Latest on Crosswork Offering​

Good morning again, and good afternoon to those joining from European region.00:10Thank you again for taking the time to join the, this excellence series or from our experts here. We have we have our customer success specialists and technical marketing engineers joining joining this.00:16Sessions as speakers and demo specialists to to walk you through or dive you through the journey of building autonomous network journey with cross crosswork network automation. This is the 1st session in the series. We have our following series scheduled one per week.00:35And and it's a continued continuity of the knowledge transfer that's gonna that that you're gonna gain from from our experts here. Today's session is led by and Fernando who are the customer success specialists and.00:55Our technical marketing engineer a lead here demoing for you. This is an interactive session, so please feel free to post your questions as, as we dive in and our panelists will be able to support you answer the questions here.01:15And, with that, if you're unable to post the question, please raise your hand and we'll unmute you to put to ask your question live. With that, at the please take it away.01:35Thank you. Thank you so much Barna. Hey everyone, good, good, good morning and good afternoon. My name is. I am a CSS specialist, customer success specialist. So the way we have planned this session is I will open up the sessions and.01:50For half of the session, then Fernando, my colleague from Europe, he will be jumping in for 2nd part of the session and who is our TME technical marketing engineer, he will be doing the demos for us. So.02:08As Barna has mentioned, this is the 1st series of EIS and we will be good. This is a building autonomous network journey with a cisco Crosswork network and what is Cisco latest on crosswork offering? That's what we will cover on this session.02:28So in agenda today, we will be going over, you know, the drivers, so what is the reason for the autonomous networking? Blueprint for the autonomous network, of course the crosswork introduction, all the products that we have it on the crosswork. We will cover a couple of the use cases for, you know, level two, level three.02:49Level four.03:09We will go do the use of the agent agents and multi agents for the AIs. At the end, will do the demo and then we will end with the summaries. It's interactive session, so as you guys have a question, feel free to stop us and or you know just put the questions in the question answer panel and.03:11People will be able to help you.03:30Alright, so what is a driver for autonomous network, right? So in today's network is scaling and it is multi domain complexities that we have, you know, out based that of single human it's impossible to do it.03:34So today about 84 % of the companies are still relying on human led execution assisted by statistic tools like, you know, telemetries, snmps, data that puts out and then we work with that. Now crossing this big distri distribution that we have it and.03:50We want to target by 20 2028 if possible by putting most of the network in autonomy, right? And that is, you know, how would that help us? That would help us by unlocking the faster service deliveries lower OpEx.04:09Improve our sustainabilities and higher service reliabilities, right? To reach this full autonomy, operator needs a practical evolution path. It's not like a rip and a replace solution that we can use it. We need to have a step by step by step process that.04:28Will help us move this partially autonomy network into a full autonomy network. Can we can give it to the customers. So today, most of the things are being done by the CLI commands as well as the some intent based SDN configuration that we are using it, but.04:46Still we are not a hundred percent that. From there we move into the policy based closed loop operation.05:05And from that, we can start with handling a real time optimization and congestion mitigation automatically. Finally, in our roadmap, we want to bring AI frameworks at stage four and five for high autonomy and full autonomy.05:12By developing these collaborative AI agents that analyze the time series metrics, time series metrics and identify the root cause toxic factors, we transition network from reactive automation to a predictive self healing autonomy, right? So what we are just trying to do.05:29Do is this partially autonomous network that most of our customer has it, we're trying to take a step by step by step and move it to the full autonomy. Now, how would that help us? Of course, once we do this, our.05:48Network monetization will be wonderful, right? We will be saving so much money. Our operational efficiency, so more than being just reactive all the time, we will be a proactive. Our resource and efficiency resource efficiency and energy saving will also increase. Last but not least, experience for the excellence.06:04Our customers will also increase cause.06:24Better solution we give them happier customer we will get it. So what do we have right now, right? In today's network, and let me just put my laser pointer. So today what we have it is called a rule based decision, right?06:27What is it?06:47Rule based is that we define a specific condition and then we take actions on that conditions that we have defined. So this works like, you know, well known scenarios that we have it in a network that ok you have some kind of interface overload, then you put the condition saying that 80.06:48Person if this particular interface gets to the 80 % of of the utilization, then give me an error or give me an alert saying that, hey, your interface is 80 80 %. So that is a rule based and we.07:08Choose to like mostly seen issues in our network and we define those rules. Now, this is this way we cannot cover each single, every single condition in the network or corner cases. That's where the AI comes in. AI enables an out driven outcome.07:21Driven approach. Instead of defining every single individual roles, which is almost hard to define.07:41We define the intent. So what we define, we define the goal or desired outcome. An AI agent then will determine what needs to be done to achieve that goal. So it's just we just tell them that this is what we want as an outcome, then AI will.07:47Determine this. Now, how will that determine it? They will determine this information by current network state, they will have historical data traffic patterns and all telemetries to determine the appropriate actions that AI has to take instead of just following the rules that I'm doing it.08:04Now let's just take an example that, you know, I just gave a while before that if you have, if you are putting your rules that, ok, interface that going from the PE to P router, I wanted to make sure that if it gets to the 80 %, then give me an alert back.08:23Now, 09:30 in the morning Eastern start on time stock markets open and that particular interface that bank is using, it's going to be overloaded because everybody is going to log