Without trying to invite debate on properly sizing broadcast domains (different topic with many variables to consider) [...]
Too late. 🙂
I have heard of this before, regarding properly sizing broadcast domains, but for the life of Brian, I have not been able to find any sources citing this.
I've heard that creating vlans (broadcast domains) too big will result in traffic drowning in broadcasts due to things as ARP and such. But nothing citing when domains are too big. Whether stated by law or rule of thumb.
Dave, suppose you could bring any insight into this? Or anyone else, for that matter?
I'm with kYutobi on this one; provided you're not going to readily run out of IP addresses, then use /24 subnets; they can never be considered too big, as broadcast domains and just because they're really easy to work with, with the third octet letting you see, very simply, whether two addresses are really in the same subnet.
These days you'd probably get away with /23 or even /22, but as soon as you started getting some kind of intermittent performance type issues, there'd be that nagging doubt as to whether broadcast/L2 multicast levels might be coming into play, which take time to allay.
One exception to this, in a big network, would be if you ever need 'transit VLANs' or similar, in which case I'd go for /29. /30 is always tempting, in those circumstances, but if you ever wanted to insert a device between - say for management, monitoring or traffic manipulation purposes - you'd be stuffed, but /29 is still pretty address-efficient.
Recently, I had an installation at a Hotel which required a VLAN specifically for Television.
They wanted an IP net, where 3rd octet referred to floor, and 4th octet referred to room number. That ended up being a /21 network. Albeit, with only about 150 hosts.
Yes - I've come across similar, with store numbers, in retail chains. I'm never a fan of doing your IP addressing based upon anything like that. IP addressing wasn't designed for it and you often end up 'standardising' on something that then ends up with anything > 254 being non-standard. A standard that works only some of the time is worse than knowing you have no standard - and using an IPAM tool - IMHO.