Live from the Summit: Build scalable infrastructure with Red Hat Enterprise Linux OpenStack Platform

Will Foster, Dan Radez, and Kambiz Aghaiepour–all senior engineers with Red Hat–wanted more automation in their environment. “What we did not want to do was be in the business of manually managing the building of [OpenStack] clusters,” Aghaiepour said. That’s a common problem for enterprises–or anyone–thinking about performance benchmarking and scalability testing on OpenStack. But it was also an opportunity.

Before too long, they had 9 racks with 200 baremetal nodes running Red Hat Enterprise Linux OpenStack Platform 4 (based on Havana) on Red Hat Enterprise Linux 6.5. They used Foreman 1.5 (part of Red Hat Satellite 6) for node provisioning and hostgroup-driven OpenStack deployment. Other tools or technologies used included:

  • OpenFlow 1.1
  • IPMI (intelligent platform management interface)
  • Nova Compute
  • Neutron networking
  • Controller
  • OpenStack storage (GlusterFS)
  • Puppet
  • Staypuft (OpenStack Foreman installer)

Recommended best practices were around utility services and configuration management. For services, the group administered Puppet, PXE, DHCP, and DNS through Foreman, which keeps things in one place and eases administrative sprawl. For config management, they used Puppetmaster through Foreman and distributed revision control systems. “Anything we do once or twice, we never want to do again. We automate it,” said Foster.

Foster also recommended doing as much as you can through Kickstart %post. Foreman, he said, makes this easy. They also used Linux software RAID. Foster noted that most modern CPUs can handle RAID overhead. His final bit of systems design advice: Keep it simple. Use shared storage for important stuff.  Nodes should be a commodity–it should be faster to spin up a new one than to fix an old one.

Aghaiepour demoed the automated environment–showing off to the packed room how quickly (and easily) a 70-node cluster could be erased, redefined, reprovisioned, and deployed. The whole process took less than 10 minutes. He also demonstrated how the instance exported data to a calendaring tool. This calendar keeps track of the node types that are going to be present in the cluster, so that engineers can figure out when they should plan to test applications or services that require a particular node type.

Want to try setting up a scalable infrastructure of your own?

> Watch the demo: Automated OpenStack deployments with Foreman and Puppet

> Get the provisioning demo tools and scripts from GitHub.

This kind of tooling and DevOps work is intended to help engineers get their jobs done.  IT no longer has to provision servers, or even set up VMs–with enough careful planning and the right automation tools, users can spin up and take down their own instances. And the instances themselves can be responsive to the intended use with proper APIs, groups, and settings.  No new hardware needed.

More information

 

Event: Red Hat Summit 2014
Date: 10:40 a.m., Wed April 16, 2014
Type: Session
Track: Application and platform infrastructure
Technical difficulty: 3
Title: Building scalable cloud infrastructure using RHEL-OSP
Speakers:
Will Foster (sr systems engineer, Red Hat), Kambiz Aghaiepour (principal software engineer, Red Hat), Dan Radez (senior software engineer, Red Hat)