core so a traditional cell phone [00:11:54.02] [00:11:54.02] network architecture the the radios known as the radio access network or the ram are really simple [00:12:00.17] [00:12:00.17] devices uh all that they do is largely forward traffic to the mobile network core where all of [00:12:06.06] [00:12:06.06] the cellular network logic happens things like authentication things like roaming things like [00:12:12.12] [00:12:12.12] routing of your voice call to the appropriate peer network uh things like billing all of that happens [00:12:18.13] [00:12:18.13] all of the logic in that mobile network core that mobile network core is usually housed at [00:12:23.14] [00:12:23.14] a data center somewhere highly connected uh with a robust power supply and these kind of things [00:12:30.06] [00:12:31.14] so this architecture which is repeated from 2g to 3g to 4g and even up to 5g networks [00:12:36.21] [00:12:38.02] has a number of limitations that make it really hard for telecoms to provide rural connectivity [00:12:42.19] [00:12:44.13] first there's very little flexibility across sites in this design because the radio access [00:12:50.13] [00:12:50.13] layer is so simple and does such a small amount of the logic uh that there's very little room for [00:12:56.12] [00:12:56.12] customization and optimization uh in that layer instead everything has to happen at the mobile [00:13:02.04] [00:13:02.04] core and so uh any kind of uh extensions that you make are similarly available uh sort of everywhere [00:13:09.03] [00:13:10.21] second this architecture doesn't degrade gracefully in a urban context um there's [00:13:16.12] [00:13:17.01] the the radio network will overlap with each other and what that means is if you lose a [00:13:20.21] [00:13:20.21] radio access piece cell phone tower crashes or whatever you're able to move over to another one [00:13:25.16] [00:13:26.12] and not necessarily see degraded service but in a rural area there's usually just one site [00:13:31.03] [00:13:31.03] so if that fails you lose all uh connectivity and moreover in these rural sites the [00:13:36.10] [00:13:38.00] opportunities for failure are significantly higher things like power being extremely rough [00:13:42.13] [00:13:42.13] in remote rural areas things like your back hauler connection to the mobile network or often going [00:13:47.18] [00:13:47.18] over a satellite link which is affected by weather instead of a fiber link as it [00:13:52.02] [00:13:52.02] would be in an urban area and lastly telcos are just fundamentally organizationally centralized [00:13:59.01] [00:13:59.14] we talk about this mobile core network but the rest of the business structures are happening [00:14:03.05] [00:14:03.05] there all the the spectrum holdings are centralized all these elements of what [00:14:07.07] [00:14:07.07] really defines a telecom are similarly just as centralized as the mobile core network itself [00:14:11.20] [00:14:13.01] so the goal of ccm is to come in and build a new system which decentralizes these network [00:14:18.19] [00:14:18.19] services enabling customization at the edge while centralizing the interoperation so still letting [00:14:24.23] [00:14:24.23] it be as telco like in its centralization interaction with the rest of the of the world [00:14:30.12] [00:14:32.08] okay so brief aside um to be clear community seller manager is a 2g [00:14:37.09] [00:14:37.09] cellular technology um and this is just because of the communities that we were working with [00:14:41.12] [00:14:43.12] we did some light studies at the time and found that most uh devices uh there's a [00:14:48.21] [00:14:48.21] plurality of devices support only 2g and so we needed to deploy that at the time [00:14:53.05] [00:14:54.13] so our solution uh community seller manager is a re-architecture of this [00:14:59.07] [00:14:59.22] mobile network core the idea here is is twofold first we have a new piece of software known as [00:15:05.20] [00:15:05.20] the ccn client that runs with the radios so the ccm client is basically a micro [00:15:12.19] [00:15:13.20] mobile core network allowing for all the kind of functionality and customization that you can do at [00:15:18.10] [00:15:18.10] a core network at the edge using open source uh core network technologies now these then [00:15:23.11] [00:15:23.11] connect to a thing called the ccm controller which is a cloud or network appliance deployed [00:15:28.19] [00:15:28.19] either with or within a telco that manages the telco resources and distributes them out to [00:15:34.06] [00:15:34.06] the clients as necessary so for example phone numbers are as our resource that telcos hold [00:15:40.04] [00:15:40.04] so in the ccm architecture what happens is your phone registers on the ccm client that sends a [00:15:45.18] [00:15:45.18] message up to the ccm controller which pulls a phone number out of the pool of numbers that [00:15:50.02] [00:15:50.02] telco has and hands it down to the user similarly things like spectrum can be uh distributed among [00:15:57.18] [00:15:57.18] the access points and uh access points can be turned off for things like licensing regimes and [00:16:03.09] [00:16:03.09] power licenses and the like now this architecture has a number of advantages the big core goal of [00:16:08.08] [00:16:08.08] it really is to provide as much localization as possible so as an example of this what we can do [00:16:14.17] [00:16:14.17] is actually have calls that route entirely locally within the community in a traditional architecture [00:16:19.16] [00:16:19.16] everything has to go back to the core network but here because the core network is local calls can [00:16:24.02] [00:16:24.02] happen entirely locally and this means even if the network is disconnected we can handle those calls [00:16:29.14] [00:16:29.14] now there's a lot more here this was a full sort of systems uh building exercise and i'll hand wave [00:16:36.08] [00:16:36.08] the rest so go take a look at the paper if you're interested in more of the the nitty-gritties there [00:16:40.17] [00:16:42.02] but the next thing i want to talk about is the deployment and evaluation of this [00:16:45.03] [00:16:45.03] a key element of the kind of work that my lab tries to do is not just build artifacts uh and and [00:16:51.12] [00:16:51.12] test them in a lab and move on to the next one we actually try to get them into the [00:16:54.12] [00:16:54.12] real world and this is one of the largest uh deployments that we put together uh so over [00:16:59.09] [00:16:59.09] three years we deployed 17 sites covering about 3 000 people with a telco partner in the philippines [00:17:04.17] [00:17:04.17] this is globe telecom there as well as a partner at the university of philippines who has [00:17:09.05] [00:17:09.05] drove much of this i want to give as much credit as i possibly can so of these sites the university [00:17:15.01] [00:17:15.01] of the philippines deployed the six sites up in the north in what are known as the aurora province [00:17:20.00] [00:17:21.05] of luzon and then globe telecom deployed uh 11 sites elsewhere uh throughout the country now [00:17:27.07] [00:17:27.07] a key distinction between these two was the level of community networking-ness that we were able to [00:17:32.17] [00:17:32.17] put together the university of this philippine sites were understood to be research sites and [00:17:37.07] [00:17:37.07] so we were able to come in and build new and novel business structures around them the community was [00:17:42.02] [00:17:42.02] deeply involved the installation