Roon RAAT uses TCP.
Small but consistent audible improvement from Netgear GS105 to Cisco WS-C2960C-8PC-L as the switches upstream from my Linn Klimax streamers. I was very skeptical, but I could not deny what I heard, and even more when my wife who did not know what we were trying out noticed the difference without prompting (she has a very good musical ear). Only hypothesis that makes any sense to me is that the cheap SMPS with the Netgear injects more noise in the wiring. Replacing Cat 6 between switch and Klimax with SFP optical (the Cisco switches and the streamers support that) makes almost no (if any) difference beyond replacing the switches.
Like all things, there are well implemented devices, and then, not so well implemented....... 10-15 years ago, I was a big fan of the dark blue metal case Netgear stuff, including Access points, and their managed switches. When I began managing the network for a 13 campus regional school district (Cisco), I questioned the cost of a Cisco 2600 48port POE switch, being over 5 times that of an equivalent Netgear model.....
I spent an afternoon with a Cisco engineer exploring the CLI interfaces of both Cisco and Netgear stuff. I was fairly adept with Netgear, not so much with Cisco at the time. I discovered some very interesting differences in the overall approach between the 2 products....
As I've said before, at OSI layer 2, an ethernet switch sits between the Physical Layer 1 (twisted pair Catx, or Fiber) and the Network Layer 3 (TCP/IP, UDP). Any modern switch maintains an ARP address table that maps the Layer 1 MAC address of each network interface on the network to a Layer 3 IP address. This allows a direct 1:1 communication between hosts on the network. A key difference between between "enterprise" switches (Cisco, HP Proliant, Juniper, Meraki.....) and consumer/prosumer grade is how this ARP address table is maintained.
What happens when a new host appears on the network? That host will send an ARP Broadcast (Adrress Resolution Protocol) on the network. Because the OSI model fully supports multiple connection methods, this broadcast sent to *all* hosts on the network. In the case of an enterprise grade switch, the ARP table in memory will simply receive an update after an IP address is either assigned via DHCP, or statically configured. This table exists persists in memory, until it is either cleared (Cisco: clear ip route, or clear ip route x.x.x.x), or the switch is rebooted. (oversimplifying things here, ignoring duplicate IP's, teamed Network adapter (multiple mac addy's), loopbacks, POE provisioning, port flapping, etc....) At the district, most switches have uptimes well over a year, with power outages being the most common cause...
Consumer switches do something *much* easier...... In response to *any* ARP Broadcast the *entire ARP table is discarded*! This results in every host having to re-negotiate with the switch, rebuilding the ARP table.... Depending on switch and number of ports this can be pretty disruptive. Network acting up??? Reboot your router (including it's cheap built in switch....)
I'm also a big fan of Ubiquiti, their gear, across the line, is very well made (although in the case of their managed stuff, it really helps to know what you are doing). Their Edgerouters are excellent, can be configured pretty much any way you want, and accept 24v POE. I personally have similar uptime results with them