 Good morning. I think we're ready to get started here. Thank you for taking some time out to have have a coffee and listen to me blab on a little bit about what we're calling the cocoa stack. Just for some introduction, my name is Jonathan LaCour. I'm the Vice President of Cloud and Development at Dreamhost. Dreamhost has been around for about 16-17 years now and I've been at Dreamhost for about four. My background is start-ups cloud. I've been in the Python community since the 1.4 days in the OpenStack community since Cactus Diablo and now I run the cloud division at Dreamhost. So what are we what are we here to talk about today? Well we're going to discuss a little about the history of Dreamhost and the LAMP stack which is something that you're probably all familiar about and then we're going to talk about kind of dream compute which is our OpenStack public cloud and and the journey that we took over three four years getting to release that to our customers and we'll talk about the stack that we've built on that we believe is kind of the LAMP stack of the next generation right and then we'll wrap up and talk a little bit about the next chapters. So as I mentioned Dreamhost was founded in 1997 by four students at Harvey Mudd which is a small school in Southern California and one of those founders was Sage Wilde the creator of Ceph and you know we were founded as a web host doing shared VPS dedicated hosting at the time just shared over that period we've evolved grown to 400,000 customers and 1.5 million domains hosted. We've got hundreds of thousands of instances of many open source applications like WordPress is the dominant one that we we host and you know this is kind of all built on the shoulders of the familiar LAMP stack which you know kind of emerged during the 90s and 2000s and it made it you know really easy for us to become a successful business because developers had a really easy way to develop apps right they had an extremely powerful you know relational database right rivaling the really pricey expensive proprietary databases they had great programming languages with a lot of immediacy you could just unzip a PHP app in your public HTML directory and off you go right customers really get it they understand it and on the back of this LAMP stack we've developed you know a platform a hosting platform running on Linux where people can go into the panel they can one click install apps they can develop their own apps and this is the platform that developers were used to for many years well fast forward a bit and and cloud starts to emerge and applications become higher scale you know immediacy is still available but people want more and LAMP stack really focuses on application issues right it's not going anywhere we're still going to be using LAMP for many years but it doesn't address the underlying infrastructure their architecture we've kind of out evolved that some of the big infra issues that that we were dealing with back in the you know mid 2000s and we're still struggling with today obviously everyone is struggling with storage right and shared hosting you're cramming hundreds of people onto single you know nodes right and having some sort of storage architecture that doesn't put everyone you know in a common failure domain that would would be a big problem or just local storage where you know it's very hard to recover from a failure this is a big problem you know we want to be able to scale up and out and you know networking right segmenting out customers is is a huge issue when you're dealing with mass market hosting you've got again we have four hundred thousand customers 1.5 million domains and it's a bit like the Wild West in some regards and this is true for every web host every infrastructure provider out there you need to find a way that on the network you can segment people security it really is the Wild West sometimes I feel like about 80% of the time we spend just updating you know PHP content management systems for people this is like our entire function as a business it's a big issue when when you have all these apps hosted for people on an infrastructure we really need to harden this and have something for the next generation automation scaling up and out making it easier for us to just provision more infrastructure make it available to customers so we need a new stack something for the next 15 years right a compliment to lamp not a replacement so let's talk about dream compute dream compute is our open stack cloud we started the journey about three and a half four years ago very early in the open stack days before open stack was really ready to build a public cloud we were starting this you know effort to build a public cloud an open stack we're targeting a very different market than the typical open stack public cloud we're not going after big enterprise it's not really our our our core market you know we're a mass market hosting company we track a lot of developers entrepreneurs smaller businesses and underserved market in cloud right now with the big behemoths going after you know winning the IT the back-end you know making all of those come into the cloud we really want to get the individual developers and have the have them have a place where they can develop cool new things but we didn't want to just develop this for our customers right we've got a massive amount of infrastructure that we have to support and so we wanted to have a place where we could develop a new stack that they could carry us forward so we want to infrastructure for ourselves so we had some fundamental design principles that we set upon when we decided to build our dream compute so on the storage front we wanted something that was shared and resilient a place where we could scale out and be highly cost-effective we needed snapshotting we wanted to have it be self-healing and easy to manage right so that we didn't have to have this massive team and we really wanted it to be performant but also resilient so this was extremely important to us the most important part was actually available in resiliency not performance security so every tenant we wanted to have them isolated at the layer to from everyone else so this was another critical decision that we made early on and four years ago when you're making this decision there's not a whole lot of choices in the marketplace there's not a whole lot of choices in open source we knew we needed some sort of SDN or virtual networking solution IPv6 we needed it to be IPv6 native you know we are one of the last major independent web hosts we don't have massive banks of IPv4 space available to us and in fact the world is out you know now it's just a bunch of people trading them around you know and it's it's exhaustion is here so we needed to build for the future where you know everyone's you know ring is on got an IP address you know I mean like smart everything everywhere so we wanted IPv6 from the start and and another really critical thing is we built on the shoulders of lamp completely open source we want that for the future for our next stack we can't have it be dependent upon any one vendor we want to be able to build on open standards so no magic no opaque solutions and what do we end up with well the cocoa stack seph open stack cumulus and overlay that's what that all stands for so while the 90s and 2000s were the air of us building up on the lamp stack now we're building on the cocoa stack this is a result of three plus years of building a cloud on open stack and we tested a lot of different things this is what worked for us and we believe would work for a lot of other people and these four components have helped us you know realize all of those fundamental design principles that we set out to achieve so we'll talk about seph first primarily because it was born at dream host right as I mentioned