in at the same time. When it's a rule based, it's gonna just give me an alert saying that, hey, there is a issue.08:40Fix it. Either add on the interface, revert the traffic, do something but you are going to see the issues. But if it's AI based, AI will take a look on the historical data saying that, ok, every day 09:30 in the morning we are seeing these issues. It will get maximum to the 90 %, but it will never overload it. It will never drop a tra.08:57Traffic and then you will substitally back subs you know it will.09:17Subside it back and you will have a normal traffic back. So that's what the AI will do it. So with the agentic AI, the system can evaluate the top context and historical behavior before it takes any actions. So if the utilization 85 percentage is normal for particular that time of the day.09:22And you know it will not give you an alert and he will wait until it's subsided back now if it does not, then of course it will give you an alert or it will find a solution and imply that solution so a human intervention will not be a needed. Okay.09:42So the value of the AI is not just detecting a condition, but it's to understand the desired outcome, interpret that network state and determine what needs to be done to achieve that outcome. Now, this is for enabling the path forwarder for the autonomous network. Now Cisco across.09:57Automation is not a single point of a product, but it's integrated suites of application designed to guide a customer along the path forward the auto.10:17A termination network. By combining the key components across the portfolio such as Cisco crosswork network controller CNC, Crosswork planners CP, Crosswork assurance CA or HCO. We enable end to end multi product workflow, ok? A core strength of this.10:27This ecosystem is a multi product event correlation.10:47So when a Cisco Crosswork active assurance or CA detects an SLA breach or a traffic breach, and the health insight identifies this localized packet drop or CRC error, Crosswork will correlate this distinctive telemetry signals into a sing single actionable event, right?10:51Together, this application will orchestrate the service provisioning, deliver a real time path optimization via CNC, and we can execute this automated closed loop remediation without any human intervention. So as I just said it that if there is an SLA issue or there is a packet drop.11:11Up on a particular interface, then 1st thing will happen is Crosswork assurance will find that particular interface that is dropping the traffic, right? And it will give you an alert.11:31Once that happens, then see every using FCO, we will get the live stream datas and that data will be a really a human readable data that will send it to us. Using CNC, we will find CNC and SRPC, we will find.11:45Reroutable path before we go ahead and fix that particular interface, we will reroute the traffic, we will reconfigure it, and we will forward the traffic to the new path. So that will be our automation, change automation that we will use it. And once we do that, then we can go ahead and fix the.12:05Issues, whether it's a physical level issue, optical level issue, IP level issue, whatever the issues that we have it using HCO we can resolve it back. So that's how Cisco cross work automation is not a one product that we are giving it, but it's a speed of the products that we are giving it to make sure that.12:24When we do the, do the automation of the network, it's just, you know, end to end solution, not a single like point of a solution. Okay. So as I was just saying that we have multiple products, HCO, HCO, so all these products can be a.12:43Standalone product.13:04As well as these all products can be a part of like, you know, a huge group of the product. So when we are implementing, you can implement crosswalk CNC on its own crosswork planning for the planning tools, crosswork workflow for the process automation and of course NSO for the mass configuration change of.13:05For add of the configuration. Crosswalker assurance will be for the network assurance to, you know, see all the datas that we are doing it. So instead of having like, you know, to manage a different product, we can manage all this product using.13:25Sorry, all these product multi layer products are you under one umbrella, right? So, these products, what it will do, it will simplify the things like pushing the configuration in a master template or do unable to approach that same service provisioning. All these products.13:46As I just mentioned will be products available as a standalone.14:06All combined suites at CO will be talking to the CNC. CNC will also merge with the Crosswork assurance as well as the NSO. Currently we are working with CNC has many features of the workflow manager.14:10So that is also part of a big suite. Now let's deep dive into the each products that's Cisco crossword product that Cisco is offering, right? Starting with HCO.14:26HCO is a crosswork hierarchical controller, is a Cisco solution designed to unify and simplify the management of a complex multi layer and multi vendor transport network. It provides us a single pane of a glass for the visibility, automation, and assurance across the multi layer networks.14:42For the IP and optical domains. So, e.g., like as you are looking at this, we have, you know, a multiple layers, we have five physical layers, then we have opticals, IP and services. So HCO will give you like a visibility store every single layers of networks that customer has it.15:02This multi layer view, and multi layer and multi domain view will give you an end to end visibilities of your network. So e.g. in a use case that it includes like a RON automation, right? And multi layer IP network manage.15:23Management and multi domain optical network of visibility's assurance.15:42So in case of a troubleshooting, in case where you are having some interface issues CRC errors, instead of just like troubleshooting to find exactly at what point on that interface is dropping the packet, whether it's a physical layers, you have some issues with the fibers or if.15:47On optical layers where you know the issues on the optics on the fiber or IP level.16:07HCO will give you in detail view of that, that exactly it's where the problem is so that's how you're not wasting your time in finding the problem like pinpointing a problem, but you have the ideas and you can resolve that problem.16:14And if you have these three layers or four layers of a visibilities, if for any customer it's easier to manage the network instead of just going blindly in it Our next one is a crosswork network controller.16:30CNC. CNC is the, Cisco CNC is unified open standards SDN controller designed to manage the entire life cycle of IP transport network, which includes the SRMPLS SRV six and five G architecture. So pretty much CNC.16:48To give you a visibilities as well as it will also give you.17:08Automations of you know configuring your tunnels, configuring your pushing or doing using now we have workflow manager, so using