Architecting the Cloud

The array of possible hardware and software combinations that can be used to create an OpenStack cloud is pretty amazing at this point. A phrase we typically hear these days is that having an integration for OpenStack is table stakes for a hardware or software product coming in the market. As of the Queens release of OpenStack (March 2018), there were 75 Cinder storage drivers and over 40 Neutron network plugins. These integrations cover a wide range of products from traditional EMC storage arrays and Cisco switches to various software-defined storage and networking products. OpenStack supports a number of hypervisors and compute platforms, ranging from commodity x86 hardware to IBM Z-series mainframes.

A few of the decisions we make as architects affect the bottom line as much as hardware and software selection. Although we approach the deployment of our cloud or the development of our software in an iterative fashion to reduce the risk of mistakes, it is typically very difficult to iteratively purchase hardware.

Luckily, OpenStack ships with a set of reference plugins for each of the subsystems that we can use to mock out our cloud before purchasing actual hardware. For Cinder, the reference implementation involves using LVM and iSCSI on a Linux system. For Neutron, the reference implementation uses Open vSwitch and iptables on a Linux system. Even Compute can be mocked out using nested virtualization with KVM. The approach that we recommend is to begin an OpenStack project with the reference plugins until you get a feel of how the software and hardware will work together before making any large hardware or software purchases.

In this chapter, we will cover the following topics:

• Picking an OpenStack distribution
• Computing hardware considerations
• Network design
• Storage design
• Expanding the initial deployment