sage one of our co-founders created it really partially because of the problems we were experiencing across our infrastructure a dream host and we incubated it developed it for six seven years and then spun out a little company called ink tank which we sold to red hat last year I guess it's been a while now and that's been a huge success it's really changed the face of storage right and it's all open source from the beginning not not late to the game now we're free now we're open no from the beginning it's been open source so that was very important to us and it's got good performance great if you if you really tune it just just right and it's easy for us to manage with a small team you know our our public cloud we've got multi petabyte clusters and we've got one two people who are managing our seph cluster that's pretty phenomenal right that's that would have been unheard of if we were buying big magic boxes that that we were appliances from from vendors so just dealing with you know spinning rust and commodity servers and fast networking and open-source software we've got this incredible storage cluster it's also cost-effective and it's vendor supported if we want it so obviously we have a good relationship with red hat ink tank you know but there are other people you can go to if you want support that's a big thing for us so the technology is open we can make changes if we need to but there's support available it also supports things like live migration copy-on-write for hyper fast boots you know this is a really big issue right now with with the kind of resurgence of containers I hate to call it the rise of because they've been around for a lot longer than people remember but you know we can boot instances VMs in like 45 to 60 seconds on top of KVM and seph because of the copy copy-on-write technology right we are able to very quickly boot a VM because we're not having to download an image to the hypervisor or anything like that also it's thin provisioned which means that you know while it may be an 80 gig volume right for the boot volume for a VM the odds are is that it's just a thin provision copy-on-write of a standard Ubuntu image right so it's it's saving us a lot in terms of cost efficiency and it's got incredible integration with open stack that's just constantly getting better so just software no proprietary boxes the first first part of the cocoa stack obviously the oh the first oh is open stack and I'm not going to dwell on this much because we're at the open stack summit if you don't know why open stack is is so important you're you know you're probably not listening but for us it was really all about the massive community the momentum and we saw from the beginning that the technology wasn't necessarily there at the time but we knew it would get there and we wanted to be a part of it and ensure that it was part of a stack that we could believe in for the future so we dived in early and we're all in on open stack that brings me to cumulus cumulus networks provides something called cumulus Linux and we were one of the first public customers of cumulus I think we actually outed to cumulus as a business at the open stack summit two or three years ago when we first started talking about what we were doing with dream compute and this really changed the game for us you know it's built on Linux open source wherever possible obviously there's some proprietary silicon that makes it difficult for open source every single thing but for the most part it's it's an open source very low-cost vendor supported platform and it it does two things for us why is this important well first from a capex perspective we're able to go out and buy white box switches from any number of supported vendors and they're all the same to us we develop relationships with these vendors they test them you know it's supported by cumulus and we put Linux on those switches not only that we don't have to buy you know interconnects from the vendor we can buy them from wherever we want to so it saves a huge amount of cost there and then from an OPEX perspective your network goes from being this special case this you know walled garden of you know proprietary CLI's no API's right it's not magic anymore it's just another Linux box on my network right so now instead of having a specialized network team manage my my network my cloud engineers manage it just like they would my hypervisor nodes or my storage nodes it just happens to have you know 48 10 gig ethernet interfaces right we use the same tools we use things like chef collectee Grafana log stash Nagios to to monitor manage provision and automate all of our network this is a huge game changer for us so that's been a big foundation for us at layer 2 on our stack and that brings us to the overlay we actually knew that this was going to be critical for us for you know that layer 2 segmentation for isolation from each and every tenant in the cloud and we also wanted to virtualize layer 3 and up this has been a very tough part of our journey you know we got in this really early we were one of the first customers of NYSERA prior to them being acquired by by VMware and MVP was a great solution for us for a while it became VMware NSX multi hypervisor but as we've kind of grown and expanded out we've really wanted to get into that back that open part here and the choice there has become a lot better recently between you know kind of the advancements with VXLan the the rise of Trident 2 switches which can do accelerated VXLan end cap and decap right in the silicon even you can get Nix to do this as well this has made the the choice a lot better and and now even with some of the hierarchical port binding features that are available now and enabled by Neutron and OpenStack and Cumulus we can actually get a full overlay fully automated with no controller no magic we have to patch our kernel no OVS and all just works it's so much simpler than a proprietary black box so we're big believers in this and a critical factor for us is this Aconda project which just like ink tank that we spun out a few years ago we've spun out Aconda as a project and it's built open from the beginning it provides layer 3 network virtualization and above and it actually does full-on network service orchestration so we can run any network service that we want from routing to firewall load balancing inside virtual machines that are orchestrated in our OpenStack cloud and it interfaces fully with Neutron supports a lot of great things and this again open source go check them out on GitHub they're on Stackforge and you can go to Aconda.io to check them out as well they have a booth so this is another big part of our stack it simplifies our Neutron deployment developed by us from the start and no magic right that's kind of a real critical thing so let's summarize right now dream computers in public beta you can sign up for it today we've got thousands of instances running on this Cocoa stack developer entrepreneur customers primarily unlike any other OpenStack cloud that's our real focus purpose built for that market and it's a proven platform now you know we're showing that this stack can work and and we're big believers in it so why should you choose a Cocoa stack and and join us and and help learn from each other about this well Open platforms you can contribute you can make the the community better we want to benefit and help you benefit from collective kind of improvement in these technologies right there's vendor support available if you need it and from multiple choices right you don't have to go to just one company to get support for these things and over time that'll that'll continue to expand and it's very low cost to deploy so that's the Cocoa stack and what we're doing a dream host a dream compute thank you and I'll stick around up front if anyone has questions