workflow manager, we have a couple of applications that you can do a mass upgrade, golden configuration, change configuration, all that. So CNC is.17:16IP.17:33Networks visibility end to end tool. So the reason I said it's end to end tool cause it's end to end automated provisioning provisioning, it will give you an end to end automated provisioning that will help you save a time that requires to bring a new services into the market, right? So if you.17:36Have a profile created then you will constantly proof using that profile.17:55It's also closed loop automation which and proactive traffic engineering optimization for the network health that will reduce the condition, improve the mean time to detect for the services issues. So if you're looking in this picture, you will see that it will give you.17:59Every single channel that is creating between the doors and if there is a congestion then it will give you an alert and it won't only give you an alert, it will also give you a resolution with that alert, so it will be easier. It has a multi, it is truly multilay layer was multi vendor programmable and extensible.18:15Solution that aligns with modern software defined network, right? So the bottom line for this is CNC transforms the complex network infrastructure into a self optimization.18:35Programmable assets that enables you to manage provisioning, monitoring, and assures that a single unified instance, right? So interfaces. So that's what the CNC gives you. It gives you the a a full end to end architecture view.18:48So next is a crossword planning. In crossword planning, we have three pillars, right? A discovery, analyze, and optimize. So what happens in discovery? In crossword planning, it will collect the traffic information from the network and then it will learn all these traffic patterns that how the traffic is being routed.19:09Which L two and L two channels are being most used? Which one are being list used?19:29Based on that information that it has discovered it, it will analyze those information. And now that we know that traffic moves around this part of the network better, then we can test it and we can abstract it. So what happens?19:35So, pretty much crosswork planning is a simulation software, right? So let's see that if this is my network and if a link between router one and router two fails, then what will be the simulation? How traffic will be rerouted? How many packets will be dropped, whether there will be a major packet drop or not.19:51It will give me that that simulation so you know I can proactively add a solution to that. Or if I'm adding a link, then what will be the result of that? If I want to see that, ok, at four years from today, I'll be adding adding 5000 more customers. In that case would.20:11This network.20:31Would be enough? Do I have to add some more routers? Do I have to get some better routers? What do I have to do in order to make sure that when I grow, my network does not fail. So crossword planning is pretty much a simulation of the software that we provide and proactively provide so when customer is growing or something is failing, then we are more.20:33Prepared. So message for this one is that.20:53Modeling the network by creating a custom models by simulating different issues like link failure, node failure or or you know confusion what happens if the traffic or certain network increases. Once we simulate this failure and access the network impact, then we will put the.20:58Solution to it.21:17Now how would it help us? It will help us by, you know, reduce the operational cost because we are already prepared. It will improve our network performance cause we know that if this link will fail, what will happen to our traffic and how will be maybe the least impacted. It will enhance our agilities, our proactive.21:18Planning will be much better. We won't be in reactive mode, we will be in proactive mode. And we will simplify this capacity planning, right? So at least we know that if we are growing, if our network is increasing, then what will our future will look like it. So that's what the crossover planning will help us.21:39Now Crosswork assurance, like I just said it in the slide below before that Crosswork assurance also has visibilities inside layers, action layers, and last but not least, it would have a user experience. So what is visibility? So we focus on how we see the network. So for.21:591st, let's look at how we gather the information. So think.22:18This as an eye of our network. We just don't rely on the sources, but we use this the combination of a hardware and software sensors that acts like a digital health monitors, ok? Now it will allow us to see exactly what is happening across the system in real time down to the microseconds, ok? So we.22:23We will get the real information from the network every single seconds.22:43So nothing slips through the crack, the entire visibilities of the network. Now we have the information, we have the visibilities, then we focus on making sense on these data that we have it. So once we have all these data, it flows into our brain, which is our inside layers.22:48This is where the magic happens. Instead of looking at the role numbers, the system will analyze this data to find a pattern and return. Whether we run this into a cloud or our own servers, our custom or our customer servers, so you can run these data as you can have this crosswork assurance either on your.23:08Server cloud or customers the rest, it turns this complex data into a simple actionable insights telling us not just to do something in or something that is happening, but why it is happening. So it will give you a whole view that.23:27What is happening and why is this happening? Now we have visibilities, we have inside. The last thing will be the all hand on the deck action layers. We focus on fixing the problem. This is where we move from seeing the problem to fixing it.23:41Because the system is open and flexible, it can automatically either triggers the alerts or even fix common issues if it's, if it's owned through the automation. This means we spend less time manually troubleshooting and more ensuring that the network is running smoother.23:59Now, we focus on who it benefits, right? So this is just not the engineers. Not only engineers like us who is troubleshooting will be focusing will be benefited, but whether you're a service manager looking for oral health or capacity planner preparing for the future.24:19This solution will give you give everyone single.24:39Clear windows into our network performance. So that's what the crosswalk assurance will give us. Now, most of this product that I just talked about a while before, all of this product are not only supporting Cisco. So using EA and TC interrupt.24:42Testing or abilities, we are adding, we are expanding our horizon by adding more and more multi vendors into all of these products. So cause today's network is not made by only Cisco networks. Every single customer has a multiple different vendors that they have in their