From Chaos to Control: A Prescription for Managing Apps on Private Cellular Networks - Luis Ariza

This talk explores the challenges and advancements in deploying private cellular networks, specifically 5G, utilizing Kubernetes and artificial intelligence. The speaker, Luis Arisa, proposes a novel solution for simplifying the configuration and deployment of these networks through the use of Custom Resource Definitions (CRDs) in YAML files. He emphasizes the need for a streamlined approach, suggesting that this technology can be accessible even to individuals without advanced technical knowledge. Arisa discusses the integration of AI tools to aid in network management and recovery, while also addressing the relevance of 3GPP standards. The session includes demonstrations of deploying a private cellular network with a natural language interface, showcasing how end-users can easily initiate complex telecommunications setups.

Hi everyone. So some years ago I started to study telecommunication especially deploying 4G, 5G and 5B broadcast subsystem and the idea is at that time was so difficult because everything have to be deployed on premise a lot of repetitive task and a lot of configuration but that must change today because we have a lot of new technologies. So I have a question for you guys and have

you ever deploy a 5G private cellular network using natural language? Please raise your hand. Somebody here is from the teleco industry from the cloud native industry from the artificial intelligence and what about the others? Okay, the idea is to make the deployment of any private cellular network or any network that is 3GPP standardized very simple. So something like this. >> Can you deploy a 5G private cellular

network to call to my grandson? I miss him too much. >> Hey grandpa, see the picture I made for you. >> So the idea is the most important is to connect people. how to use the technology the complex technology make them make it so easy to deploy it. So this is from cows to the control a prescription uh for managing applications on private cellular networks. My name

is Luis Arisa from Vtor that is my startup. I am a cubastronaut. I work one telco artificial intelligence and cloud native and open infra community. See for the for next 25 minutes I will be your 3GPP doctor. See if you see in the middle the RX and when you go to the doctor the receipt started with this. So seven years in the telecom at Cubicon started in

2019 with professor Raymond built its first 5G private solution network using kubernetes. Then it was organized at the telecom user group the telco day the scale and secure introduced the artificial intelligence and machine learning. the keynote from 5G to 6G and in this year in the mobile world congress something happened uh I don't know if you know SRS run was moved to the Lenos foundation umbrella so

those companies are helping to make it that possible so from 2019 to 2025 can kubernetes handle telecommunication and today uh the answer is yes but not the way we used to do it. And earlier this week, Yudel and Asan Seir explained the challenges of running a 5G core on Kubernetes. And today I am following up that path toward a good solution. So what is the disease? The

disease is in the telecommunication industry. We have a lot of odd document. the documents to read specifications that are too technical. For example, if we have to orchestrate a solution, we need a lot of memory. If we are using onapp, the infrastructure is too cost. Maybe the tele the teleco industry or the operator uses vendor locking maybe and the expertise must be from five to 10. for

example, the expert for the core, the expert for the the run, the expert for the kubernetes, the expert for artificial intelligence and so on. So what is the diagnosis? The diagnosis is the patient is overweight, over engineer, over budget and the industry built cathedrals when it needs clinics. So if that continues, it will explode. So we have to find a solution. because we are going to nothing.

What? What is the prescription? The prescription is we are going to use cellular CRDs. I created a CRD to use only one fi YAML file to deploy a private cellular network for example. But for example, if you want to deploy an arrowan internet of things of if you want to deploy a multiccast broadcast subsystem, you can do the same. I created also a 3GPP telco expert that

read all the information the specification of the 3GPP and talks with the the developer because it is a lot of information and everyone is that information is everywhere. So it is a good idea to have an expert that would can help us. We can use something ship like PCs SDRs like something like this to transmit in into the radio. We can use open software and GitHubs for

example using flux reconciliation and also artificial intelligence and MCP to connect all the LLMs and to make able to do something amazing. the guy have the problem is a jaml file with the jaml file is very happy now. So one CRD is mapping the CNCF any CNCF tool that is the the future I want to do. So remember that this can be like a um human being.