the service generation uh the local university helped do [00:17:47.16] [00:17:47.16] maintenance really as we built them from the ground up to be uh proper community networks [00:17:52.12] [00:17:52.12] the global sites were a little bit less than that global business processes didn't really [00:17:56.12] [00:17:57.05] mesh with that very well and so instead they handed it over to an ngo there were [00:18:01.12] [00:18:01.12] still elements of community uh networking going on things like rev shares into local community [00:18:06.13] [00:18:06.13] and things like this and i want to be clear this was not a toy deployment so here is a trial from [00:18:13.12] [00:18:13.12] the installation it's about a year and a half where you see the overall uh monthly active users [00:18:18.15] [00:18:18.15] in the network and so you can see the original test globe sites going up and then picking up [00:18:24.15] [00:18:24.15] and then the up sites kick on and we're ready to go what we see here is 50 000 hours of calls [00:18:30.08] [00:18:30.08] and over 3 million sms over the test period so lots of usage lots of users real real deployment [00:18:36.21] [00:18:38.10] i would say largely i would consider the technical elements of this study to [00:18:42.15] [00:18:42.15] be a success we were absolutely able to enable cooperation between community cellular networks [00:18:47.18] [00:18:47.18] and telecoms we were able to provide autonomy for community cellular networks those up sites [00:18:52.10] [00:18:52.10] were very highly customized for the local community and and sustainable in that context [00:18:57.11] [00:18:57.11] um they were able to degrade gracefully in the face of failures the technology supported that [00:19:01.22] [00:19:01.22] we have these wonderful stories of um a super typhoon coming in which hit the aurora province [00:19:08.02] [00:19:08.02] regularly and the community actually able to tear down the equipment put it into storage [00:19:12.15] [00:19:12.15] after the typhoon passes come out put it back up and provide connectivity from there and so [00:19:18.12] [00:19:18.12] things like the satellite link being down for some period beyond that was was acceptable and workable [00:19:23.09] [00:19:23.09] and lastly this was the largest community seller network deployment to date [00:19:26.23] [00:19:28.19] another thing that we do if this style of research in our lab is sort of follow the core technical [00:19:35.09] [00:19:35.09] intervention with a number of related studies surrounding it i want to call out a few of those [00:19:39.09] [00:19:40.06] the first is a paper at kai 2018 which explored how to crowd source repair information from [00:19:45.22] [00:19:45.22] community members the idea of this was basically we could send a broadcast sms [00:19:50.08] [00:19:50.08] to everyone the network saying hey does anyone have a wrench does anyone know how to do basic [00:19:54.23] [00:19:54.23] electrical work these kind of things and see who came and what kind of capacity they actually had [00:19:59.20] [00:19:59.20] now this was a surprisingly successful study i'll say the capacity absolutely existed in [00:20:04.00] [00:20:04.00] the community but people were very reluctant to engage in the active repair because the [00:20:09.05] [00:20:09.05] equipment was not necessarily viewed as highly local this was at one of the globe sites i wish [00:20:13.22] [00:20:13.22] we had been able to repeat the study in one of the community more community oriented sites to [00:20:17.14] [00:20:17.14] see if we could have gotten farther at cscw in 2019 esther also went out and explored the [00:20:24.02] [00:20:24.02] repair ecosystem in the codes of aurora this was a really interesting wrinkle in the sense that [00:20:28.15] [00:20:31.03] a number of studies that we had seen elsewhere had explored repair and remote rural areas [00:20:36.12] [00:20:36.12] and they had found really robust repair ecosystems um but in the coves of aurora we [00:20:41.22] [00:20:41.22] found nothing of this part instead we found just piles of decrepit electrical equipment these two [00:20:47.01] [00:20:47.01] televisions are a great example of it and so it was a real counterpoint to that and we think it [00:20:54.13] [00:20:54.13] was really just how far away it was because you needed to take boats to get to the coves rather [00:20:59.20] [00:20:59.20] than road infrastructure and so a lot of the repair infrastructure couldn't make it all the way [00:21:03.09] [00:21:03.09] out there but there were some successful examples largely top-down repair ecosystems that being the [00:21:09.18] [00:21:10.10] local grid repair system and we thought maybe we could replicate those to create a better repair [00:21:14.21] [00:21:14.21] infrastructure in the codes similarly there's a little bit of data science work i mentioned uh [00:21:21.07] [00:21:21.07] prior a study on phone adoption showing that smartphone growth was important and would be [00:21:26.06] [00:21:26.06] viable in 2020 leading to our future work and also this whole deployment the up sites were part [00:21:31.18] [00:21:31.18] of a long-term randomized control trial of mobile network impact and so we had a paper at ictd 2020 [00:21:37.09] [00:21:38.00] on what kind of features given a social network analysis were most important of [00:21:42.17] [00:21:42.17] adoption and it turned out to be income which is not terribly surprising but the other [00:21:46.17] [00:21:46.17] features were surprisingly less important and so that was an interesting set of conclusions [00:21:52.00] [00:21:53.09] so okay where are we now well unfortunately um global actually pulled the spectrum license in [00:22:01.05] [00:22:01.05] 2019. uh essentially they were unable to operate the cloud software themselves but [00:22:05.05] [00:22:05.05] unwilling to farm it out because it was holding a sensitive state [00:22:08.21] [00:22:10.17] there were hardware sourcing vendors not really globe or anyone else's fault uh with a vendor [00:22:14.23] [00:22:14.23] uh at the end of the day and i don't want to uh signal this is globe's fault like they [00:22:19.14] [00:22:19.14] were i think being very very aggressive and engaging in this kind of research activity [00:22:23.20] [00:22:24.13] and so uh kudos to them for for for engaging at that depth um now we were able to convert [00:22:31.07] [00:22:31.07] any of the sustainable sites which were sites that were receiving enough revenue to pay for [00:22:35.01] [00:22:35.01] themselves over to 802.11. um unsustainable sites or largely correlated with existing coverage [00:22:41.11] [00:22:41.11] or growth of coverage so that the community network was competing with an incumbent mno [00:22:45.20] [00:22:46.12] the wi-fi is back called using the same vsats deployed during the project now this is [00:22:50.13] [00:22:50.13] unfortunate in that many users are now left out specifically those feature phone users we were [00:22:54.12] [00:22:54.12] talking to are talking about uh but we do provide general internet access which has its own benefits [00:22:59.09] [00:23:00.15] so after this study and after this reality we came back and we sort of thought about what [00:23:04.15] [00:23:04.15] we were doing and why we were doing how did we wanna uh to build things in the future to avoid [00:23:10.04] [00:23:10.04] these kind of uh issues and first just kind of thinking of the limitations of working with [00:23:16.04] [00:23:16.04] a telco or known as a mobile network operator the other day the