network.25:03And you don't want to use like ten different automation process for the each vendors. So that's the reason what we are trying to do is we're trying to expand our horizon by adding all the vendors into our crosswork applications like what we are seeing in this one right now that we are supporting in CNC we are supporting Hawaii, we are supporting Nokia.25:22Have you're supporting Juniper.25:42Xer all of these products, so that's how you guys have a single automation tool in your network to operate. And each year's we are, you know, adding more and more new vendors to expand our horizon so that's how customer will have a full.25:45Access to entire network. Now I will hand over my rest of the presentation to Fernando. Fernando, do you wanna go through the next slides? Yeah sure thank you thank you.26:04Just let me know. I will share now my screen just a moment. Before Fernando steps in if there's any questions that anybody has, this is the time for questions. Please raise your hand and we'll unmute you to ask your question.26:20Yeah 1 s Yes, please go ahead with your questions. Hello. So my my question is related to the.26:37Crosswalk discussion you guys explained before.26:57I was wondering which type of models could be implemented to foresee the traffic in the network talking by like a machine learning type of model, if it's applied any, let's say I.27:01Algorithm or some types of algorithms specifically.27:21To to foresee the crosswalk. So can you repeat your question? I'm a bit confused. So are you? Before you, you explain to us in the slides if you go, I I can pinpoint the slide you were talking about.27:25Can you go back, Fernando?27:45To the cross to the crosswalk when you were talking about expanding the network and implementing models that can that could be, that could sustain a radical change in the traffic. There was a specific slide and you explained to us that's the one we talked about.27:49This one?28:08Yeah, in, yeah, in those type of models that we apply, to give you an example, ok? As of now in our internal platform in fibercop, we have implemented the PCA, the provider connectivity assurance from Cisco.28:09As a central orchestrator to observe traffic. I was wondering whether there could be some solutions that are being implemented that rely on machine learning algorithms to foresee the traffic instead of monitoring traffic, but you.28:30You're saying that instead of current Exactly. Exactly. Are there any sort of automatic system that extract data?28:49From the let's say machines to foresee and not monitor the traffic. Yeah. So crosswork assurance as of for now currently sees what is currently happening in the traffic, not four seas what will happen based on that. But for the crosswork planning if you if Fernando if you go couple.28:59Of a slide back, maybe that would help you, that in crossword planning.29:19You would be able to, yeah, you will be able to simulate your traffic grow and you will see using that. So that's where the crossword planning will help you to stimulate that problem you know and then which type of which type of machine learning models are you used?29:24The basic of AI. I was wondering just the type of algorithms to be implemented, like, are there any type of specific algorithms or.29:44I don't know what kind of a machine learning it will be used. I can figure out if or anyone else know please chime in otherwise I I will, I can definitely get back to you on that. Okay, ok, thank you. And so what's just out of curiosity, sorry. Yeah.29:54So, just if I have to answer I mean I'll just give my fair share of thought here. Today we don't have any model that is sitting locally, so you can, this is an algorithm based thing.30:13Okay, the crosswork planning is more of algorithm based and going forward yes, we have I mean like we are building an agent of crosswork planning which will be sitting on crosswork AI where you can use the of your own choice and make it more predictable. That is one thing and we are also working with.30:28In future you are going to see an integration with one of the very key articles it's it's too early to talk about it right now, but yeah, we are, we are working with one of the prominent vendors to, you know, integrate it with their model.30:48Okay. Yeah. Okay, thank you. Thank you for the question. Are there any question any more questions? If not, I think Fernando please go ahead, thank you.31:03Thank you, so now we will start talking about a little bit of the kind of scenarios that we can just model into the into the CNC and the use cases that we can have. 1st of all, I want to show you this challenge that we have over here. So the thing that we have is that.31:20That we have the manual reoptimization based on the network changes is not scalable and process risk to target the eh service level agreements, ok?31:39Now, a link failure, a maintenance event, a new circuit or even a metric change can invalidate the assumptions behind existing traffic engineer paths. Now in a large network, these events happen constantly. So nobody can re verify every segment routing policy against the.31:50Service level agreement objectives. So in practice.32:10Path silently drift out the compliance of the 1st sign of trouble is a customer compliant or an SLA credit. Now, what do we have here? We have this objective that is the latency IGP team metric minimizations, and then we have those constraints that we have such as affinities.32:14Joint plans.32:34And bandwidth. Now, which is the solution that it can come we can find here we can have we can have here on the real time network customization. 1st of all is that you are going to define the policy internal ones, meaning that you express the objective and the constraints, not the.32:35Now secondly, we are going to automatically detect the topology changes. The controller has a live real time via view via V BGPLS and the PCP. So it knows the moment the network changes.32:56Now on 3rd side on 3rd side we have the real time recomputation. Now this is going to be of the parts that are in violation of the optimization metric that is inbound in violation of intent. And last but not least, we have the optimized path this that is automatically provisioned.33:12Now the loop closes without a human in it. You can see that all of this mo all of this solution is going to be.33:32It's a closed loop operations. So, here what we are going to have on the outcome is going to be preserve the service policy intent and associated SLAs. Now the network stays compliant by construction continuously, not just that we are turning it off.33:41We also have the enhanced operational agility with real time actions, meaning that the reinsponds in seconds at machine speed rather than at the speed of a changing a ticket. And also we have the optimal utilization of the network capacity. Now you can run the network closer to its real limits.34:01Because the system is actively managing it, which differs the CapEx. Now, this is something that we are that we are going to have this