The human being the brain could be the core. The heart could be the UPF. The lungs could be the run. The nervous system the monitoring like Promethus agraphana and the immune immune system solution like captain for example. In day zero you can designing and provisioning. In day one you deploy and activate using helm for example. In day two observe and optimize and in the next days the

idea is to self-healing and evolve. This is a fast history of how those open source software went directly with the umbrella of the Lenos Foundation. Now in the mobile world congress okudu is part of the Linux foundation but with the only jamo I mentioned it before we can also use any other vendor neutral solution like open air interface open 5GS free 5G core and any solution for

example that a provider can give us in a helmch what is the idea the idea is we have the the jam file There is a universal 3GBP cell network blueprint. We have AP version the kind the metadata. We can select the t the generation of the cellular network the software of the core the software of the run and anything more everything control with the shark. >> I

am king of many cloud native universes. So remember one jam with one share any cellular network. In this moment I have working 5G standalone LTE but in the future is to introduce nan roban internet of things 5G broadcast LTM and everything using any software available. >> No one in this room can doubt my power bestowed by the cloudnative gods. So what is the convergence? The convergence we

don't have to work in different aisles. If we go for example with the artificial intelligence, it will elucinate 3GPP configs. If we go to the cloud alone, we are going not telecom domain. If we go to telqu alone, maybe we are bend locket uh we use manual configuration. But if we use two is better than one always. What happened with Nephew? Nephew uses artificial sorry Kubernetes and

telco intent. The key agent uses kubernetes and a a agents but no telco and the llm plus the 3GPP don't generate config for the cubernetes. What is the solution? The solution is integrate all the three the cloud native solution the artificial intelligence and telco in order to deploy any cellular network using Additionally introduce self-healing with 3pp domain less megabytes in memory less pots and less money and

is ready for the EUR compliant. So this is a a solution that I created in the left part you can observe that this is a CLI. So the CLI is very important in this moment because you can create an MCP and connected with the artificial intelligence. So the artificial intelligence we execute only the commands of the CL I have but also I create something that is very

amazing uh in real time you can see the notes how is deploying in time. For example, first I create the Kubernet cluster. Second I create the tools that I will use for my application. And finally I will put the Saluran network ports with the user ends. The idea is to use artificial intelligence. So we are using in that NGM telco graph solution some different models that can

be local or can use key APIs. For example, Neotron is launched in the mobile world congress and the idea is that they want to create Neo Neotron like a telco neotron neotron. So in this case we have telco emocron enabled here and we are going to see it. There are more models like Mistral Quen, YAMA and here is very important because recently Dodge Telecom with Google create

an association a strategic association to create MI and this is how to introduce the artificial intelligence into the cellular network to resolve automatically problems into the network. So here but it is very important that the CLI orchestrate everything but the AI only assist and never takes the wheel. So please if you can scan here you can send me some comments about the speech. Uh we are going

to see as you see four demos. It's very important here that no transmissions will occur because I have to adance the authority for digital infrastructure regulatory requirements here in Amsterdam. So I can do that because here this is the the solution to implement it. This the next thing is to observe what will be the demos. So this is an implementation. I don't know if you see well

but the idea is to use a complete life cycle of an implementation of a private cellular network completely from the ground. So I started from kind then the cubernetes then solutions like a monitoring like graphana and promeus then captain and then the application. So here also is very important because the idea is to observe what is happening inside the network. So if I have not enough resources,

I can trigger automatically a rain mediation using the solution that I have in this moment. Why are three GP? >> Sorry. Why are 3GPP standards too? >> Sorry, I can't control the the video right now. I will put it here. So here you can observe what is happen for example with flux. This is the answer the graph that I told you before in real time. We are

seeing all the and here we are ready where the artificial intelligent chat that is called Kim to do everything. We can ask Him please h tell me what is the difference between one generation or another generation. And Kim also can have the ability to manipulate any cluster any container using docker compos helm and Here you can observe how it is implemented. And here you can observe how

is working the graphana to visualize some metrics for example the throughput consumption and everything is controlled with the the flu. multi-tenant solution that is only created with just one um command. At the end we are going to test with a normal smartphone with outward transmitter and the smartphone if it works. And obviously we can define what is the name for example of the network. What is the

bandwidth of the network? What is the band whether we want to configure to our private cellular network it could be 4G or 5G and after it is working >> everything controlled just with one comment. This is our first demo. So I have the second demo. >> Why are 3GPP standards too complex? >> If you want the short answer, it is not complex anymore. Wait, let me reconfigure

my learning screws. >> Yes, sometimes we have these answers of the artificial intelligence. So we have to make sure that everything we do with artificial intelligence is good. So the idea of creating this 3GPP expert is to create a gra. So I use it the LLM's or local SML uh with um repository of all the documentation that is called testspec LLM and that recommendation uh repository I

put it into the obsidian bolt. So I don't know if you know obsidian. So with obsidian I can organize very well all the information. So both of the of them are using for the solution and here I have the same problem. For example, here the CLI works. If I put help, I can see all the comments available. The first one is for uh telco expert. And when

