chain Team Structure Architecture Contention Coherence (of your organisation and processes, not just your software!) CI/CD DevOps Containers Microservices Cloud How fast can you deliver software? + How fast can you respond to runtime changes? Orchestration
[able to be] 100% on the public cloud (any cloud). It’s just obvious. But we can’t be only on one public cloud…” - Don Duet, Head of Technology, Goldman Sachs Typical enterprise target: AWS + 1 + Data centres
move aggressively towards public cloud with confidence. “We’re attaching to massively funded cloud projects left, right, and center...” - Scott Johnston, Docker
efforts are open and standard ◦ Partnered with Mesosphere and Docker on Azure rather than develop own schedulers ◦ ACS is Linux-first, with Windows coming ◦ Windows support is being done by contribution to Docker OSS “This is important to customers because in the fast-changing cloud ecosystem they want open systems without lock-in…” - John Gossman, Microsoft Azure Architect
Change 2. Private to Public 3. Proprietary to Open Scale of the challenge: • 8K technical employees • 5K applications • 75K database instances • 39 PB data (59% growth) • 165K servers Vast majority already on internal cloud infrastructure Source: Don Duet, GS European Tech Founders Summit May 2016
Hide underlying platforms (AWS, GCP, Azure, and vSphere) • Flexible deployment topologies • Local development • Decouple packaging and topology Why kubernetes? • Needed to be universal, not just AWS (e.g. ECS) • Liked the pluggability of Kubernetes (logs, volumes, network, …)
containers to basic kubernetes was incredibly smooth. It took about two weeks and we had a full stack running… All-in-all it was about seven weeks to get fully-integrated with kubernetes for volume management, service handling, log shipping, monitoring.” - Rob Harrop, CTO
right through to production. The same containers can be arranged in clusters of varying size on platforms ranging from AWS to vSphere. We've isolated all the platform-specific bits into the thin layer that is the K8S cluster. Even their K8S tooling makes cluster config easy.” - Rob Harrop, CTO
cloud, focusing mostly on resiliency to failure and evolvability. We also wanted an active, preferably open platform. Kubernetes and Docker Swarm rose to the top of the list… Kubernetes appeared to be geared for large scale deployments and have better engineering around ops. Docker was more geared towards developers…” - David Dooling, Atomist
hospitality, media and entertainment, and other industries have embraced Docker and committed to going into production. We built Amazon EC2 Container Service (ECS) to be the best place for these customers to run Docker in production securely and at scale.” - Deepak Singh, Amazon ECS. • ECS is a result of the traction the AWS team were already seeing • Container management, scheduling, and deep integration with AWS platform • Examples: Linden Lab, Empire
a few weeks • Serious usage already - sustained large deployments • Usage going up every day, not dropping at weekends “The team is also getting far more inquiries that we can handle quickly, so interest is higher than anticipated” - John Gossman, Microsoft Azure Architect
most of its services, including the storage services, to run on top of Mesos” - Matthias Eichstaedt “Service style applications, batch jobs, and stream processing alike, from a variety of use cases across Netflix rely on executing container based applications in multi-tenant clusters powered by Mesos and Fenzo.” - Sharma Podila, Netflix
and Docker come up together. Customers tell us that they value the scale and agility provided by the AWS platform when coupled with the developer productivity benefits of Docker’s dependency management and deployment capabilities.” - Deepak Singh, Amazon ECS
bottoms-up adoption in enterprises” - Scott Johnston, Docker 1 2 3 LAND: lift-and-shift existing app (no microservices), adopt CI etc. 6-9 months PRODUCTION: getting the first app from there into production, about 6 months PLATFORM: floodgates open and big re-platforming project begins... 12-15 months
in one data center EXPAND 1: several apps within one data center EXPAND 2: bridge several data centers “[This customer] wants to leverage existing data centers (each of which have thousands of nodes) while increasing their cloud footprint. They will migrate everything to Kubernetes running everywhere over the next two years.” - David Aronchick, Google
all about apps, not containers... Once the platform is in… • Repeated experience: 500+ devs, 1,000+ apps in first couple of months • V. low ops overhead: e.g. 1500 apps onboarded in 6 months, < 2 people to run
(May) • CoreOS $28M Series B (May) • Rancher Labs $20M Series B (May) • Pivotal $253M Series C (May) (+ $400M debt/equity swap) • Sysdig $15M Series B (April) • Docker $95M Series D (April) • Mesosphere $73.5M Series C (March)
simple “let’s run containers!” to “we’re building apps… and of course we’re using containers.” - John Gossman, Microsoft Azure “We’re seeing a shift with many customers starting with the application, not the infrastructure.” - Deepak Singh, Amazon ECS
◦ By definition, not tied to any one cloud vendor • An important ‘cloud’ is the developer laptop • Applications may run in several different environments as part of the development lifecycle ◦ Even if production migration is rarer • Need to consider not just compute, but also networking and storage ◦ Data gravity vs data agility
Pipeline Value chain Team Structure Architecture Contention Coherence (of your organisation and processes, not just your software!) CI/CD DevOps Containers Microservices Cloud How fast can you deliver software? + How fast can you respond to runtime changes? Orchestration
what services depend on what other services is something everyone seems to be hand rolling • Even basic things like knowing who to talk to about a service is difficult in larger organisations • Very little is being done to help developers, outside of platforms that make things easier to deploy • Data is still problematic - replication, and moving process to data