type of of level three use case that we are just mentioning, but it's not the only one. The one that I want to mention as well is this.34:21Congestion mitigation. Now the LCM detects congestion by comparing the interface utilization against the interface threshold. Here we have the, the, the, the challenge that we were just mentioning. The dynamic network stayed with risk of congestion leading to degraded service level and the user.34:41Experience. This is some typical this is some typical use case that we may have.35:01What we're just trying to get here is to quickly mitigate the impact on the end user experience. Now, the LCM determines minimum amount of traffic to divert from congested interface and compute spath of local and the tour DSR policies.35:07Now, eh, here the recommendations that we have is that the SR policies are going to be deployed via the SRPC and based on changing network traffic and topology, the LCM will recommend further actions for the user confirmation.35:27Meaning that this is going to be the solution that we are having here on the right side in the on the bottom where we are where we are handing the concession handling using the tactical traffic engineering policies, the automatic, the automated path recommendation, the user approved or automated more for path of settings and provisioning.35:47And and finally the historical record of past conge congestion events and mitigation actions.36:07So every time that we are trying to face this type of situation, it's not only that I am going to just book it and know what was happening. No, the system is going to learn from the issues that we were just having. Now.36:13Another thing that I would like to talk to you about is like the risk factor de detection. Now, toxic factor detection is a generic capability that identifies the toxic patterns of factors that drive multiple failure types.36:29Now the inside behind it failures are rarely caused by one only sync. They are caused by combinations, a particular software version on a particular hardware revision in a particular role, in a particular location with a particular configuration. Now individually.36:45Each factor looks harmless and passes every threshold check. In combination, they are statistically toxic.37:05Now NO human can find those combination by I across thousands of devices and months of event history. Now a machine can, and that is precisely the kind of work that justified the AI rather than another dashboard. Now identify toxic patterns of factors that drive multiple failure.37:13Types. Analysis maybe periodically scheduled and corresponding the alerts maybe provided to the users.37:33Now, the thing here that we have is that every time that we are going to find a toxic factor detection, it's a combination of many many things happening. Here's an example. Software version seven.seven.one plus the hardware version em.37:42X Y Z rated major and location Lexington Avenue plus the device group lex.38:01Two, this is rated severe. What am I going to do with this? Because this is a combination of factors. That's the toxic factory detection. Now, this is how it's going to looks like. We have our network infrastructure here.38:10Then we have all of the network events that we are just, finding, and then under my network infrastructure, this is what is going to happen. I have an amount of of operating systems, I have.38:27Not only the operating system, the version of them, the model of the devices, the hardware, and so on, and so on. So this is the path that we are just going to follow. Easy pc, but at some point we are just going to have this type of problem, right? What is going to happen?38:441st of all, I am going to start the pattern identification algorithm. And then this is what we are, this is what the path is going to to to measure. We have from one side the toxicity that is going to be high, and then on the other side the toxicity that is going to be medium. You can see.39:02The percentage of the events the percentage of inventory and that's the analysis that the pattern identification algorithm is going to.39:22Show you. Now the system correlates common factors among the network failures. Just to mention some of them, we have toxic factors that are categorized by their toxicity and related toxicity, helping the operations teams to prioritize and addressing the highest network risk rather than just the symptoms.39:30Now the outcome enables informed decisions for software, hardware rollouts, and proactive network updates. Improving mean time between failures and overall the network reliability. Now the toxic factory detection agents work across various devices and operating systems without requiring.39:50Prior setup or configuration.40:10Now the future enhancement includes two automated actions to address identifying problems and sharing anonymized reports to build a community knowledge base. This is what I was telling you not only just going to detect it, store it and that's it, it's going to be learning.40:12And also here we have the deep network troubleshooting is just going to summarize just a little bit of all the things that we can just troubleshoot inside our networks. We are dividing here in three pilots, which is the deep research and the deep research deep parallelization.40:31And the semantic resiliency.40:50These three pilots are going to act individually in order to start this troubleshooting. However, I believe that it's better if we are just going to see how he's working in the real life and I will pass the ball here to our TME and he's going to show to show you a real.40:53Interesting demo.41:13Showing the value that we have on on these type of scenarios. So the floor is yours, my friend. Hey, thanks Fernando and team team I'll get started with the DNT demo to begin with and then we will go into the, the next layer layer three.41:14Use case, which would be the path analytics today. So let me go ahead and get started. Let me know if you guys can see my screen.41:34You guys can see my screen, right? Yes, we can. Thank you. So yeah, this is a pre recorded demos just in the interest of time. Today we will be talking about how the deep network troubleshooting with Crosswork AI is aligning with the level four autonomous network vision.41:43You know, of the TMF forum.42:03D and T uses Agent AI to accelerate the issue resolution by analyzing the data at scale running multiple troubleshooting tasks in parallel, and it also breaks the complex problems into small and manageable steps. With that being said, let's get started with the demo. Troubleshooting agent.42:06So, as you can see here, you know, let, I'll I'll keep pausing it and I'll keep running it. So, just let me know if you have any questions. Here today we'll be looking at, you know, monitoring a service or like we will be debugging a segment.42:25Issue in this in this demo and we will bring it down manually and then we will try to figure it out using our DNT, what's when what went wrong, what is the, what is the root cause for it and let's go