Structure 2014 - The right and wrong way to scale - Rackspace

Editor's Notes

• #3: two videos URL? Zach will come out for comments in middle of his presentation
• #4: Thank you, ____________________
• #5: I’m going to talk today about an advance in cloud technology and services that will change the way the world thinks about scaling production applications.
• #6: I’m excited to be here with Zack Rosen of Pantheon, and with video appearances by a couple of other leading innovators in our industry.
• #7: So what’s on their minds — and the minds of other leaders of fast-growing Internet companies? What’s not on their minds is the unit price of compute. Instead they are concerned about more significant costs:
• #8: [FAST]— One is the cost of dealing with retired instances, which can disappear on fairly short notice. [CLICK to start flameout animation]
• #9: Another is the cost of noisy neighbors. That’s not just a price that Seth Rogen pays for living next door to Zach Efron’s frat house. Noisy neighbors force cloud users to deal with inconsistent behavior and performance on networks and disks.
• #10: Here’s a simple readout. The multi-tenant cloud is great for getting a business started quickly. But as a business gets big and tries to scale on the multi-tenant cloud, the result is inconsistent behavior, poor performance and variable availability.
• #11: These pain points are inherent in the multi-tenant nature of today’s public clouds — including ours. They are driving massive increases in application complexity and computing costs for users who are big — or are getting big. Those costs are rising even amid the much-ballyhooed price cuts for units of compute by major cloud providers. Infrastructure isn’t the only cost to factor in.
• #12: This rising complexity and cost on the multi-tenant cloud is hitting users in four main ways: [CLICK] 1. They spend more on engineering time and talent to architect for failure on the multi-tenant cloud, which is complex and hard. [CLICK] 2. They also spend more on engineering to deal with inconsistent performance, which is even harder. [CLICK] 3. They spend more on infrastructure, because over-provisioning is one of the major ways to compensate for inconsistent performance. [CLICK] 4. They spend more through the virtualization tax, which can diminish network and disk performance by 5% to 20%
• #13: Here’s what many companies face. [CLICK] In the beginning, the multi-tenant cloud is great for prototyping and growth. [CLICK] But eventually as growth accelerates and they need to optimize, they start thinking colocation.
• #14: They like the elasticity of the multi-tenant cloud, but they know it’s not simple to scale. [CLICK] If they move to colocation, they get simpler scaling, but they give up elasticity and agility. And they take on a lot more IT work.
• #15: What’s striking is that they — and we — all want the same thing. We want the simplicity, consistent performance, and predictable cost of colo. And we want the elasticity of the multi-tenant cloud. Well, I am here to tell you: you don’t have to pick between the two anymore. Rackspace is announcing today the limited availability of a new product... [CLICK TO NEXT SLIDE]
• #16: ....called OnMetal. OnMetal offers bare-metal servers that users can spin up and spin down as quickly as VMs — in just a matter of minutes. These servers are accessed through an API, just like VMs. But unlike VMs, these servers are highly specialized for the workload you’re running. There’s no sharing of metal. And there’s no hypervisor.
• #17: OnMetal is entirely open source. Everyone here is free to see how it works and improve upon it. It’s built on OpenStack software and OpenCompute hardware, leveraging two open-source projects that Rackspace helped to launch. The API for OnMetal is not just any old proprietary API. It is the OpenStack Nova API. So you don’t have to worry about vendor lock-in.
• #18: We designed OnMetal around the same principles applied by many of the world’s leading Internet companies. They don’t use VMs or off-the-shelf servers for their core operations. And neither should you, if you’re big or plan to get big. Our aim -- with onMetal -- was simplicity at scale, with a high ratio of performance to cost. Here’s how we achieved that goal:
• #19: [FAST] All of the servers are solid-state. There are ZERO moving parts.
• #20: [FAST] The fans are detached from the chassis and can be serviced without stopping the machines. That means no heat. No vibration. This greatly increases the mean time between failures.
• #21: [FAST] We believe the network should have consistently low latency. That’s why we’re arming OnMetal cabinets with 10-gigabit networking and minimal network over-provisioning.
• #22: The configurations are highly optimized for specific workloads. We have servers optimized for databases [CLICK] low-latency caching [CLICK] and for handling web requests. Specializing the gear helps keep costs under control.
• #23: Here, for example, are some of the sexy specs for our OnMetal server customized for users who are running data stores like MongoDB, Cassandra, MySQL, and Postgres. IO is the heart of a database — and one of the areas where we see the toughest customer challenges in the cloud. We designed this server to crush IO and deliver exceptional performance and value.
• #24: You may be asking yourself: why is Rackspace is offering this single tenant cloud in a bare metal form factor? What about virtualization?   The answer is this: if you’re running a big and rapidly growing Internet application, virtualization is not the way to move forward. You’re not interested in fractions of a machine. And the virtualization tax on performance starts to matter as you get big.   So when you no longer need to share a host with other users, why do you need virtualization? Maybe for ease of management? Not anymore. We see a new crop of tools, like Docker and CoreOS, taking over.
• #25: The final element in the formula for successful scaling involves your most-precious asset — your engineering talent. As you grow fast and get big, you need to think hard about which IT tasks to handle in house, and which ones to buy as a service. If you work with a company like Rackspace, you can leverage scaling engineers and devOps engineers to run these OnMetal machines and other components of your stack. [CLICK] This way, your company can stay fast and lean. You can focus on building your app, instead of swelling your payroll with a lot of engineers to run IT that doesn’t differentiate your business. The fast, lean shop in this picture is Pantheon.
• #26: I’d like to invite Zack Rosen, the CEO of Pantheon, to join me on stage. Pantheon is a great use case for what OnMetal can do, because scaling customer’s applications is at the heart of its business. Zack, please tell us: what does Pantheon do for its customers? Why does Pantheon use bare-metal servers rather than VMs? What will instant provisioning of bare-metal servers mean for your business, and for others operating at scale? Thank you, Zack.
• #27: Now let’s hear from a leader in container technology. Here’s a quick video of Ben Golub of Docker.
• #29: Let’s hear now from Alex Polvi, founder of CoreOS, a software guy who is excited about the possibilities of instantly-deployed bare-metal hardware with an API.
• #30: [Core OS video]
• #31: As you can see, these industry pioneers share our view of the right and wrong way to scale cloud applications. [PAUSE]
• #32: In conclusion, let me admit to something most of you already know. Rackspace is not a hardware company. We are a service company that listens closely to its business customers and works to make computing simpler for them. We are proud of OnMetal because it addresses a big pain point that we heard about from customers. We’ve designed OnMetal in an entirely open-source manner meant to encourage emulation and improvement. As it catches on, we believe that the technology behind OnMetal will address many of the issues — around isolation and control and security and compliance — that until now have kept many companies running IT in do-it-yourself DCs or in colo facilities. It will also make it easier for innovative companies to grow. That will be good for all of us, in different ways. Where we at Rackspace will differentiate ourselves is by providing managed services on top of OnMetal.  We will help fast-growing companies scale their most-precious resource — their engineering talent.
• #33: Speaking of which — we invite all the engineering talent present here to test drive OnMetal, at this URL. We welcome your feedback. Once OnMetal becomes widely available, starting in late July, we think that users at scale will find that it’s just what they’ve been looking for.... elastic computing that they don’t have to share. THANK YOU. [Pause for applause.] With that, I’ll open the floor for questions. Please step up to one of the mikes on the floor.
• #34: Thank you, ____________________