 Good morning everybody. Are we awake? Yeah good. I Was it this morning? I was I think about five o'clock. I went I went out of my bed and drove drove here I live in the middle part of Holland and I want to be early because I didn't want to go in a traffic jam With apparently failed. So I was in a traffic jam and the story. I'm about to tell I think today from a business perspective is That we also had a traffic jam and we are finding a solution with all open shifts to solve it But that leaves me I hope I is there a clicker somewhere out here Who am I? I Have the honor to be One of the managers of the open shift and now what the manager of the open shift and Linux A team we have at Akhmeya. Akhmeya is one of the largest insurers in Holland We do help insurance. We do Mortgages we do claims insurance about 15,000 people working for us We have I have a team of about 30 Well, it says over here great people and they really are great people Divided into three five three teams for Linux Ansible and two teams separate for the open shift development and we have the responsibility for approximately 200 Deployments service either in cloud or on-prem and about 250 namespaces that we use also development acceptance Etc. And that's what we do and a time ago redhead asked me Can you tell us something about an implementation that you did an implementation? I think is interesting for a lot of people to at least understand or get some some knowledge about and it was for us well, and Also a journey to do this, right? Let me tell you something about that as you can imagine Akhmeya as a financial organization has a lot of financial models They we have a lot of with kids Calculating a risk premiums, etc. Etc. So and we have a lot of financial data also customer data financial data, whatever So those those guys and girls by the way, they need to calculate with this data What the next premium is because if there is an increase and for now, well, let's say A half-year life expectancy Premiums have to be recalculated. So that's a lot of work that has to be done and they have we have models for that and One of those models is a KPM. I will tell it in Dutch. It's a capital prognose model Basically, it's a stochastic process calculating Some financial data, whatever the financial data is. I don't think that's too interesting for here It's we used to do that with mudlap, but apparently from a business point of view that was too expensive and They build a new solution and they build it in C sharp. It was also But they tried to do it in Python and and are but basically C sharp was better for parallel processing and We had a lot more C sharp knowledge internally than Python and and are And what they did they they built something and we implemented that it was implemented by the way It wasn't implemented by me, but I'm just a manager And What what they did they implemented it on 10 calculation engine and the process has to run we have to do about 2000 sub-processes and It runs 24 7 27 to 30 hours straight And that takes a lot of time and if you have to do it more than once a month it takes a lot of time and And it's not efficient And we also have an interest in doing things a little bit more efficient and cost-effective Because if you have then Fixed machines in our data center and you only use them for let's say. Well, maybe maximum two days a month It's rather expensive and Not very efficient so What we What what I always say we but basically it's day who did it. I'm just telling the story What what what they did is Rewrote the architecture the original architecture was and I tried to make it as simple as possible Was that they have a complete process and they divided it up in single steps and they wanted to optimize each step And I thought oh, that's a good idea. Let's optimize each step and we can run those parallel hence the C sharp programming Eventually that wasn't The good way to go. So what they did. Okay, we have one complete process one step and we can better Run those parallel and that's what they did But if you want to scale that up that means that you need more Well processing power basically And as I said That's a little bit expensive. I'm not very efficient In the same time I may as a Company and probably we are not the only one in this world is making a transition to what's the cloud We have traditional data centers, but we are moving our data centers to the cloud We already have a very large SAP Setup in the cloud But basically what we are going to do is Move all our stuff from the data center to cloud and we do this on an edge a cloud um Sorry and We want to do that because we think it's more dynamic You can use it as pay as you go use it whatever you want and that's what we are currently doing and we have of course implemented open shift on Open on on cloud. We also had it on prem, but Happily we are getting rid of that It was a bit of a pain in I'm sorry for my language in the air but We are having this now on cloud and The combination of open shift and clouds gives us the possibility to scale up and scale down because that's the the basic of of Open shift and it also gives us the capability to make a cost efficient solution Which is KPM now I come back to my traffic jam What we currently have implemented is There's a drawing over here, but never mind. It's too small. I think for you to have We have a Calculation part and then we calculate it and it takes about seven to eight minutes to do one part complete and One calculation engines is for CPU and team gig memory But we need to have two thousand of those calculation part to do the complete Calculation the complete projection But what we can do is scale up and basically zero or one to a hundred Machines in the open shift and we can do that Within approximately one hour time you can scale up within one hour from basically one or two machines which are constantly running because it has to run somewhere to One hundred and that means that instead of 27 hours to 30 hours We now have well approximately two and a half to three hours to do the complete calculation Instead of those 10 fixed machine and of course if there is less Less power needed. All right, we scale we can scale down if there's more needed We can scale up, but there's a catch But this is what we do and On a cost-efficient point of view We did some rough calculations. I think we save the company about half a million a year only for this calculation only And because we don't have to have 90 extra fixed machines in our data center Which costs a lot of money We have a quarrel a Marlene. We still have to do some some subscription Negotiations I think but we'll get there One way or the other I hope But that's that's that's the idea and it's huge so Yeah, that's what we did And I say we again, but one of the engineers is in the back room. Actually he did But what did we learn? Sorry, what did we learn? The first thing is that there is a difference in How the business views as your cloud or cloud in general I think and how we look at it I have many conversations with business managers and They say yeah, but the cloud we can do it and scale up and fell down and then