 Okay, so hi everybody, thanks for being here today. My name is Raphael Ferreira and I'm the CEO of In-Avance, which is a French company specialized in open-cloud technology. I'm here today to give you more visibility regarding the financial aspect of OpenStack project. So I'm sorry, Fabia is old, living in Paris, I have a beautiful son. I'm running In-Avance since 2008 with the best partner ever, Nicolas Marchard, which is a CTO of In-Avance and Pierre Mona, which is a COO. What do we do? So we are based in Paris and Montreal. We are focused on open-cloud technology, like OpenStack. We are also core contributor on the SAF storage and Perpet. We are a good member of the foundation. And we are an active contributor to the code. Like for example, we created CELoMeter one years ago for the metering and bidding in OpenStack. We package OpenStack for DBR, we created the Perpet module for OpenStack, et cetera. And in the grizzly release, about what to be the seven biggest contributor in this release. And after that, we provide several types of service like design, build, run of critical platform. So today, we will talk about how you can reduce your IT cost from 30 to 60% by using OpenStack. To illustrate that, let me introduce you to Ed. Hello, Ed. Ed is a 35 years old man living in London. He created AdoptAmen.Love in 2008. And he's a successful start-up. The company grows fast, and there are now 150 people and they generate 50 billion in revenue. So AdoptAmen is a dating website dedicated to women. You know, they can choose a man they want and they put it in the card and that's cool. It's a really website in France, but I have to change the name. Because they want to grow faster and open new branches in several countries, AdoptAmen need to become more agile and more scalable. They decided to outsource their hosting, but they hired an internal team to manage the platform. AdoptAmen have a seasonal activity. They need twice more servers, three months per year. And they also have a continuous increase of their audience, which is very cool for us. This is a typical pattern for this type of activity and for any business website. Here is the current architecture of AdoptAmen.Love. Because there are pure players that decide to split the platform in two data centers in London. They are using standard technology, HP and Cisco for the equipment, Varnish and GINX for the web and the caching, PHP, MySQL, MongoDB for the application and the database, NFS for the storage. And they also started to play with Puppet to automate the configuration of new machine. Here are the costs of the current architecture. They have 80 servers, four load balancers, eight firewalls for NFS equipment, for a total monthly cost of $20,000 per month. They are hosting two data centers in London, Quinix and Teddy City. And they use the one gig internet access for a total monthly cost of $15,000 per month. There are five engineers in the team to manage this architecture with an average of $100,000 per year and per person. The monthly cost to manage this platform is about $40,000 per month. So the global cost, actually, for this architecture is around $75,000 per month. But what's the problem? That's work, but Ed has several problems. To launch a new service, the average length is around three months. And you know in this area, the competition is very strong. So it's a critical problem for Ed. Because they need to install manually new dedicated equipment, the developers have to wait four weeks to have new environment to play with. There is a complete separation between the dev team and the ops team. The dev team have no access on the machine, no access on the operating system. This is a major problem for the dev and the test platform because they spend too much time waiting, building new environment to play with. Last week, Ed was invited to a great TV show on the BBC to talk about Adopt Amen. In only one hour, the traffic increased by 300%. Even if the code is well optimized after 10 minutes, the website failed. Ed felt that he lost 20,000 new potential customers for his website. He missed its fantastic opportunity and degraded his brand image. So Ed has now several motivations and he can summarize his expectations in four categories. Be free of his destiny. Probably most of the most important things for a critical business platform. We will explain this later. Be able to massively scale. Be aggressive in terms of investment. And be interoperable. This word is too complicated for a French guy, so I will tell it in French during all the presentation at the interoperable. Of course, we can tell that there are many other motivations like quality of service, like features. But for me, everything is in this slide. So let's have a look. First of all, you don't have the same expectation depending on your size. If you are a small player on the local market, you can choose between two type of strategy. For a time-to-market reason, you will be looking for a package solution like Cloud in a Box to start fast. And for niche strategy, you will be looking for a specific solution for a specific type of market. For example, Cloud for SaaS provider, Cloud for gaming, Cloud for e-business activity, anything else. If you are a big player on the global market, we need a Cloud framework because you absolutely need to drive your own map. You will be the custom solution to fit