u p sites the community sites [00:23:22.10] [00:23:23.05] were significantly more sustainable indeed many of the globe sites were simply non-starters [00:23:28.00] [00:23:28.00] there are protocol issues here but really at the end of the day we were forced to partner [00:23:32.15] [00:23:32.15] we had sort of our arms twisted because all of the 2g spectrum in the philippines was licensed [00:23:36.19] [00:23:37.09] and so when we came into the philippines and we wanted to deploy a network we had to work with [00:23:40.23] [00:23:40.23] a telecom and again globe did a fantastic job of being a big m o trying to engage on these issues [00:23:46.00] [00:23:46.00] but it's hard these are hard problems and it's fundamentally just not in their business model [00:23:50.06] [00:23:51.05] uh concurrent to this we started realizing the issues with gsm i mean smartphone adoption was [00:23:55.11] [00:23:55.11] growing there was a big desire for internet connectivity and phone numbers as a costs are [00:23:59.09] [00:23:59.09] a huge problem globe was able to give them away for free but in our deployment in indonesia it was [00:24:03.20] [00:24:03.20] literally the cost of our backhaul half of our office numbers and so these two things coupled [00:24:09.12] [00:24:09.12] together really got us thinking how could we build instead an independent community lte network let's [00:24:15.16] [00:24:15.16] get to our internet roots let's try to build out internet infrastructure and you might ask well why [00:24:21.12] [00:24:21.12] haven't people done this right like why is this why is this hard and it's hard for a whole suite [00:24:26.08] [00:24:26.08] of reasons gsm and spectrum and all these things like the the lte model changes some things but [00:24:32.08] [00:24:32.08] not everything but fortunately there's really big industry movement in the direction that we want to [00:24:37.09] [00:24:37.09] go so for instance on spectrum while 2g spectrum is entirely licensed out in most markets lte is [00:24:45.22] [00:24:45.22] so hungry for a new spectrum that we're starting to see new and unlicensed or shared licensed bands [00:24:51.03] [00:24:51.03] uh so this is lte unlicensed and lte license assisted access these are protocols for allowing [00:24:56.17] [00:24:56.17] cellular equipment to actually operate in wi-fi bands um similarly there's a brand new regime [00:25:01.18] [00:25:01.18] from the fcc that went live in april 2020 called the citizens broadband radio service i'm going [00:25:06.06] [00:25:06.06] to give a lot more detail on that soon but it's this really neat multi-tiered licensing regime [00:25:10.12] [00:25:11.14] where okay i just realized i have a chat box so if people want to put things in the chat box i [00:25:17.14] [00:25:17.14] will look at it uh multi-tiered uh radio access network and so the idea of this is basically that [00:25:24.04] [00:25:24.04] someone can pay a lot of money and get priority access but if you don't pay money you still get [00:25:27.20] [00:25:27.20] sort of secondary access and if they don't if someone's not using it you need to take it [00:25:32.08] [00:25:35.11] the next piece of this is hardware lte access points are actually significantly cheaper [00:25:39.16] [00:25:39.16] than 2g access points it's crazy and this whole ecosystem is rapidly commoditizing there's things [00:25:45.01] [00:25:45.01] like openran which let smaller manufacturers participate and especially as we move to 5g [00:25:49.16] [00:25:49.16] you're entering small cells which are inherently commoditized small scale and produced at large [00:25:56.06] [00:25:58.13] the software ecosystem here is significantly better there are two open source gsm stacks both [00:26:03.05] [00:26:03.05] of which were pretty hairy while there's multiple s lte implementations and indeed even elements [00:26:10.08] [00:26:10.08] of 5g are allowing for the kind of designs that we have network function virtualization [00:26:15.20] [00:26:15.20] specifically is focused on pushing services out to edge and building private networks which we'll [00:26:21.16] [00:26:21.16] talk about here next all right after this one uh interconnect as mentioned the ability for phone [00:26:27.22] [00:26:27.22] numbers generating those is a really difficult expensive problem but when you're in an uh an ip [00:26:32.15] [00:26:32.15] network uh which lte and 5g both are you get to use over the top services as your default [00:26:37.12] [00:26:37.12] so don't provide calls you can just do whatsapp and let them handle everything or let uh google [00:26:44.02] [00:26:44.02] handle the the oauth the authentication into the network all these kind of things and lastly i [00:26:50.02] [00:26:50.02] think most importantly this technology is being driven towards smaller and smaller networks [00:26:56.04] [00:26:56.04] wireless isps in the us have been using lte for a decade now to provide wide area connectivity [00:27:01.12] [00:27:02.02] there's a big motion towards a thing called private lte the idea of this is essentially let's [00:27:07.07] [00:27:07.07] say you run a factory and you want to automate your lifters or something else you don't want to [00:27:13.05] [00:27:13.05] send that over the public internet because there's too much variation in the quality that you receive [00:27:19.05] [00:27:19.05] so instead what you do is you build your own lte network uh with very specific properties that you [00:27:25.22] [00:27:25.22] want and then you're able to run this so there's a big motion towards providing small organizations [00:27:30.13] [00:27:30.13] companies buildings uh the ability to run their own networks and this dovetails into what is known [00:27:36.00] [00:27:36.00] as multi-operator core network uh which is where you put in an lte network and then you lease it [00:27:41.03] [00:27:41.03] out to multiple uh other carriers for them to provide access let's say you're running a um an [00:27:47.14] [00:27:47.14] apartment building you install this and you take a cut of everyone in your apartment building finally [00:27:52.00] [00:27:52.00] having their own internet access at the end of the day the reality here is that um there is a great [00:27:58.10] [00:27:58.10] opportunity for lte community networks right now and so we're going to pick that up and run with it [00:28:01.20] [00:28:02.19] and so the next piece of this talk is to talk through uh our designs for community lte thing we [00:28:09.07] [00:28:09.07] call a culti and so what we did is we really took three big planks um and and and built around those [00:28:15.20] [00:28:16.17] the first was trying to embrace being the neutral pipe so in classic satellite as mentioned you [00:28:22.12] [00:28:22.12] provide all of these other services voice sms ussd you provide phone numbers roaming and you charge [00:28:27.05] [00:28:27.05] everyone for everything but we just want to build the internet we want to connect people in the way [00:28:33.11] [00:28:33.11] the internet connects and so we disregard almost everything provide none of that the only wrinkle [00:28:38.10] [00:28:38.10] here is emergency services which i'll talk very briefly about uh later so that's design principle [00:28:43.07] [00:28:43.07] one design principle two is embracing the edge and as mentioned in a classic cell at architecture you [00:28:48.12] [00:28:48.12] send everything to your mobile core network and then those mobile core networks connect [00:28:52.12] [00:28:52.12] to each other using big money roaming agreements with each other