ahead and see this in action.42:45You know, so as a part of this demo, you know, let's watch, we will, you know, walk through how to quickly identify and resolve an issue with the service policy. In this case, we are seeing.43:02Yeah, we are monitoring the SR policy with color 1291 and just to show that I mean like if you see, it's operationally down right now.43:22You know, because we have operationally brought it down and we want to see why we are going to investigate why we brought it down. You know, you can click on the top here on this blue button and it will open the crosswork AI assistant, which can be cross launched from the CNC itself directly. You can have a single view like this or you can.43:32Have a full screen view. We'll go ahead and start with the full screen view where we can, we can directly, I mean like if you go here to the Crossbook AI, this is this can be one of the directly.43:52Direct pages, and as you can see, I'm changing into the full screen view here. Once you have the full screen view, you can pick the deep network troubleshooting. Yeah, so there's a question what CNC version is the demo with? Yeah crosswork AI runs with CNC seven.21 SP.44:05And Crosswork AI, the demo that we are running is on Crosswork AI two.zero.44:25Okay, we have these things these images available today. They're going to be crosswork AI is going to be beta version and it's going to be available starting 1st week of October, ok? Yeah, once, once you get to the screen, you need to provide, you need to provide your.44:31Problem description, what, you need to provide as much description as possible. And again, this is just to show how we do it, you know, again, you can automate it fully by saying that I'll create an agent.44:51Which will, you know, pick up the alarms from the network and then create a problem statement and then provide it to the DNT and the DNT will start investigating, ok? So this is possible, but today, just for the understanding of how D and T works, we are going to go through the process of filling up the.45:06The problem description here. And once you fill up the problem description, as you can see, you know, you can select what type of, what is the window that you want to, you know, look at and how much is the time frame and those things. You can have a troubleshooting mode, which will, which will define how many hypotheses it's going to come out with.45:25Like e.g. in echo mode you are going to get with it's it's like an economy mode. It'll come with it'll come up with three hypotheses.45:45It'll come up with three hypotheses and in Optima it would come up with like seven hypotheses in a fully performant mode, it would come up with a maximum of 13 hypotheses. And again, I'm talking about the maximum number of hypotheses it's going to come out with. You would also provide the observed symptoms that you have in a.45:56That you have observed during this particular issue when you're debugging it, in this case we are selecting performance just to show that how many hypotheses it's coming up with. You know, you click begin investigation and as you can see, here the agents are, you know, it's breaking the this complex task down into small sp.46:13All things and it's reaching out to the LLM of your choice that has been pre configured here on the crosswork AI and it comes up with the hypotheses that it has identified that are related to the.46:33The issue that you have described. As you can see, these are the hypotheses it has come up with. We give you an option to select and deselect the, the things that are that you think are relevant for this issue and there are some things that are not relevant. In this case, we'll only pick two of the hypotheses.46:46Which we think are relevant just to, you know, just to show that it is, it is done. Okay, let me pause it here very quick minute. So if I scroll back quickly here, yeah. So I'm just defecting here and then we are going to let it run in a minute.47:03But what I want to show it to you here is we also give you an see before we click the start validation, there is an add hypothesis button as well. If you think you want to add something very specific to it, you can add it, we let you customize it as well.47:23Once you customize it, once the hypotheses are set in place, it will, the sub agents are again triggered. They will go ahead and collect the network data that are related to this particular hypothesis. In this case, they are collecting the data.47:39Related to the two hypotheses that are running or that have been that have been identified, and we have an observability page that we that I just let me just scroll back. So yeah, let me be here very quick minute. So 1st we are collecting the data. We are, we are preparing the validation checks for those two hypotheses.47:55We are trying to investigate based on those two hypotheses if the hypotheses have show commands, it is collecting the the agents, the sub agents are collecting the show commands from the devices and then they're trying to correlate the information that is relevant and then it's going to provide you with an output in a minute. And, as you can see here, it will go.48:15To the observe, I mean if you want to look at the observability page, we also provide you an observability page where you can see what LLM calls are made and what is happening behind the scenes as well. So once the investigation is completed, you know, the assistant will provide you with the likely root cause of what has happened. It will give you the.48:35Different different.48:55Details about why it thinks that is the cause and it will tell you clearly what is the configuration that is in this case a flex algo 01:29 mismatch configuration has been identified and you can see that the hypotheses is being.48:57Cross verified and you know you can see what is the timeline, what is the cost, what is the effect, and how did it, how, and what needs to be changed to, to resolve this issue.49:17It will provide you these things. Once you think, this is good, you can provide your feedback and then there's the learning that happens behind the scenes going forward for the future use cases. Now in this case we are again just going and verifying what is happening here.49:29We are looking at it. We are looking at the flex algo. Again, let's look at the traffic sessions here, flex algo. I'll go 01:28. We are just going ahead and, you know, making the config changes that we made in the past.49:45So, whatever it has told us to or whatever it has recommended us, we are just changing the priority here and then you know you're saving the config, now we can go back and see that the color one to nine one is operational.50:05There's an operational state as you can see. So it's good. So, you know, this completes the demonstration for the deep network troubleshooting. Okay, and the key things here are like again crosswork AI is a multi agent framework.50:22We are coming out with three agents out of box, which is config drift, deep network troubleshooting and risk factor