exactly to your needs. This is a way following by head for Adopt-a-man.club. OpenStack can address both of these two needs. Being a package solution with, for example, Python Cloud or Nebula or TechOps. And being a framework with specialized integrator like IneVents. As I said, OpenStack is a framework. At IneVents, this is the main reason why we choose OpenStack two years ago. All that you can imagine in terms of infrastructure, you can make it with OpenStack. This is a reason why I'll resume the expectation of Ad in only four categories. Being sure your destiny allows you to build different type of platform. I am Cloud with high performance, high availability, complex network, high end support. Locals Cloud with cheap hardware, flat network, very aggressive price. Cloud storage for block or object storage, massively scalable and S3 compatible. Or HPC Cloud with compute pizza box or bar metal provisioning, rendering workload or specific scheduler. For the question is, how do you see the life? Like this, buy a software and build the same cloud on your competitors? Or this, though being special, being unique increase your creativity. I think you can find my preference. I know Cloud data centers, network, blah blah blah boring for most of us. But you can share that and use a framework to really innovate on this exciting market. The someone expectation is to be able to massively scale. So sure, it's very easy. But sorry, I have no money right now because I start a new activity. I know that because it was my case when I started in advance five years ago. And it was the same thing for Ed. The point is, you must be able to massively scale, but you must be able to start small. Even if you are a big player, it's better to provision the infrastructure progressively to minimize your power consumption and to get the new generation of hardware. And for that reason, OpenStack is the right solution because you don't need to follow unique reference architecture. You can start small or use your existing hardware and evolve with your business activity. So my favorite part, this is a financial part of the presentation, one of the two financial parts where I'm sure you will enjoy my English level in this topic. So you want to build your own private cloud. Of course, we need to be excellent in terms of quality of service, innovation and support, but you also must to be competitive in terms of investment because even if you really want to play with OpenStack, even if it's currently the hype in the IT industry, you must be in line with the market. I think it's not necessary to present AWS in this session, right? They are good, they are very, very good. Personally, I love this company. They innovate, they provide large-scale infrastructure, they are present in several countries and they reduce their price two or three times per year. As you can see, they can provide you a medium-sized VM for 17 cents per hour, around $120 per month. Rockspace, another player, they also ran a public cloud with an aggressive price and a very high-end fanatical support. They are most expensive than AWS, but are focused on quality and performance. A new player in the infrastructure market, Google, you probably never heard about this company, but they can address specific market for compute workload. This infrastructure is based on their homemade hardware and on a very aggressive price. So, as you understand, on this market, there is a real price war. So, is it interesting to build your own open-stack private cloud? Quick analysis. Imagine your target cost for a medium-sized VM is 14 cents per hour. With that price, we'll be in line with the market. For this price, you can propose to your team a VM with two CPU, four gig of RAM, and 100 gig of disk. We estimate that we can run 20 VM per node in a standard period, sometimes more, sometimes less, but 20 is a medium number. Based on that, you need to achieve a monthly cost per node of approximately $2,000 per node and per month. So, imagine you want to build a big private cloud, huge private cloud with the total capacity of 4,000 VM. So, it's a huge one. You will need to build your own infrastructure with 200 physical nodes, 20 by 20 VM by node. Maybe a secondary storage with the block to provide volume to your users, to your team. It can be a NAS like NetApp or distributed storage like CIF. You will need an IN security system keeping in mind that the security is a big part of your investment. So, you must protect yourself from an external street but you also must protect yourself from an internal street and it certainly is the most difficult thing to do. You will need network and node balancing, certainly in high grid line with the one gig and 10 gig for the storage and a minimum of one gig for your internet access. And you also need an information system and several production tools like LDesk, monitoring, backup, automation, stuff. So, we can discuss about the number but based on my experience, approximately price for that is around $80,000 per month. So, now you have your platform, it's great. You will need some rack and transit and you also need some good guys to do the job around 10 guns. So, your global direct cost is around $250,000 per month. Based on the capacity of 4,000 VM, your monthly cost per node is around $1,200. So, to resume, to be competitive with the leader of public cloud infrastructure, you