Networking Upgrades 2026

Background

For a number of years my partner and I were using 2.5gbs switches as well as an old consumer router from Netgear that capped at 1gbs...I believe that router was like $600. Eventually the router was marked as to be EOL at the end of the year which in turn kicked off a roughly 24 hour long period of fighting with the network configuration to maximize our uplink and address any issues from the upgrade.

How we got here

This should be obvious, but once you are getting into a $500-600 consumer you will likely have far better results with some small business gear whether it be off of ebay or new. The old router was purchased back in 2021 and I working a hybrid office job where I was experiencing issues with an SSH connection into a machine where some hardware I needed to test was located on. Ultimately replacing the entire thing due to severely degraded performance following me getting some virtual machines on my network. This was fine...only because I lived alone.

I would meet my partner in 2023 (through VR actually) and move in with her in 2024. Shortly after we moved in the network...worked...but definitely was overloaded. I looked at it and just chuckled, having moved cross country I felt investing in a new router was not a good decision and it was unlikely that we would hit the limits on a $600 router with 2 people.

When things started breaking

Eventually I would end up deploying a K8s server cluster and my partner would deploy a server cluster running running EXO (MLX so MacOS AI inferencing). As well as just rebuilt the homelab in general. Overall...a very nice bit of infrastructure...and very enjoyable to use\work with.

Unfortunately, the moment we added in network storage and migrated some things to K8s the network very quickly started to show signs of degredation with things like random increases to ping (ex 300ms on the local network), connections dropping, and lengthy load times to navigate the router web interface when we needed to set static IPs. Most of the time this wasn't visible to us, but on rare ocassions it did show in our usage of our systems.

Silver lining

The router we had would end up going EOL which would result in us really having no choice, but to replace the hardware we had. And for the price point we were looking at we ended up looking at small business routers. While I do know how to configure things like Cisco gear, I wanted something that could also be configured by people other than me (ex my partner). In our case this was the Ubiquiti Dream Machine.

Installation

The Ubiqiti dream machine is only 1gbs ethernet ports, but SFP+ ports at 10gbs. What this means is you can connect a switch to the LAN SFP+ port and hookup a WAN SFP+ port that you can break 1gbs on outbound. The problem here is you need to actually be able to negotiate a 10gbs link or have an SFP(+) adapater that can support 2.5 if you want to use it on a 2.5gbs internal network. So when we attempted to run SFP+ what happened is we ran it into a 2.5gbs. The simple solution to this is just...deploy a 10gbs switch with a SFP+ to ethernet from the LAN SFP+ port.

My partner was able to export the routing table from the old netgear and then went through and manually put in the individual IPs. And sure a few were missed and 1 or 2 were keyed in wrong, but at 27 static IPs it was inevitable that 1 or 2 would get keyed in incorrectly. Most of the time this was minor; for example the NFS share rejected connections because the IP address was incorrect or an SSH connection fails over an upgrade.

Improvements

While this would allow me to fully utilize our outbound connection in terms of bandwidth (SFP+ is wonderful) it had a very little impact on stability for most applications within our network. In the case of a moonlight application stream there was a small amount of stability improvement on this as even with all the changes we had made a disconnection wasn't unheard of and was something you might see once a week afterwards that mostly disappeared. For VR applications when my partner and I were both in some fairly heavy worlds (100 people present with a lot of custom user content) we did notice a pretty solid improvement in general load times as well as better stability to the remote servers in general.