We operate a network where we have 'tons' of VM servers that are physically hosted on many many blade chassis. The thought behind the original design was to simply put them in one very large broadcast domain so there are no restrictions on where you move blades or VMs to. Need more IPs? Just allocate a new block and keep adding blades.
As we own our own IP space this lead to allocating a
- /22
- /20
- /24
- /24
All on one L2 segment. I understand this is against 'conventional' IT wisdom but besides the potential for a loop is this really "that bad" on modern network equipment? These VM servers are essentially just plugged into one of two switches that are daisy chained together.
Most of the traffic is unicast and besides ARP there isn't a lot of broadcast traffic. All the servers are connected w\ 10G links so a few Mbps of ARP seems to be nothing. We push around 60 Gbps of BW.
With that being said my network equipment consists of:
- Juniper MX204 routers.
- Cisco N9K-C93180YC-EX switches
- Servers are standard HP Moonshots.
We tightly control the network so the potential for a misconfig taking down the entire stack is small. Do you think this could be causing performance issues or in the grand scheme of things is nothing?