detection. We also have the SDK that is open, so customers can create their own.50:39Agents and they can onboard it to us and they can leverage off our infrastructure needs that we have, like we have discussed, like our knowledge graph and other things, the dras and other and the rest of the things that we have already built to to make our life.50:56To make the life of the customers easy when they create their agents. And again, you can also create east west traffic between the agents where they can communicate with each other. An example is like if you want to create an agent, let's, let's think of an agent like an operational agent which is constantly pulling the cnc's operational.51:13Alarms page and then whenever there's an alarm that shows up on CNC and.51:33It constructs a query and then a query is automatically pushed to a DNT and then the DNT troubleshoots it and gets back to you with an answer. Again, you know, this is just one example which which may not be the.51:39Which this example is trying to show, showcase how, you know, fast and, you know, we can learn from the network and we can debug the network, you know, and again, you don't need to have an agent at DNT triggered for all the alarms that are present on your system.51:57Okay, with that being said, any questions, any other questions on the DNT side? Yeah, we have a question here from Christy Altoy and he's asking what? Yeah, I answered that. I'll log in, i'm happy to answer that question again.52:15So CNC seven.21 SP, that is the latest release, is integrated with Crosswork AI, which is a beta version. We call it as crosswork AI two.zero. So these are the versions that, that I've used in this demonstration here today.52:35Okay, so if there are NO other questions, I will start with a path analytics video ok so yeah now let's go ahead and get started with the path analytics.52:53Demonstration here today.53:13So I'll just run through a couple of slides 1st, you know, just to explain what it is and then we can take it from there. Now let's start with the overall architecture of the IPM at the foundation are the measurements measurement sensors, probing capabilities. These provide the visibility into the network performance, including latency pass.53:15Loss, liveliness, latency histograms, and absolute loss. CNC Essentials provides the path analytics layer. It collects and analyzes the path information across the network.53:35Path information across the, across the network, including different DCMP paths that the traffic may use. The crosswork assurance layer provides the measurement and analytics infrastructure. It maintains the time series network history, performance measurement path correlation and provides visualization.53:47The integration between the components uses between the components using GNMI and GRPC, the data is exchanged with the assurance platform. While the IP measurement, IP performance measurement insights and actions can be, can be sent to the crossword network controller. This creates an.54:07The complete operational loop, the network is measured, the path and the service impact are analyzed and the results inside and the result insights can be used to initiate an action or a remediation workflow through the CNC in this case. You know, from an operational perspective CNC provides service and network topology visualization, service health monitoring.54:26Bringing dynamic traffic engineering and closed loop automation. The key value is bringing the performance data and the routing and routing context together so that the operators can understand.54:46You know, operators can understand not just the services degraded but also why it has degraded. Let's look at how the path analytics information is exchanged between, between CNC and the assurance platform.54:57This, this slide kind of let me this slide shows the integration model of CNC and our crosswork path analytics API, you know, the path.55:14The path analytics component maintains the visibility into the member notes. It also has the links associated with the flex algorithm.55:33It receives the routing information through from sources such as SRPC and BGPLS. With the crosswork with the with the path analytics container and a comp a companion process manages the path registration and the path updates. It receives crosswork infrastructure notifications and uses them to keep the path information current.55:42The path analytics process then registers on registers the paths as needed I mean it depends it can register or unregister the path as needed. Computes some path updates stream and streams those updates through GRPC to the crosswalk assurance in this case we are calling it we are going forward, we are going to call it as.56:01Previously it was called as provider connectivity assurance. The important point is that this is a dynamic integration. The system is not simply exporting a static topology snapshot. It, it is doing routing conditions, topology change, the path information can be updated and made available to the assurance platform. That allows.56:22Some the PCA or the crosswork assurance to correlate the service performance events.56:41With the actual routing path that is being used by the network. In practical terms, you know, in practical terms, the API provides the bridge between the routing intent and the performance assurance. It enables the assurance system to understand the path while CNC maintains the routing, routing and the network context to.56:46Needed for analysis and automation. With integration in place, we can look at how the operational capabilities are exposed through routing analytics.57:06I will use, I will use postman collection to, you know, to post, to post to the register for a path on a flex algo member 01:28 in the demo between the source between the sources where between the source router and R nine and the destination router R one in this case in the demo.57:15And then we will subscribe to all the paths, which is if you are registered for multiple multi path, you know, this is the token API that you'll be looking at or that you will be used to subscribe to all the paths or you could subscribe to a specific path also in this case in the demonstration case, we are going to use flexalgo 01:28.57:34But you all know.57:54I'll take I see the question come across but I'll take the question once I'm done with the demonstration. But know that you will have to give the path key as I mean, just, just to remember, just to reiterate what I was just saying.57:58This is the one that I'm showing here as a token API that you can use to subscribe to all the paths or you can, you know, you can pick a specific path that is in the demonstration we'll be using flexalgo 01:28. But, but know that you will have to give the path key as a source and destination.58:15In base 64 encoded IPv six formats.58:34IPv six address, ok? With that being said, let's get started with the demonstration. Just give me 1 min.58:38I think I embedded into the into this one, so now it just running on two nodes and.59:01So, yeah. So