Posts in this series
I am using a cheap-in-the-extreme 10/100 DD-WRT wireless router for ingress/egress that I found at the Thrift store for less than $3. I converted its 12V power supply to 5V using a USB step-up transformer shown here:
I want my network design to reflect a traditional Spine/Leaf architecture seen in most data centers, with the egress router attaching to the spine, the object storage approximating a leaf attachment (non-blocking architecture) by connecting to the spine, but notionally representing devices connected to any number of leaf switches. My data center control plane management systems need to reside inside the core somewhere but be logically separated form the leaf nodes since they directly manage the physical hardware and share duties with the egress router. To accomplish this I would need at least 2 more switches. I found these little Netgear 5-port units that should be perfect for the task. They are managed switches and hopefully they support VLAN assignments. I will update once they come in, but usually a web managed switch has this functionality, and my DD-WRT router should support it already.
Obviously I will have to strip the cases off of everything for aesthetics and void the warranties.
Since I already converted the router power to USB, these are also 12V, I have plenty of spare amperage on my 5V rail for additional devices, and still have several step-up transformers lying around from the last purchase, it makes sense to power these via USB transformer as well. As long as these guys do not pull more than 2A of current each (I think they were rated at 1A apiece) it makes sense to convert these to USB as well to keep all the network gear the same.
I will be adding one of these to the NUC rack, and building out a separate “network rack” with one of these (spine) and the router in it.
My 3 Raspberry Pi 3B Swift nodes will plug directly into this spine switch in the network rack, as will my router leaving 1 port free for my uplink. Meanwhile my TOR leaf switch will connect the Neutron network node and Openstack controller (NUC w/wifi adapter and SSD drive,) the 3 hypervisors, and an trunk uplink to the spine switch. The PXE/MaaS server will connect directly to the router since nothing it does is very throughput intensive, and it is likely that I will be managing it via the wireless connection on the router anyway. This works out well because PXE requests will need to be forwarded to the MaaS controller from the router on the base VLAN anyway, and DHCP assignments will need to come from here for that subnet.