In in in general we can't because there's a limit in the amount of hardware that Microsoft has in his data center you can't scale endlessly and As from my point of view we have a constraint We only want to be in the in the vesture Europe as a cloud. We don't want to be somewhere else. I Think there's a cost thing and a compliancy thing involved But West Europe is a heavy used zone so If you want to have I'm sorry, dude. It's not part of the show Evacuate the building. I think it means somebody is a fit happens My time is up Yeah, go down. Yeah. Yeah. Yeah. I've missed mine. It's over stay down. Stay down. Yeah Yeah I'm sorry if you was in the in the audience when you look if you if you look from from from teams or whatever zoom the fire alarm just went off I Have the opinion that you Maybe you wanted to go that might be somebody put a lighter on there. I want to go So what I was stating is that from a business point of view Capacity is not always available. And if you want to have a certain type of machine in the as a cloud There's a limit to what they have Unfortunately, so what we did to get those under calculations engine is try to get all different types and put that in open shift And it worked work perfectly so we didn't put one type of a machine that you can have in the as a cloud, but we Tried different ones and it works. So also that gives some flexibility, but it's a different It's something you have to discuss with your with your business partner because they They don't understand So that's what the availability is The second one is Is that the application of course and from from our perspective probably this sounds like the But the application should be Perfect it should be able to support And scalability Because if it's not it doesn't work I have in my we have an ODF cluster and with with a stateful cluster and there's one application and When I do an update They get all panicky because it did the application doesn't support that so we have to Forcedfully do that during of hours of offers hours to the update With which basically I'm sorry again for my language sucks. It's not we shouldn't do that So you should have an application that is capable of doing that But this also means that the developers developing the application should be Kubernetes aware of scalability and auto scaling options and that's not always the case. So we also have some well, maybe some some a job to do in communicating to them or telling them what the options are and Currently the KPM engine is perfectly suitable From a CPU point of view, but of course there are more options And the Kubernetes event-driven auto scaling option gives you more flexibility to do that and I Was just telling you that we can scale up to a thousand within an hour a hundred within one hour I also tell you that there's a limit to what you can do and This is the reason why if you Scale up it takes time You don't only need to scale up the node or whatever, but the complete cluster has to scale up So that means the logging everything has to scale up that takes time If you have a calculation that takes one hour Complete to run or two hours then after a certain moment It's no use anymore to scale up new machines. So there's a difference Because it takes more time to scale up new machines than to do at the actual calculation So there's a limit to what you can do. Of course what you can do is Say, okay, we scale up five hours before they do the calculation to 2,000 machines, whatever you want But that's not cost-efficient That's that's not efficient. So that's not what we are doing. So that's the limit that we learned and the the last thing we We have done is if you scale up scale down too often It gets Now well, it's not good for the help of the system. So we we put some dampening in it Don't ask me how they did it by the way the guys but what we did is not go like this but make it more smooth it takes about three three minutes delay we have To actually do the open shift Well, turn the shift between scaling up and scaling down a little bit more smoother and it runs more efficient That also doesn't help of course if you want to scale up very fast But for us, that's the the efficient thing to do So that's what we learned and the last thing we're not ready yet Certainly not ready yet. What are we going to do next? Of course, we are currently busy trying to implement arrow Why haven't we done that because when we started arrow was not from a compliancy point of view not suitable for our mayor yet Currently, I think it's it's it's good enough. So we are currently able to use it and the next step for us is to implement it The the thing the developers currently are doing is trying to have the data transfer to the models to finance Models more smoothly via ETL tooling and We are currently I'm currently in conversation with either BI guys data warehouse guys or other Financial models to have them also implemented and what we see is a growing wish to have during of hours of office hours, sorry Be able to To do quick calculations with a lot of power and then scale down again And that's what we are currently doing and of course, we are constantly improving ourselves Basically, what we are doing is of course looking at our processes and do are we still efficient and improving? But I think everybody is doing doing that I think that's it So what we have I think is a solution That doesn't have a traffic jam, but we have a much a lot of roads that we can build up scale down to do the calculation efficiently fast and That's what we did and it runs smoothly without hindrance Business is happy and if they are happy we are also happy Thank you very much if there are questions by the way, I will be here But you can also ask him now, but please not too technical because I'm just a manager No, no, no, no, the data comes from all the systems Yeah, yeah, yeah, just do the calculation and throw the way Yeah, yeah, good question. I know I know I know There's two two reasons and we have currently have one as we also are available in Northern Europe in Dublin But we are scaling it down and eventually trying to get rid of it and West Europe you have three zones From our perspective, that's more than enough To in Middlemeer and one in Amsterdam. So that's where our open shift cluster resides We can't go outside the European Union because we have to adhere to European compliancy rules and that means Sorry Yes, yes, yes, but There's a company company policy that says, okay, we just want to be in West Europe and and I have two less Not enough stripes on my my shoulder to make to do that undo their decision But basically it's a cost efficient and a compliancy issue that we did and we we are in in Dublin But we trying to get that was to I think it was too expensive. I haven't seen the calculation, but I Can I can if you want to know I can show you somebody who can help you with that, but it's not the interesting Other questions I have to close by the way. Now you're asking a tech technical questions. I Don't know, but there's an engineer at the back Maybe he can can answer it for you. And if you come to me, I will show him Yeah, all right