Post Google I/O 2016, an exploration of Google's open source strategy as a feeder to their cloud strategy. Focus on Kubernetes, TensorFlow and Parsey McParseface (sigh)
2. Google’s cloud strategy in a nutshell
Build new capabilities to support the business
Open source the tech behind these capabilities as a clean rework, managed as a
public project (not a line of business)
Let a standard / ecosystem develop around the tech
Have a great implementation for hosting systems built on the standard
7. The OSS: SyntaxNet & Parsey McParseface
What it is:
● a way to teach computers to manage the ambiguity of human conversation
How it’s used:
● build a model of the way humans speak about a subject
● build a model of an individual speaker
What it enables:
● highly accurate conversational parsing
● the ability to quickly identify missing information and only ask for that
8.
9. What standard is being developed?
Increased demand for “bots” to be conversational
A model for processing written texts, acting as an example for driving new models
A way to build new models -- SyntaxNet, built on TensorFlow
10. How do you help engineers model users
and make predictions about them faster?
11. The OSS: TensorFlow
What it is:
● A framework for defining, executing and training machine learning models
How it’s used:
● Build predictive models of complex human behavior, such as preferences,
schedules, speech patterns
What it enables:
● Interfaces that adapt to users, cultures and improve over time
12. What’s a tensor?
A multi-dimensional data structure, used
to hold transformations of a data set.
13. What's a flow
Programs that chain together operations into a directed
graph. Flows describe the movement of data and
operations but specifically don't imply an execution
technology or order -- making them ideal for scaling in
to run on a laptop or out to run on a cloud.
14. What standard is being developed?
Engineers focus on building ML models using standard tools.
Models are tradable.
Visualizations & tools arise to improve TTM
18. The OSS: Kubernetes (K8S)
What it is:
● Container orchestration & cluster management
How it’s used:
● High leverage primitives produce addressable services on top of Linux nodes
running Docker (more OS & container format support coming!)
What it enables:
● Management of services at a high level of abstraction.
● Automation of scale out, failover, advanced deployments
19. New in Kubernetes 1.2
● Persistent volume provisioning support
● Rolling deployments
● Cross-zone failover (“Ubernetes”)
● TLS support & load balancing (“Ingress”)
● Autoscaling
● Graceful node shutdown (patches…)
● New documentation website (with very good docs)
20. What standard is being developed?
Engineers have a way to construct scalable, firewalled services using docker
images and configuration
Scaling, rolling deployments and failover are built in
Short-term statefulness provided through load balancing
Stateful containers aren’t as much an anti-pattern as they used to be (even less so
in K8S 1.3!)
A new class of technologies -- Cloud Native Computing -- and a Foundation to
support their development (the CNCF)
21. ...I’m not good at ending slideshows
Apprenda has recently acquired a Kubernetes company (Kismatic) and will be
offering an Enterprise
We’re working hard to bring Kubernetes to Windows Server Containers (don’t
laugh, it’s working)
Lots of exciting Enterprise PaaS and Container Orch stuff coming!