now in colty we're going to move [00:28:58.06] [00:28:58.06] exactly like we did in the 2g network everything to the edge co-locate it with the ram and provide [00:29:02.19] [00:29:02.19] services locally again this is growing support for this through in 5g new radio and then we can [00:29:10.00] [00:29:10.00] interconnect using those protocols designed in the early days of the internet things like bgp [00:29:14.13] [00:29:14.13] let's say the early days mid days of the internet so that we can interconnect as peers there lastly [00:29:21.07] [00:29:21.07] we want to embrace cellular security this is a bit of an aside but cellular has wonderful [00:29:26.17] [00:29:26.17] links layers security primitives um you know the network both validates you and you validate [00:29:31.11] [00:29:31.11] your network through a physical token that being your sim card which has a physical key on it um i [00:29:38.04] [00:29:38.04] mean it's digital but you know what i mean it's a physical device and so this is great this means we [00:29:43.01] [00:29:43.01] can build networks that don't use passwords or mac addresses because we can uniquely identify [00:29:47.05] [00:29:47.05] users based on uh their sim card so again we built it um on the top you see traditional traditional [00:29:54.02] [00:29:54.02] core network architecture uh the access points in 5g are known as 4g are known as enode b's and [00:30:00.13] [00:30:00.13] the core network is known as an epc i'm going to try to avoid that terminology but you might [00:30:04.08] [00:30:04.08] see me slipping into it on accident so here you can see one of the access points talking to the [00:30:08.06] [00:30:08.06] core network over a satellite link and the rest nod in community lte we move those core networks [00:30:13.20] [00:30:13.20] co-locating them with the communities and then we back haul there from the satellite link moving a [00:30:18.04] [00:30:18.04] bunch of traffic over making life easier so yeah the um the core network in a data center versus [00:30:25.20] [00:30:25.20] being a co-located with the ram um thousands of access points per core network versus just a few [00:30:31.11] [00:30:31.11] access points per core network um and then operations far from the community and some [00:30:35.16] [00:30:35.16] data center or the entire network within the community really a lot of those 2g ideas taken [00:30:40.00] [00:30:40.00] up but trying to remove the telco partner and trying to see if we can move forward like that [00:30:45.09] [00:30:46.23] so we have implemented this hardening a system known as open5ds which is our favorite [00:30:51.18] [00:30:51.18] uh open source 5g and 4g network core a whole bunch of engineering happened again [00:30:57.09] [00:30:57.09] feel free to read the mobicom paper if you'd like to dive into the engineering efforts [00:31:00.13] [00:31:01.18] and then we deploy and evaluate so this was deployed in the highlands of papua this is [00:31:06.21] [00:31:06.21] actually where i did my grad work and so that speaks a little bit to them liking the research [00:31:11.05] [00:31:11.05] that we do and that we actually help people and that they had us come back five years later to [00:31:15.07] [00:31:15.07] install a new system in deploying it in indonesia you need to customize it for indonesia first that [00:31:21.18] [00:31:21.18] was making it look like an indonesian telecom meaning you have prepaid pay as you go bundles [00:31:26.12] [00:31:27.07] a small number of uh package sizes and then an indonesian language [00:31:31.18] [00:31:31.18] landing page so that people can use it credits were sold in small shops as is done in indonesia [00:31:38.00] [00:31:39.14] we deployed this in mid 2018 using a staged rollout we saw a bunch of really interesting [00:31:44.23] [00:31:44.23] technical foibles here just uh handsets that do not fit the the protocol and not do their job [00:31:50.15] [00:31:50.15] but that's life and in general again on the right just to say this is not a toy we see [00:31:55.05] [00:31:55.20] about 200 registered users and a unique monthly of between 80 and 100 most of the time [00:32:02.08] [00:32:05.01] so total cost of this system and this is just a single spot right is about ten [00:32:08.17] [00:32:08.17] thousand dollars with the monthly cost being 391 and uh the most of that being back home [00:32:14.02] [00:32:16.15] again uh we do a bunch of we like we deploy this artifact and we do this technical intervention [00:32:22.08] [00:32:22.08] then we surround it with a related work um and so the first of these was a paper at hot nets [00:32:27.12] [00:32:27.12] which was arguing for a slightly different architecture here uh really going to lead [00:32:31.18] [00:32:31.18] to phase three of the project saying there's a lot of room to be really clever here and get [00:32:35.12] [00:32:35.12] get weird with it and then matt johnson also put together a paper coming up at ccw this next cycle [00:32:42.04] [00:32:42.04] on trying to do congestion control as a community project and a process and so he's building some [00:32:47.14] [00:32:47.14] tools right now for the community uh it's a different lte network it's one of mexico to [00:32:52.08] [00:32:52.08] engage in congestion control and traffic prioritization all these kind of elements [00:32:56.00] [00:32:56.00] that are normally automated and making them into a community-based process we also have some [00:33:01.09] [00:33:01.09] deployment papers so matt has a paper at www this year on the usage patterns of the network that we [00:33:09.05] [00:33:09.05] just talked about the big focus or output here is that there are a couple anchor tenants who's [00:33:13.11] [00:33:13.11] really paid for the network and this is different than the prior literature which really argues for [00:33:18.02] [00:33:18.02] rpu average revenue per user as the primary metric average revenue is not great when you just have [00:33:22.19] [00:33:22.19] three people who spend you know 70 of the credits uh and lastly spencer a postdoc in my lab and now [00:33:28.04] [00:33:28.04] research scientist in my lab has been building these networks and building a bunch of tooling [00:33:32.04] [00:33:32.04] and support for getting them out and letting other people repeat these models so one of these [00:33:36.10] [00:33:36.10] was a book chapter detailing the manual of sort of setting up your own lte network if you wanted to [00:33:40.23] [00:33:42.10] current status of this network much better this network remains operational it's functionally a [00:33:47.01] [00:33:47.01] canary testbed for much of our core network innovations these technical innovations are [00:33:51.03] [00:33:51.03] fantastic but mostly i think the really exciting thing is it is super sustainable from both a [00:33:55.09] [00:33:55.09] financial and social perspective this thing is making a ton of money for the operators which is [00:34:00.06] [00:34:00.06] a primary school up there and it's maintainable we do of course cause technical issues and [00:34:04.23] [00:34:04.23] then fix technical issues but these are all resolvable at distance um and so that the network [00:34:09.12] [00:34:10.06] subsists with with sustains with with no issue i think the biggest thing here is [00:34:15.09] [00:34:15.09] that this is actively competing with and beating a nationwide mmo in the area we installed this [00:34:20.06] [00:34:20.06] before the telecom came to town but they came to town installed a big tower big satellite [00:34:25.05] [00:34:25.05] link