I’ve been playing with Openstack off and on since Grizzly. I decided it’s time to go back and redo my lab and install Kilo on it. I’m using a mix of hardware from my Grizzly install and other pieces I have lying around.
Acer 1820t i3 laptop with 6GB running as the imaging and configuration management server.
Acer 1820t i7 laptop with 6GB and a 100mbit ethernet USB dongle running as the network node.
Thinkpad X230 i7 laptop with 16GB running as the controller node
White Box’d i3 Ivy Bridge with 32GB running as a compute node
White Box’d i7 Ivy Bridge with 32GB running as a compute node
White Box’d i7 Sandy Bridge with 32GB running as a compute node
4 BeagleBone Blacks running as VPN end points
All the boxes have SSDs. The compute nodes also have WD Black 1TB drives for instance storing. The controller node is also running cinder and exports a 500GB SSD that way. The main bottleneck is the 100mbit ethernet usb dongle on the network node. All network traffic goes through this machine.
Everything connects to a 1820-24T network switch. I split my network into different VLANs to better segregate traffic. This includes backend management, the api network, and multiple networks to segregate projects within Openstack. Each compute node uses multiple VLANs off a single NIC. The network node is using a usb nic to route each of the VLANs associated with a project.
The cluster is also segmented from my home network through the use of VLANs and VPN endpoints. I have to log into a vpn to perform any maintenance to the system. I also have to log into a separate vpn server to access most of the lab systems I have running.
Initially I worked on getting this setup deployed via puppet. However, I ran into bugs with the official puppet openstack scripts. I also had a hard time figuring out how to get puppet to configure neutron for VLANs. Instead, I ended up writing my own bash scripts to install the cluster. You can download them from my openstack github repo.
Drawbacks to my setup
The main issues with my setup is the lack of shared storage. While I don’t have enough nodes to get true HA, it would be nice to be able to perform live migrations of instances. The 100mbit USB network adapter is also a pain to deal with if I have to upload data to an instance from my laptop.
While this setup works, it’s not ideal. I’m looking at rearchitecting my setup to use chef to deploy and manage Openstack. I’m also working on increasing my network speed and decreasing the reliance on the slow networking node by updating neutron to use Distributed Virtual Routing.