need to target a maximum cost of $2,000 per node and per month. As you see with OpenStack, you can make it. You need to optimize your design, you need to have a good team or to find the good partners, but it's possible. So, what about proprietary solution? What about if you need to buy license, support specific reference architecture to do that? Last thing to understand, if you want to be the private cloud, your customers are everywhere, customer or users. So, be interoperable. Your customers are everywhere. You will need to distribute your application on several platforms. Maybe your OpenStack cloud and other public cloud. Your audience is unpredictable, so you will need to be able to scale fast and it's not sure you will have the capacity on your private platform. So, think hybrid. Be interoperable. Make your life easier. Help your team to easily move the workload between private cloud and other public cloud. OpenStack is the best way to do that. You can provide open APIs, you can guarantee the reversibility, you can be compatible with AWS, Rackspace, HP Cloud, and this is the best way to simplify your team life and allowing them to be more agile and more reactive. So, for Ed, what is the bill now for his new platform? He decided to move to OpenStack and create a private cloud hosting hosted in two data centers. So, it's okay. It was architecture designed by the Adopt-A-Man team. Now, to do that, they took advice through the community and through a fantastic and amazing startup name in advance, I think, something like that. So, this architecture is split into two data centers because this is a business-critical application, as I said. Instead of having custom server by functions, they decided to build a pool of identical servers and configuration for all the platforms. They also decided to give up NFS and to build two cloud storage platforms, one block storage platform for the database and one object storage platform for all the media and unstructured data. For the first one, they choose SEF and for the second one, they choose OpenStack Swift. So, regarding the need to be more scalable and more distributed, they have implemented Sedexys Global Load Balancing, allowing them to redirect all the users on the appropriate platform. It could be their OpenStack private cloud or any public cloud like a no-cloud workspace or WS. So, Ed have big ambition and he knows traffic will increase in the next few months. So, for that, you need 500 virtual machine, distributed on 30 physical nodes. If you need more computer resources, you will use a public cloud. That's it. In terms of storage, Adopt-A-Man need 40 terabytes of block storage and 40 terabytes of object storage. And because all the data are replicated three times, they have to provision at least 120 terabytes for each storage. This is a detailed build for the infrastructure and the hardware. So, depending of your contract with HP, Dell, or IBM, or Cisco, it will be something like that. You have four rack. You have one gig of internet transit. You have a hybrid LAN in 10 gig for the storage and one gig for the web server. You have a global load balancer. This is a SAS solution named Sedexis. You have four load balancers, the open source one. You have 30 OpenStack compute server. You have five OpenStack object storage server, five OpenStack, five step block storage server, and 10 management server. So, with our existing team at Adopt-A-Man, our total monthly cost will be around $70,000 per month. To be more comfortable, you can also add an optional support contract between $5,000 and $10,000 per month, depending on your SLA requirement for OpenStack NSF. So, it's a quick comparison between several solutions. So, on the infrastructure cost, OpenStack wins the price war because it's allowing you to use your existing hardware without specific requirement. The density in terms of number of VM per node is good, 20 per node, 25 per node. In existing platform, Ad had a problem of consolidation. He used too much server because he had dedicated equipment to specific function. And also because this platform has increased with the time. Another problem was that all the server were not virtualized. In comparison with a public cloud like AWS, for 500 VM at a medium price of $150 per month, including transit, snapshot, et cetera, the difference is significant. You have a maximum flexibility, no investment, so it's clearly a good solution, but it's more expensive in the long term. Regarding VMware, I've used this morning the TCO calculator on VMware.com. And in my point of view, the cost of the license is not the biggest cost in the VMware architecture. It's significant, but the main difference is in the reference architecture. To be able to use all the features of VMware, you will need some sand equipment. You will need a five-hole channel network. So for 500 VM on two data centers, you should invest at least twice on your architecture and the license. Regarding the team, it's very interesting because in this case, Ad don't have the critical mass to reduce the number of employees dedicated on the IT operation. If you want to provide 24-7 service, you need at least five-seven engineers. Ad can start with five people because even if the architecture is ready for 500 VM, you don't need to hire now. We estimate that for that size, that size of project, the team is the same for