all this wonderful stuff and it's not as performant as our network so uh people still [00:34:30.17] [00:34:30.17] use our network despite there being an mno in place uh i think it's all in all a great success [00:34:35.07] [00:34:36.08] but despite this great success there are still limitations and so as we built this network we [00:34:42.00] [00:34:42.00] started talking about like how do we scale how do we repeat this how do we do this and [00:34:44.19] [00:34:44.19] you can just see the limitations essentially this little island of coverage just has limited utility [00:34:50.00] [00:34:50.15] the big value add for going to the mnos going to the big operator the um the carrier the telcom in [00:34:56.10] [00:34:56.10] the country is that you get to roam around you get to go to jayapura you get to go to [00:35:01.16] [00:35:01.16] jakarta and use the same sim card and these little islands of connectivity aren't going to do that [00:35:05.22] [00:35:06.23] and essentially scaling these networks is impossible [00:35:10.08] [00:35:11.14] because what you actually need to scale is your partner organization [00:35:14.17] [00:35:15.11] and you have to make them into telecoms like this primary school is not going to be a primary [00:35:21.07] [00:35:21.07] a telecom as their their business their job is to provide education uh to community members [00:35:26.17] [00:35:27.22] and so uh they're you know they're just small and local and so the next project is trying to [00:35:32.17] [00:35:32.17] resolve this what we really want to do is let a whole bunch of these small community networks [00:35:37.14] [00:35:37.14] exist and then have them cooperate to provide coverage and build an mno out of them rather than [00:35:43.03] [00:35:43.03] expecting them to become an mno and that's the third part of the talk cooperative cell either [00:35:48.08] [00:35:50.00] now the idea for cooperative seller is exactly this what we're going to do is allow multiple [00:35:54.10] [00:35:54.10] small community lte or new radio networks peer together in exactly the way i mean not [00:35:59.18] [00:35:59.18] technically exactly where but in logically or say philosophically the same way that we let [00:36:05.03] [00:36:05.03] um that the internet was built and those those networks appeared together to provide wide area [00:36:10.10] [00:36:10.10] access to be one network where everyone peers together essentially you can think of it as you [00:36:15.20] [00:36:15.20] you take some equipment uh you put it on your roof you configure it to join this collective [00:36:21.18] [00:36:21.18] and then you get and give service like you do with everyone else now technically what we do is we use [00:36:27.11] [00:36:27.11] a distributed storage layer to share critical state between the individual network cores again [00:36:34.00] [00:36:34.00] co-located with the radios some examples of these that i'll talk through are authentication [00:36:38.15] [00:36:38.15] billing and neighbor tables for handover and then we leverage some other advances [00:36:42.02] [00:36:43.01] that are ongoing in order to make this work so this is a diagram of what we're trying to do [00:36:47.11] [00:36:48.12] what you can see on the top is a traditional mno where there's a large number of radios a [00:36:53.12] [00:36:53.12] small number of core networks and they talk to each other over something now in the cooperative [00:36:57.22] [00:36:57.22] cellular model we again have a small number of radios a large per core a large number of core [00:37:03.16] [00:37:03.16] networks and now they talk and they share state over this distributed ledger solution the output [00:37:09.07] [00:37:09.07] of this the critical thing to understand is that what we do is we now create this cooperative [00:37:13.12] [00:37:13.12] coverage so in a traditional telco what you do is have multiple networks overlapping uh each [00:37:18.06] [00:37:18.06] competing with each other to provide connectivity uh and and going from there now in the cooperative [00:37:23.16] [00:37:23.16] model what we do is we have all of these networks all these independent organizations work together [00:37:28.15] [00:37:28.15] and let a user move around them so it looks like one big telco so first to talk through our [00:37:35.14] [00:37:35.14] distributed authentication in classic cellular your authentication is generated in a thing [00:37:40.21] [00:37:40.21] called your home hss this is the home subscriber service sorry again for the terminology we're just [00:37:45.16] [00:37:45.16] in cellular it's going to be there now users have to talk to this home agency when you authenticate [00:37:51.05] [00:37:51.05] to the network you have to talk to this hss because it holds the other side of your key [00:37:55.20] [00:37:55.20] now what you'll usually do is if you're roaming you'll go over ipsec and talk to this particular [00:38:00.04] [00:38:00.04] thing out there and connect potentially over long distance in cooperative cellular we still [00:38:05.14] [00:38:05.14] have this limitation you still have a home hss this is still the thing that makes your sim card [00:38:10.10] [00:38:10.10] but through some really rough abuse of the lte protocol we now figured out that what you can do [00:38:16.19] [00:38:16.19] is pre-generate what are known as authentication vectors now these are basically pre-computed [00:38:22.23] [00:38:22.23] calculations from the hss that are supposed to be used for caching so you cache them in the radio [00:38:29.09] [00:38:29.09] and this allows a phone to re-authenticate really rapidly but if instead we cash them on the ledger [00:38:34.19] [00:38:34.19] that means anyone on the ledger can authenticate the phone and the phone can similarly authenticate [00:38:40.08] [00:38:40.08] that their home hss and so this allows effectively roaming you're allowed to move to any other [00:38:46.17] [00:38:46.17] network as long as you've pre-generated this key this stuff this vector and put it on the ledger [00:38:51.18] [00:38:53.01] now building we do basically in the same way in classic cellular building is provided by [00:38:57.01] [00:38:57.01] a thing called the policy and charging rules function again i apologize but this monitors [00:39:02.17] [00:39:02.17] user properties mostly how much money they have and modifies their rights in the network [00:39:06.10] [00:39:06.10] it touches everything it's actually one of the worst pieces of infrastructure i've seen [00:39:09.18] [00:39:10.08] you can imagine a system that allows some sms and not some sms and local calls and 911 calls but not [00:39:15.12] [00:39:16.04] international calls and like all this sort of uh difficult thing that they had to engineer [00:39:20.04] [00:39:21.05] now in cooperatively in cooperative cellular what we actually do is cryptographically [00:39:24.15] [00:39:24.15] couple the billing to the previously mentioned authentication vectors so now users buy bundles [00:39:29.22] [00:39:30.12] um when they go to their home hss and say i would like one gigabyte and so that one gigabyte is an [00:39:35.07] [00:39:35.07] authentication vector that says one gigabyte and so then whenever you join to a new network [00:39:40.08] [00:39:40.08] that network then spends down that bundle now these authentication vectors are one-time use [00:39:45.16] [00:39:45.16] so you do have to generate another bundle in the future but this works so great now [00:39:50.13] [00:39:50.13] we have authentication or roaming and now we have distributed billing so it works like