 Good evening, my name is Kobe Olzer really excited to be here tonight. I Lead the DevOps group in a live person You're going to hear today about our great experience with the open stack How we change the way live person consider and decides on its technology how it took us to the next level in terms in terms of Infrastructure and cost efficiency To the tech guys here. I'm going to talk about some of the tools that helped us Maybe something of that you can find they're useful for yourself as well So taking you through the agenda for this talk I'm going to introduce you to live person a short introduction just to get you in the context of what live person does and how it applies on the open stack Going to go through the story how it was in the old days and how it is today Going to talk about some interesting stuff. What's cooking in our lab and Finally some tips and tools that you might find a useful Want to start with a short introduction about myself? Started my career in 99 being with a live person since 2012 have married with three kids and Love everything to do with virtualization automation big data and beer So live person live person been around since 1998 Sus company even before this word was invented. I think it was called the ASP before We have a 8,500 customers eight out of ten a fortune 500 and our mission statement is to create Meaningful customer connections through our a rich a multimedia platform, which I'm going to show you an example in the next slide You can see the real Verizon site. You can see a customer lingering in the checkout page Not sure if we want to go through this is the buy our smart system detects that they haven't maybe some doubts so a Windows pops with a invitation for a live chat with a real Verizon representative and if the visitor accepts then our rich multimedia SAS platform with a voice video and Chat a taking for the sale Increment in sale reducing bounds and helping the customer through the through the process Okay, so before we start a quick Show about how a live person infrastructure looks in a nutshell So we have three tiers the first tier is the web tier like any other a normal SAS company in the web to we have the Apache and the HHA proxy and we have some IIS servers for the legacy systems In the uplayer we have the mainly the Tomcat and the JBoss. It's all about the Java and the Scala for us And in the data store, we have a bunch of them We have a Cassandra Vertica couch base my sequel Oracle and a dupe especially proud of the addupe. We just crossed the one petabyte So as you can see the open stock is used for us For the first two layers, which are our non-persistent layers. I'm going to talk about what is non-persistent for us in a second Okay, our use case when we started this junior and a half ago We knew that we wanted to go as much as virtualized as possible so We had the minimal requirements The first requirement was to find a solution for our non-persistent servers such as the web tier and the up tier and The what by saying non-persistent I mean that those instances don't need to have any data that we need on them. They can go down We can bring another round up. So There is no value to the data there The other use case the other minimum requirement is to be able to scale to thousands of VMs with commodity servers We had a lot of physical servers and we wanted to reuse those servers with the open stack And the last one we wanted to transform our infrastructure to platform as a service style so Before before we started with the open stack If we wanted to deploy a new service, we needed our operations To work with a VMware or physical servers It took a lot of time and money and then we got a new service Up and running It's not that the VMware is the best solution on the contrary But when we started when we wanted the when we knew that we wanted to go to As much as virtualized as possible We had a very small footprint of VMware and they're looking at our minimal requirement and The fact that we had a lot of physical servers Already in place the open stack was the best cost-efficient option for us So today we were managed to bring our infrastructure To become a building block very easy with the rest API Scripts using puppets and open stack. We are able to build as a cloud of Services in a matter of minutes and even maybe seconds. We have new services running almost new services every day So in the old days as you can see here We used to have a most of our servers physical and some a footprint of a VMware with a none, of course open stack and today most of our servers are a Open stack some VMware the same didn't change and the physical servers which are mostly for the data store And of course for the open stack hosts Okay jumping in into the Components and architecture as you can see here. We mainly use Our KVM is the hypervisor. We use my SQL as our database our distribution is a Ubuntu distribution and we started a long time ago with a Diablo distribution and Moved from there started went to a production with the Essex and today we are running a Folsom and Grizzly on the staging environment We use local disks like I mentioned before we use Our services as a mainly as non-persistent. So local disks work for us We are looking at the solutions for a distributed storage such as a safe, but it's a Will come only later next year We are a managed we are managing everything with a puppet and to end and Starting from a from bringing use new servers up to new services up in production up and running and ready to serve We have a for production data centers two in North America two in Europe One staging and another one in a dev to two other small data centers This is a real snapshot from our production Open-stack as you can see here. We mainly use a the Sentos 6.4 and the the windows the 2003 images and in flavors we have a lot of flavors, but we use mainly the Two cores and four cores with the four gigabyte run It's also possible to of course to use a much bigger flavors like you can see here We also use a 16 cores and the 15 gigabyte of from but So we don't need to be afraid and they can use a much more if you need Okay, so this graph is very interesting You can see the whole life person story. In fact, you can see that we started in mint 2012 built our confidence slowly since then and You can see that the in a me 2013 We got a decision where we are confident enough with the open stack So we moved everything else to open stock with P2VD with V2VD Everything we had left to open stack and since then just continue growing with the open stack We have a 40 we have a 1400 instances with more than 6,000 cores Okay, another interesting stuff with the open stock that we do and we use it for our QA and dev environment so In our lab as you can see here We are planning to a provision with the knock Using foreman foreman is a one-stop shop application for us to provision your services to provision your instances To use it with a to provision new instances even in easy to and to deploy new services using puppet Another very interesting tool for us is them collective of course like we are puppet users So and collective come naturally to us with them collective. We orchestrate Our clusters we use the the plugins out of the box with them collective to manage the the data centers But writing our own plugins to each service We use it to orchestrate the the clusters Another very interesting thing that is going on in the lab is Auto scaling we are looking at the heat and cilometer To and also scale out which is not written here to a auto scale It looks very promising and of course it's it's part of the of the day of the Havana distribution Which is going to be much easier to implement So like I mentioned The open stack is also used in our R&D environment. It's a it's a It's very dear to our R&D because they They can get their own quota to provision VMs. They can set their own private load balancers They have the same keys and metadata deployed on all services and then don't need big budgets to for and settle for few VMs Like they used to do in the past Okay, some tips and tools as I promised Okay, the first one I get to us I get this question a lot. How did I get my management organization to support? so It's a we started very small we started with a couple of servers in a QA and dev environment and Then we took some servers into production for servers that they Were due to decommission we started a new cluster and we started very very small with Not important services like a maybe some alpha services for new For a new services so Starting small beating our confidence slowly but surely Got us until here and of course the the cost efficiency Was it was a big part of the picture when Talking to the management and getting them to support us another tip is like I mentioned before the foreman Combined with puppet and then collective we have a good experience with it in the lab It's a I'm not going to go through of course each technology right now, but Using the foreman To for a one-stop shop to provision new services to deploy to for a puppet dashboard it's a it's a proven to be a Very good asset and last when if you are thinking of starting a new deployment with the open stack Or even you have an existing one. I would suggest looking at the filter scheduler for the soft provisioning It's something that they proven for us Very useful to to look deep into the soft provisioning and figure out exactly what fit for us How to a soft provision because it's not a straightforward you can you can of course out of the box use a memory This space and the CPU soft provision, but you can do it with a Much more Details for your organization and your needs, okay to summarize Like you saw before one a open stack is is a running It's a core business on 100% of the services at least one server and up to 100% And it's also being used in our staging dev and QA environment So if life person could do it, I think that many more companies can do it as well If there was one thing I want you to remember from this session is is this and Hopefully Our use case life person use case can convince most of you to try open stock in production and enjoy the cost effective solution Thank you very much