each scenario, VMware, AWS, or standard platform. Even for AWS, as I said, to provide 24-7 service and the application layer, you need at least five people. And in this particular case, you can reduce your monthly cost by outsourcing the management of your cloud. The economy in this case will be 25% higher. So where are the other savings? Between 15% and 50% of savings, just to start is great, but I'm sure we can do more. So this is a current organization of the majority of large companies in France. It's a basic team but we have typically two teams. The developer team and the operation team. I'd like to draw three lines. The first one is a metal line. And there's this line, it's all about data center, physical network and stuff. After we have the operating system line and the application line. Currently, under those three lines, this is the IT Bastion. Don't touch my machine. Don't touch everything. And this is the only way to do things. Whatever you are in the private cloud or in the public cloud, you have to reconsider the way the developer use the infrastructure. This is a way you have to reconsider your team if you want to maximize the value of your new cloud Christmas gift. Because as you know, the profit area is not in the infrastructure side. You have to give more autonomy to your developer and if you want to innovate faster. Keep in mind that infrastructure is just a tool to do something to create value who to develop new business. It's just a tool. We could take several hours on this topic but there are just a few concepts you should keep in mind or to explore to increase the value of your open stack cloud. The first one is the infrastructure as a code. Through the API and by hiring DevOps engineer, this is a man versus machine. The goal is to automate, automate, automate. Second one is continuous delivery. To have a good time to market and days instead of months, you have to be able to release every week or every day. To do that, you need agile test dev platform. So test dev environment ready to launch in a few minutes with for example, heat, our puppet chef Jenkins, et cetera. Design for failure. The infrastructure will fail. You have to deal with that. So explore chaos monkey project and design your architecture to replace than repair. Auto scaling. There are several projects in the open stack ecosystem around that one is incubated in a sailometer for the cloud watch. And you can also work on Chef, our puppet and the API is to achieve the auto scaling. And if you are a global audience, to target your global vendors in several countries and to scale, you need to practice modular design in your software development and also in the design of your architecture. So to resume the session, keep in mind that when you address the financial part of your cloud project, you must work on two areas. The cost of the infrastructure and direct saving. The savings are always pretty amazing in the open stack project. And the profit you can increase by improving your agility and your reactivity. It could be huge. How much would you value having three months of head start on your competitors? 10%, 50%, 300%. Think about that and thanks. If you have question, you have 10 minutes. Go. A question, you'll hold the numbers, it's okay. Perfect, yeah. Instead of Gluster, yeah. But Gluster is not a block storage. It's a file system storage, distributed file system. And we choose, we are very involved in the Chef project because the design is pretty amazing to be able to have object storage and block storage maybe after NFS in the same cluster and to allow us to manage this cluster depending of the customer needs. And in terms of performance, Chef is pretty amazing too. Is that a question? Yeah. I saw nothing, so go. I think we will release some performance testing in the next few weeks. On our website and with Ink Tank, we have a big Chef cluster, actually in France. And we develop some performance tool homemade by using some tools like a Tung or open source solution like this. And I think we will publish some results in the few weeks. Yes? Yeah. In your experience, is that a realistic? For a startup, yeah. Not for a large company. This is one of the biggest problems we have currently with some of our customers. They have an IT team which manage physical machine, physical network, et cetera. And they have to re-sync their job and it's one of the most difficult things to do, actually, is our large customer. That seems to be right. Are you selling a product that depends on a skill set that simply doesn't work? Yeah. Which is good for us about... Yeah. In France, like I said, in another session, we have the, I don't know if you are familiar with the syndicat, the syndicat, of course. It is a very powerful organization to protect the right of the employees, which is very, very good. But it's time to change anything in a company in France. So we don't have training for this kind of change management. It's just the beginning in the Europe and as I said, the problem we have right now to accelerate the development of OpenStack in Europe is not a technical problem. We have the technology, we know how we can implement the technology, it's okay. It's about people. We are not ready yet. This is... No. So, Chmuel is one of the senior developers in advance, he's a core developer on Swift. Okay? Another question? Okay. Thank you, everybody.