this [00:39:55.22] [00:39:55.22] users connect only to their home network in a traditional mmo but in cooperative cell you're [00:39:59.20] [00:39:59.20] able to talk to any core network that is on the ledger and has your shared authentication vector [00:40:04.12] [00:40:05.14] lastly we have handover in a traditional cellular network a handover happens between [00:40:10.12] [00:40:10.12] different radios this is the process of you keeping a call going as you move between radios [00:40:14.15] [00:40:15.09] there's a protocol for this known as x2 and most notably they coordinate through what are known as [00:40:20.00] [00:40:20.00] neighbor tables this is your cell phone network knows the nearby radios and tells your cell [00:40:24.23] [00:40:24.23] phone about them so that it can look ahead and try to move over so we're implementing a novel [00:40:30.08] [00:40:30.08] quarter core handover mechanism that's the first tier and then we're going to [00:40:33.22] [00:40:33.22] share neighbor tables on the ledger the exact same way we're sharing authentication vectors [00:40:38.04] [00:40:38.04] in this way everyone will know about networks that are nearby that should be able to [00:40:42.06] [00:40:42.19] support this user and will also be able to hand over from between cores rather than between radio [00:40:48.10] [00:40:48.10] elements so here you're able to hand over between networks and a tradition only within your network [00:40:54.15] [00:40:54.15] in a traditional mmo but in cooperative cellular we can hand over networks now we've built this [00:41:00.02] [00:41:01.03] these graphs are really really early the big points here is more nodes is better [00:41:04.21] [00:41:05.18] and that we're not overloading the the ledger um through the generation of these [00:41:10.00] [00:41:10.00] authentication vectors but the core here is to say uh this works and we've done it [00:41:14.02] [00:41:16.04] now for spectrum we're using cbrs and i want to go into a little bit more depth in cbrs here because [00:41:20.04] [00:41:20.04] cbrs is really cool uh so cbrs is basically uh 150 megahertz of what used to be naval radar licenses [00:41:27.03] [00:41:27.16] um and so what they've done is they've opened it and the idea is that your [00:41:31.03] [00:41:31.22] radio comes up and talks to what is um a um a database in the sky and that database reports [00:41:38.23] [00:41:38.23] back and says hey here's some frequency you can use um and so what can happen or the way that [00:41:44.23] [00:41:44.23] it works is you can give money to be known as a higher tier customer you can kick other people off [00:41:50.02] [00:41:50.02] but if no one else is using that spectrum it's available to everyone and indeed 50 [00:41:54.06] [00:41:54.06] megahertz not five megahertz i've lost a zero somewhere of of that spectrum is always [00:41:58.23] [00:41:58.23] available for for unlicensed access you can't pay to get priority and this is known as gaa [00:42:03.05] [00:42:04.10] so we're actually using this google has given access to their sas and so this is going to be [00:42:08.23] [00:42:08.23] our spectrum solution going forward okay so now we have a solution do we really have a problem if you [00:42:14.13] [00:42:14.13] slow down and you like think through the logic of the story i put together this seems actually very [00:42:18.23] [00:42:18.23] poorly suited for rural connectivity we don't need handover we only have single installations [00:42:23.14] [00:42:23.14] uh blockchains can be very very chatty and we're operating over a low bandwidth links uh and so [00:42:29.20] [00:42:30.21] the story here is that esther uh who had been working in the philippines had was originally from [00:42:35.20] [00:42:35.20] new york and she came back uh from spending some time in new york where she had been volunteering [00:42:39.20] [00:42:39.20] for nyc mesh she came back and said you know this community seller stuff i think it's really really [00:42:44.12] [00:42:44.12] good fit for urban areas actually and i sort of hemmed and hot i'm from alaska i don't really work [00:42:49.01] [00:42:49.01] in urban areas um but she convinced me and so we went and we started looking into what connectivity [00:42:54.13] [00:42:54.13] was like in seattle now the city of seattle has a really nice report on this and they found that 95 [00:43:00.02] [00:43:00.17] of seattleites have access uh which seems really high right but then you realize like [00:43:05.14] [00:43:05.14] we're the ict research group that five percent is where we live these are the kind of people that we [00:43:09.18] [00:43:09.18] live to work for um and so we went in a little more depth and the study goes into a little bit [00:43:14.06] [00:43:14.21] we found exactly that it's like so if you live in poverty one quarter of the people who [00:43:19.03] [00:43:19.03] live in poverty in seattle do not have internet similarly living with a disability of being older [00:43:24.04] [00:43:24.17] english as a second language or non-english speaking all these people are [00:43:28.06] [00:43:28.06] are disproportionately unlikely to uh participate in the internet in seattle and so yes there's an [00:43:33.18] [00:43:33.18] opportunity here these are the kind of people we like to serve and so we will uh we're building the [00:43:38.17] [00:43:38.17] system for them now the reality of any of these kind of deployments is that you need partners [00:43:43.16] [00:43:43.16] and so we started finding local organizations that were really focused on digital equity [00:43:47.20] [00:43:47.20] people trying to maneuver or make it through digital financial barriers um things like insecure [00:43:54.04] [00:43:54.04] housing populations around here things like those sort of bodies of organizations and so we picked [00:44:00.00] [00:44:00.00] a number of partners um tacoma cooperative network is a cooperative network handling [00:44:04.17] [00:44:04.17] operating in tacoma a tacoma product library is a black brains research [00:44:07.20] [00:44:07.20] project which is an extremely large research endeavor going on in king county trying to [00:44:13.01] [00:44:14.08] build capacity among those communities the city of seattle has an internet for all initiative [00:44:18.13] [00:44:18.13] and our partner autheia networks who's uh somewhat local they're in oregon building neat distributed [00:44:24.04] [00:44:24.17] routing solutions and so here's the technology uh here's the left is an installation on the roof of [00:44:29.22] [00:44:29.22] the computer science building on the right is matt johnson that student doing a measurement [00:44:34.15] [00:44:34.15] and what you can see is at 2.1 kilometers we actually see significant we see broadband level [00:44:39.09] [00:44:39.09] access 60 megabits per second down eight megabits per second up at 2.1 kilometers using cbrs in an [00:44:45.14] [00:44:45.14] urban environment and indeed you can go and see a map this is the cbrs frequencies available in [00:44:52.21] [00:44:52.21] the greater puget sound we actually i think one of the worst parts in the country because there's [00:44:56.21] [00:44:56.21] significant naval holdings here and so the naval radar isn't always turned off but there's plenty [00:45:01.16] [00:45:01.16] of opportunity and at the bottom here what you can see is our installation in tacoma [00:45:05.22] [00:45:05.22] which was our first trial installation this is installed on a rooftop in the hilltop neighborhood [00:45:10.19] [00:45:10.19] providing connectivity to