True Cloud Technology

Capturing the cost effectiveness of true cloud requires a hyper-converged architecture, such as Null-Tier™, but in order to achieve that architecture, you also need a multi-server runtime environment. Piston OpenStack achieves this with a system called Moxie Runtime Environment™.

Intelligent Systems Management

Most infrastructure services can be trivially run on every server. But some components, typically load balancers, queue servers, database servers, and other data stores, need to run a single instance at a guaranteed location. While typical configuration management approaches address this using a single “master” server configuration profile for the database, with “slave” profiles for one or more HA failover servers, Moxie RTE™ uses a master election approach: at runtime, any one of the nodes in your cluster is elected to operate each service. If the process, the network connection, or the hardware in that node fails, the service is seamlessly re-elected to run somewhere else.

No More Split Brain

While there are a number of solutions that provide master election, what makes Moxie RTE unique is the way that it orchestrates “strong fencing.” In order to guarantee that your cloud avoids “split brain” scenarios (where more than one node tries to operate a critical service at the same time), the Moxie RTE master election mechanism uses Zookeeper as a distributed consensus voting system. This means that if your cluster is chopped into smaller pieces by something like the failure of one or more network ports or switches, only the portion of the cluster that contains greater than 50% of the nodes can continue to “vote.”

In addition, to guarantee that a node that fails catastrophically cannot hold on to storage or network resources that are required for that service, Moxie RTE will terminate the node using the out-of-band IPMI connection over a physically separate management network. (The official term for this is STONITH – which stands for Shoot The Other Node In The Head, and is one of the precursors to the now-infamous “Puppies and Cows” analogy).

Add & Remove Capacity Without Downtime

Master election and strong fencing are already a powerful combination, but they’re only the starting point; Moxie RTE also implements two separate state machines (technically, non-deterministic finite automata), one for node state and one for service state. This allows Piston OpenStack to treat the full set of dependencies between services as a state machine, and orchestrate upgrades across dozens of processes and hundreds of nodes. It also manages the scale-in and scale-out processes, allowing operators of Piston OpenStack environments to add and remove capacity without a maintenance window.

The First Truly Complete Runtime Environment

Many of these technologies (master election, strong fencing, systems-level state machines) have been applied to infrastructure before, but Moxie RTE is the first time they have been combined into a complete runtime environment. For our customers, this means less time managing cloud infrastructure, and more time working on stuff that matters.

Video

Live Demo on Valley View

Watch a live demo of Moxie's high availability capabilities (11:18)

Web

'Welcome to Continuous Infrastructure'

Read more about adapting to the era of scaling-out, by Piston CTO Joshua McKenty


Plan Your Pilot

Ready to plan your deployment? Click below to view hardware specs, access reference architectures, and plan your Piston OpenStack deployment, down to the last detail.

Start Now