let's get started with the demonstration today and here as you can see, this is the CNC topology which is running on two nodes, two nodes here. As we can see here on the topology, there are traffic engineering that is running. I've gone to the traffic eng.59:18Engineering tab to to learn more about the flex algo algorithm. So I would start with the you know.59:37Postman collection to, to register, to register for a path for IPv six, which is between the source and destination. Again, I'm going to send this again as you can see it is send, I see all the status as ok. I have, I should have a registered path. Let's go to the get registrations and you know.59:46Just to confirm that we have the path here, you know, I see the path.01:00:05Now, now I'll turn to the GRPC stream to to subscribe to the paths where I should start seeing that I'm receiving the notification through path analytics. So the GRPC stream and it tells me that the status about the path, which is the source and the destination, the path is stable, the source, the source name.01:00:12The source name and the destination name, the metric type, the value and the value and then the link and then the link level information are all there. If there are ECMP paths across the pass it will give me a status for for all the nodes within the information. This can be subscribed.01:00:31This can be subscribed through a steam as well and once the operator is done, you can click on the unregistered path just to delete the path. In this case, which I'm going to show, which is I and I'm going to send. And once I, once I send this.01:00:49You know the the notifications on the GRPC entries are stopped. The use case for, the use case for route analytics is also on the VPN services as I go here. Let me pause it here quickly. So here, you know, the use case for the route analytics is also enabled on the VPN services where for.01:01:05VPN 01:28, now sorry.01:01:25I shouldn't have paused this ok I'll let it fast. So again, we are starting the the VPN thing here 01:28 which is using flat Cellgo member notes for.01:01:31VPN config, which is in this case the service topology and then the VPN instance profile can be expanded to everything so that we can, we can see, we can see the details and the requester is the locator latency, which is I'll go 01:26 in this case.01:01:51I can go to the transport. The next step is to go to the transport layer. As you can see, I'm seeing the low latency here. Let's go to the transport next. I can also click on IGP flex algo path, and I can select between to stay, give the path.01:02:10Between R one and R nine and it gives and I can click through so that I can go to 01:28 and get more information. You know, it gives me both the baths A to Z and Z2A.01:02:30I can also overlay them with the metrics to say IGP and the delay and it will tell me which path, which path is being used for the service as well. We will, so we'll I mean again I just want to, you know, the APIs that I have used, we have put it on, I mean they're available on the devnet.01:02:41And you know also on the GitHub based upon what, where you want we can we can provide you with all the information. Again, I have included this information in the slides that are upcoming and you should you should have more information there again this is this is the yeah again going back to.01:03:01For our next slide. Here we are talking about again like this is the same thing that I just showed, you know, but, what we just have seen so.01:03:21Let me go back. Okay, so again, but the but the overall use case I mean again these are the slides these are the screenshots from our APIs that I've just put in just for just for everybody's reference, when somebody is trying to use or trying to do it with with the slides, they can have all this.01:03:41Information available here.01:04:01Okay, again, once we are done with this, we can just, you know, I just want to show that you know here are the references for our documentation. The APIs are available on devnet and also the GitHub, you know, depending upon how.01:04:02We want to pick them up, the samples are provided here. And again, I just want to reiterate the use case I mean in this case, you can today subscribe to the paths through path analytics, view the responses, and then you can list, you could list your registrations that are currently available or once you're done, you can register the register.01:04:22Paths through the APIs. So the idea is route analytics is used as a function to tar to target initial use cases on traffic engineering and optimization and also the point analysis for the policy creation and overlay them over the VPN services. As I've shown in the demo, you can see for a VPN service, what does the flex.01:04:42Salgo member, then you can also visualize the IGP paths between the two nodes. We have shared the references for your usage. Please feel to try them out and, you know, please let me know if you have any questions.01:05:02Okay, with that sorry with that I I conclude again thank you for listening and and I'm happy to take the questions if there are any. I saw some questions come across.01:05:16When I was demonstrating Yes, I mean, the it's a follow up question.01:05:36I think that the the questions have been answered on the chat window. That's cool. Yeah, I I'm seeing that. Yeah. But if there's any question you'd like to answer verbally among them, please.01:05:45Are these capabilities supported for 3rd party devices on, onboarded on CNC? Would you like to take that question from? We will get back to you on the specifics of the 3rd party devices that.01:06:01Are, that can be supported with this Mohammed and the team. The reason is.01:06:20Yeah, we support a lot of multi vendor devices and I just want to confirm like what is the matrix of support for path analytics on these devices. Cool, thank you. Thank you for the question Mohammed and.01:06:27If there's any further questions, please feel free to raise your hand and I'll unmute your unmute you to ask your question live.01:06:45Okay, if there's NO further questions or you can also reach out to us via email. Our team here will be happy to address any questions you may have after the session. All the session recordings will be posted on the community page.01:07:02I've posted the links for the session recordings registrations for the upcoming C sessions and additional related resources for crosswork network controller on the chat window, please feel free to you know get them and.01:07:19There's NO further questions, we'd like to appreciate the time you've taken to join us to learn from our experts on this topic. The next session on this series is will be next Wednesday, the same time and the topic will be on.01:07:38Leveraging the crosswork net network controller and it will further deep dive in covering on the evolutionary evolutions that's happening on the CNC. We look forward to seeing you again on on the next session until then.01:07:58Please take care. Thank you everybody for your time.01:08:18