now less than 10 customers tests and 10 households [00:45:14.17] [00:45:15.09] but a very good canary for us in terms of running gear and and setting stuff up again [00:45:21.07] [00:45:21.07] there's a bunch of related technology research i'm running a little long so i'm going to go fast [00:45:25.07] [00:45:25.07] um training and technology capacity is really important here so we're [00:45:29.03] [00:45:29.03] working with detroit mesh's digital stewards pro program to put educational content [00:45:33.22] [00:45:33.22] around these community networks we're doing participatory design of in-network services [00:45:38.17] [00:45:38.17] um these are for example conflict resolution that avoids 9-1-1 this is very important to the dvr [00:45:43.20] [00:45:43.20] folks some technical work on wpac based content distribution networks similar participatory design [00:45:49.09] [00:45:49.09] of sensing and ml applications soil quality air quality noise collision detectors again [00:45:54.10] [00:45:54.10] trying to avoid 9-1-1 and then security in these shared network infrastructure environments this [00:45:59.09] [00:45:59.09] is largely a technically solved problem but something people worry about quite a bit [00:46:02.23] [00:46:02.23] um disinformation among communities is another interesting uh security angle and we sort of [00:46:07.09] [00:46:07.09] phrase that as just building a less racist next door and see if we can get a pretty ambitious [00:46:11.14] [00:46:11.14] topic i'll say um but you know we're here for that in conclusion um our group believes the [00:46:17.22] [00:46:17.22] internet specifically access networks should be more distributed uh industry forces are enabling [00:46:22.19] [00:46:22.19] these models of connectivity after having worked on them for so long it's really interesting to see [00:46:27.03] [00:46:27.03] that pivot and all of this excitement and energy and we've learned that working with telecoms [00:46:31.20] [00:46:31.20] and trying to scale partners are both really hard problems um and so we're attacking those [00:46:36.04] [00:46:36.04] um and so what we've done is we've built a cooperative network architecture that weaves [00:46:40.08] [00:46:40.08] community infrastructure together to make a nation scale nationwide mno at least in ambition lastly [00:46:47.05] [00:46:47.05] i want to briefly call out that you are welcome to join if anyone here is interested this is a [00:46:51.09] [00:46:51.09] cbrs map from atlanta then you can tell me what's going on in powder springs it looks interesting [00:46:56.13] [00:46:56.13] but in general there's a lot of spectrum here and so you should be able to set up gear as well [00:47:00.12] [00:47:01.20] lastly i want to thank our funders and new america foundation facebook city of [00:47:05.01] [00:47:05.01] seattle google's providing some support university students it's really hard to [00:47:09.18] [00:47:09.18] just mention how many people work on projects with these scopes if you are running a network [00:47:13.22] [00:47:13.22] or 17 networks in the philippines for three years uh the the lift is enormous um and so [00:47:20.00] [00:47:20.00] just thanking everyone who did all the heavy lifting thank you thank you curtis um let's see uh [00:47:30.04] [00:47:32.06] i don't know curtis if you want me to read you the questions uh actually i don't know [00:47:36.06] [00:47:46.15] a few words about pushing higher level services down to the network edge what [00:47:49.22] [00:47:49.22] sort of opportunities are you looking at for hyperlocal services that are community design [00:47:53.07] [00:47:53.07] or might even be hosted at the edge infrastructure really there's an interesting trade-off space here [00:47:57.22] [00:47:57.22] and you know as engineers especially for these talks you kind of have to sound like you know [00:48:03.16] [00:48:03.16] what you're doing before you do it i've always found that really frustrating in the sense that [00:48:08.12] [00:48:08.12] the answer to this is like we're going to run participatory workshops and we're going to talk [00:48:12.04] [00:48:12.04] to our community members about what they want now that is a wholly unsatisfying answer to this [00:48:18.21] [00:48:18.21] question in a sense and uh so what we've done is we've talked with the partner organization [00:48:23.03] [00:48:23.03] it kind of spitballed a little bit uh the biggest exciting stuff has been trying to get away from [00:48:29.01] [00:48:30.04] the role of the city and the role of law enforcement in the city and so trying [00:48:36.15] [00:48:36.15] to build this capacity uh for services that are about connecting people to other people [00:48:41.16] [00:48:41.16] um this is i think most exciting in that context of conflict avoidance and saying what do you do [00:48:48.00] [00:48:48.00] when um there is some kind of event happening in the community where you don't necessarily want [00:48:52.17] [00:48:52.17] to bring law enforcement in on it and so that has been the angle right now which is a very [00:48:57.11] [00:48:57.11] hcie focus there are other applications that we're excited about is networking people uh i mentioned [00:49:03.01] [00:49:03.01] wpac lightly this is a new standard coming out of google for redistributing web uh sites the idea [00:49:09.22] [00:49:09.22] being that you can download a website and then sign it like have google sign the google website [00:49:14.19] [00:49:14.19] and then you can redistribute that and it remains secure but not private and so caching some of [00:49:20.12] [00:49:20.12] those things is really exciting to us it's more of the rural use case for that kind of stuff [00:49:24.00] [00:49:25.01] but is is an angle that we're planning on taking [00:49:29.20] [00:49:39.09] um uh so i guess when when you were um describing uh culti um and uh this idea of making a an mno [00:49:53.20] [00:49:53.20] out of small community networks i wasn't expecting distributed storage to be the um i guess i wasn't [00:50:02.08] [00:50:02.08] anticipating that that was going to be the big technology piece um and i'm wondering [00:50:08.06] [00:50:08.19] uh i guess two things one was that kind of clear to you from the start that that was gonna be what [00:50:14.10] [00:50:14.10] was necessary and then two um you know it is is blocked did blockchain enable something here or [00:50:22.21] [00:50:22.21] or um or not yeah yes and no right like we are very aware of the blockchain hype [00:50:30.10] [00:50:30.10] and you can see me trying to avoid it by saying distributed ledger i slipped up at the end [00:50:34.19] [00:50:34.19] um yes and no is the is the right answer to this right at the end of the day like we're going to [00:50:40.17] [00:50:40.17] need to do distributed state among these the the the core networks that's just how it's going to be [00:50:46.17] [00:50:46.17] um and so i think there's a lot of room to be more clever we want to make sure that it's clear [00:50:50.23] [00:50:50.23] that our innovations are not at that layer we're just using off-the-shelf um hyperledger stuff at [00:50:57.01] [00:50:57.01] the moment and while we think that there may be problems to attack like trying to operate [00:51:00.19] [00:51:00.19] these things behind the vsat links that we talk about in rolex's especially as something like um [00:51:06.04] [00:51:07.01] oh i'm forgetting it's not one web it's the other one that just went live starlink um comes around [00:51:11.20] [00:51:11.20] uh there may be room for innovation there and good