Agree or disagree?
Yes, it always depends on the type of workloads, size of the infrastructure, however, network switches are normally a 5+ years investment, so why limit yourself right off the bat?
Ports are backwards compatible with 10GB and 1GB (with the right adapters), price difference is not as noticeable as before (the gap in price has truly been reduced) and you are putting yourself in a much better situation to support whatever comes out next (who knows what the network requirements will be in 5-7 years). So tell me, why not just get 25G to begin with?
Wifi technology (5G, or Wifi 6) continues to improve, therefore all the endpoints require more and more bandwidth from the access points as time goes by. Storage bandwidth consumption continues to increase as well with the introduction of NVMe and Optane technologies. And what about modern GPUs? Artificial Intelligence? I would even say that 25G may not even be enough for those workloads, depending the size of it.
From my experience, I first noticed the speed difference between 10G and 25G on an HCI deployment. It was honestly night and day difference though. Evacuating hosts in the 25G cluster would take less than 1 minutes (live vMotioning 40-50 VMs in that amount of time) vs. 3-5 minutes in the 10G cluster. Storage vMotions would simply fly. Disk rebuilds I feel like were much faster too, even though I was never able to measure it.
Use case scenarios:
- Mid/bigger Edge/Robo location. Most likely still a bit overkill at the moment, however, with the right switches, your network will be automatically is released in the next few years.
- Datacenter. No brainer. May even have to increase the switch to 40G in some cases too.
- Small Edge/Robo. A bit overkill at the moment, however, maybe you do not have enough compute/storage workloads, but enough Wifi access points to at least consider 25GB uplinks at least.
- HCI cluster once it approaches 6 or more nodes. 25G all day.
