VMware's Project Zombie gives dumb servers BRAAAAINS
Virtualization company reveals 'special sauce' for upcoming cloud
VMware has taken the wraps off the gigantic task-automation system that will marshal resources in its upcoming infrastructure-as-a-service cloud.
"Project Zombie" was unveiled at PuppetConf 2013 in San Francisco on Friday by VMware's infernal zombie commander (officially: Automation Architect, Hybrid Cloud Services) Nicholas Weaver, who said the technology has cut the time it takes to do a cloud deployment onto bare metal from 72 hours, to one and a half.
"We're going to expose this service to anyone around the world" via VMware's upcoming IaaS cloud, he said.
Project Zombie was developed after VMware took a look at the major clouds of Microsoft, Rackspace, and Amazon, and realized that each of these companies has developed "their own flavor" of advanced, proprietary technology to underpin the systems. Project Zombie is VMware's own "special sauce," he said, describing it as "an automation platform that can scale out and do amazing things."
The technology will underpin VMware's upcoming vCHS (vCloud Hybrid Service), which was revealed in May. Project Zombie uses the Puppet data center automation technology, along with the distributed high availability database Cassandra, which was developed by Facebook and is now an Apache project. Other technologies to go into the tech include RabbitMQ and JRuby.
In his speech Weaver gave details on two key modules within Zombie, named Rez and Engine.
Rez is a globally distributed resource management system built around a scale-out system, he said. It has distributed locking and uses a bare-metal provisioning tool co-developed by Puppet and VMware-parent EMC named Razor to push data up into Rez, which can output data and be controlled by a complete RESTful API.
Engine is the tool that runs global puppet jobs, Weaver says, and uses a VMware-developed language named Zombie Engine DSL (ZED) for writing these major commands.
"If I have 48 compute hosts and want to change one thing of each of those compute hosts and it takes two minutes apiece and I want to do all 48 at once, what ZED allows me to do is ... just push it out," he says, describing how ZED lets him say in a single line "do all these things concurrent, wait, then do all of this."
ZED code can be loaded into Project Zombie's Engine, where it shows up in a library as an action that can be called from an API, Weaver said. This API kicks off a chain of events that sees the commands go through to processors which take the work and break it up across multiple computers, and then into brokers which can be used to create or remove bottlenecks to have a predictable automation flow.
So far, during the design of Project Zombie, VMware has built a multitude of in-house modules for Puppet which it plans to publish as open source in the future.
Project Zombie has already made radical differences to how the company creates clouds, Weaver says. "Before Zombie existed it was a factory floor. ... we had a bunch of operations guys and they sat down with a 55-page manual," he says, saying a typical cloud build can have several thousand distinct configuration points. In one example cloud build, he said, the use of Zombie "turned four to five days of human effort into one hour of hands-off [automation]".
Given Vmware's late entrance into the market, and the spin-off of advanced technical assets like Cloud Foundry into the separate company Pivotal, the virtualization company needs its vCHS service to work, and with tools like Project Zombie the company may be able to put its hypervisor knowledge to use. ®
Sponsored: Benefits from the lessons learned in HPC