research it's just about us sharing state and [00:51:17.14] [00:51:18.12] blockchain does meet our needs it does allow for multi-stakeholder organizations to share [00:51:23.12] [00:51:23.12] this context now it might be the case that the real answer is to just use a distributed database [00:51:28.00] [00:51:28.00] and do some sort of access control on that um in which case you know if you assume that you're able [00:51:33.14] [00:51:33.14] to keep organizations filter organizations that come into admission control into that then that [00:51:38.19] [00:51:38.19] might be it certainly can be more efficient um and so i i don't i think the answer is yes but [00:51:44.17] [00:51:44.17] we're not sure if it's the long-term uh bit there right it could be the case that these other [00:51:50.02] [00:51:50.02] solutions are more appropriate and we'll move over to them it's abstract to us all we need is a spot [00:51:54.10] [00:51:54.10] to put these uh um these authentication vectors um and put the the neighbor table information on [00:52:00.10] [00:52:00.10] there in a way that lets multiple organizations uh do that in a in a secure and safe way [00:52:05.07] [00:52:07.20] here's a question and another question yeah david rome asks how do you do customer service isps and [00:52:12.19] [00:52:12.19] telcos get tons of service questions and requests for service that's a great question which we just [00:52:17.11] [00:52:17.11] have not addressed um you know because of the scale that we're operating at uh so i can imagine [00:52:22.21] [00:52:22.21] multiple models like there's just a reluctance you can imagine and there's a different version of [00:52:27.20] [00:52:27.20] this talk i gave at the ietf where it talks about other organizations that have made models like [00:52:32.12] [00:52:32.12] these the biggest one is fawn it's a wi-fi network in the uk and it was huge there's a lot of fun [00:52:38.23] [00:52:38.23] things that actually went out of business and they pivoted to being uh to working at um airports like [00:52:45.05] [00:52:45.05] everyone else does because that's what money is um but fawn was a whole bunch of wi-fi hot spots and [00:52:51.14] [00:52:51.14] actually xfinity is not functionally that different from it you ins you run this fun [00:52:55.14] [00:52:55.14] wi-fi hotspot and that lets you join the fon network and so as you wander around you can join [00:52:59.09] [00:52:59.09] other people's funds uh networks but they still have that centralized component who does that [00:53:04.06] [00:53:04.06] that service uh and support element i would like to avoid that i would like to be highly local [00:53:09.16] [00:53:09.16] like if you have a problem you know there's like a button to hit to be like talk to uh [00:53:14.04] [00:53:14.17] george who's running this network or sally who's running this network or whatever [00:53:18.04] [00:53:18.04] um and so i don't know if that'll work i don't know if people want to do customer service [00:53:22.02] [00:53:22.02] um and my guess is they probably don't in which case there has to be somebody some wider [00:53:26.15] [00:53:26.15] organization handle that maybe they sell the gear and do support on top of that um right now with [00:53:32.23] [00:53:32.23] tacoma cooperative network that's their job they do that support element um and so depending on [00:53:37.20] [00:53:37.20] the capacity of the organizations it could be not exactly uh you know anyone puts gear on their roof [00:53:43.01] [00:53:43.01] but rather just a small set of organizations which agree to do that kind of service and support so [00:53:47.16] [00:53:47.16] a great question though and certainly a a a large issue we're going to have to tackle in the future [00:53:55.20] [00:53:59.01] curtis um i'm getting a note from tim that we should try to wrap up quite soon [00:54:05.05] [00:54:05.18] um so let me just open up if there's any one last question um happy to send that curtis's [00:54:14.10] [00:54:14.10] way if not i'll ask the last question um so uh seeing the words congestion control [00:54:22.21] [00:54:22.21] and community um process in the same paper title is blowing my mind um is there i'm wondering what [00:54:31.01] [00:54:31.01] you could say about um about the csdw 2021 work yeah it was it was really that like the initial [00:54:38.10] [00:54:38.10] design was exactly like we thought it would be a networking paper we really like there might still [00:54:42.06] [00:54:42.06] be a networking paper to come out of it um but you know it is congestion control it's saying like [00:54:48.15] [00:54:48.15] the the automated systems have been you both worked in these areas as well ellen like they [00:54:52.21] [00:54:52.21] don't like congestion control just is about fair sharing of your links right and when your link is [00:54:58.10] [00:54:58.10] a one megabit per second connection being shared by 300 people there's no fair sharing it just [00:55:02.13] [00:55:02.13] doesn't work anymore and so congestion control as a system and even you know most modern versions of [00:55:07.16] [00:55:07.16] it simply do not work in this ecosystem and so we were approaching how do we get the community [00:55:12.17] [00:55:12.17] to come in and like prioritize traffic in this way a non-automated not like algorithmic sense and it [00:55:18.19] [00:55:18.19] quickly became an hdi problem because it was about how do we explain these to users how do they pick [00:55:23.14] [00:55:23.14] up these services all these kind of things but it is still congestion control it's about that [00:55:27.20] [00:55:27.20] exact problem of this pipe being too thin and too many people trying to use it um so i'm happy to [00:55:33.03] [00:55:33.03] share the paper of people right i mean it's it's dc21 it's pretty far off unfortunately uh we got [00:55:38.04] [00:55:38.04] your first cycle um but that's the idea and so far has been pretty successful people know about [00:55:43.07] [00:55:43.20] congestion they live in these areas they know about service prioritization there are things they [00:55:47.20] [00:55:47.20] like in their network and they don't because of the community network they get to engage on those [00:55:52.00] [00:55:52.00] questions they get to say let's just turn off facebook it's bad it's a community network they [00:55:56.17] [00:55:56.17] get to make that decision there was a fun wrinkle here which we haven't been able to pick up which [00:56:01.01] [00:56:01.01] is me being a little stodgy about net neutrality in that in these environments net neutrality is [00:56:06.10] [00:56:06.10] largely awful like the community has priorities um and they own this infrastructure and i think [00:56:11.18] [00:56:11.18] they should be able to make those choices but that's the sort of big aside either way it's [00:56:15.09] [00:56:15.09] it's a neat agenda that matt came up with and i'm really excited to see where it goes [00:56:19.01] [00:56:19.01] from here he's hopefully deploying it here uh in the summer maybe depending on vaccine schedules [00:56:23.22] [00:56:26.06] all right well curtis um thanks again very much um uh it was great to great to have you here and [00:56:33.01] [00:56:33.01] it was great to hear the talk so um stay safe and uh hopefully see you in person in the next [00:56:41.07] [00:56:41.07] i don't know what six months here or something like that we'll see thanks everyone for attending [00:56:44.23] [00:56:44.23] and the great questions and hopefully i'll see you all whenever i can bye crimson [00:56:50.04]