I use 3GPB tco expert, I can select the model can be neotron, can be mal, can be local, can be online, whatever we want to configure or to do, we can do it. And here for example I am asking to him to Kim why is 3GPV too complex and I made a mistake at the end sorry a 4G NSA doesn't exist to see if Kim is able

to identify if I am with an error. So you see also the spec LLM is consulting with the rack the information of all the documentation the of all the specifications of the 3GPP. I here started to provide information why it is too complex in different aspects and the difference between two technologies 4G and 5G. For example, if we have a boy, we can ask the artificial intelligence,

please uh create a response like if you were of you is six years old, for example. The next demo is network slicing. The idea is to use open air interface all the software for the core and the run to implement network slicing. So you can see that the network slicing is the division of logically and isolated virtual slices based on war one infrastructure. So if we use

our CLI and we use NGN telco install slicing, it started to ask for question to configure the cellular network the number of slices and it executes in in docker compose the So in real time you can we can see it what is happening. For example, if I write docker psa we can see the the containers but we are also to see what is happening with the graph.

So when when I launch the graph I can see the difference of slices. And finally the idea is to test if it is working or not with a physical smartphone. And right now we have 5G working only as you saw with one comment with the CLI. >> Excuse me. >> Did you rap? And this is the most amazing erh demo that I have and is the everyone

in this room can be an expert using uh the technology to deploy anything what they need. For example, a kid must to connect with the grand motor. So what is the idea that the it can use natural language to deploy a private cellular >> deploy a 5G private cellular network to call my grandma. Hi Grammy, >> I can hear you. >> So the guy doesn't need a

PhD to execute a private cellular network. It's pretty easy. So let me see what is happening with cloud. in cloud. I use cloud because it has voice recognition and I use one girl here from the staff organization stop to deploy the private cellular network. She doesn't knows about the cubernetes. She doesn't knows anything about the teleico industry but she was able to Hey Kim, can you deploy

a 5G private cellular network network name and a bridge band 78 20 megahertz of bandwidth? So what is the idea? The idea is we are use a CRD by default it use the CRD. Here we are going to see here the deployment of the private cellular network also is working at here you know that to connect with cloud I have to make a MCP to connect with

my CLI NGM telco can you see the implement mentation there. Here is very important because the CRD is telling me what is the network, what software I am using for the run, what software I am using for the core. Let me see if we can observe and obviously I will connect the smartphone to see if it works. There is no service right now. And now we have

she says Anna breath. The name of the network is Annab breath. She also said that the band have to be 78. So the band we are using in that demo is So from a five year old to 120 years old who can deploy a 5G private cellular network your own cellular network is one conversation away. So from engineer friend from deploying cellular networks to create real value

extending the code and for telos that own their future application ecosystem. So the idea is instead of begging for for example the fair share and that is the prescription from cows to control. So thank you very much. I think we have five minutes more for questions. So, do we have any question or If you guys guys want to see how it works, I can show it in

my laptop. >> Hi, Luis. Thank you for the for the talk. It was really interesting. Um, can you elaborate a bit more on what hardware is needed to accomplish some of these demos that you've you've shown here? um like what do you actually need in terms of physical hardware to accomplish what you've shown here? Okay, for in the installation of the 5G private solar network using containers

with Docker Compose, it's pretty easy with a hardware with 8 GB of RAM and a processor of 26 GHz with eight B vCPUs with eight cores. It's pretty simple, pretty bas b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b

b basic but if you want to go ahead and deploy that solution with the flux with graphana with Promethus with Kepton, you need a little bit more. So if you put for example 16 gab gigabytes of RAM, eight cores with 2.6 six gigahertz of RAM, it can be possible. Now, if you want to implement a a chatbot in local, obviously you need a GPU card that manage

that amount of the tokens, the million of tokens. So if you have for example I have in this computer I have a car with 16 I have to download a five millions uh millions of billions uh billions I think it's billions of uh parameters get a little bit of space of RAM just for making or manage managing all the memory of the the algorithm of the chatbot.

>> Can I ask another question? >> Yes, another question please. More >> I should I use this? >> Yes. So um I haven't joined the party from the beginning but um I was wondering like I saw when you are deploying you are like um putting your phone in airplane mode and so on and that is it reconnects again. Have you did some attempts to um basically after

each deployment to like reuse the PTP context that was already established um after the new deployment? >> I don't understand the PDP context. >> Yeah. Yeah. So um the the internet connection have you tried like to um reuse it remain the older one so you don't have like to reconnect and so on after each new deployment >> I use the connection of the computer for example could

be the wireless >> or could be the line and I use like a true throughput >> Mhm. to connect to to the internet. So the network can use the the same internet to connect. >> Okay. So I thought it's actually really going over the um cellular network and doing GTPC, GTPU and so on, wasn't it? >> I don't understand the GTPU. >> Okay. Yeah. It's one of

the protocols that's used in >> Okay. Okay. Okay. Okay. Okay. >> Thank you very much. Okay.