Multicast is semi-broken on most networks, and devolves into pure broadcast on the current LAN or VLAN in most cases. This in turn causes chatter havoc on LANs with many devices, such as corporate WiFi. Oh, and multicast is totally insecure unless you layer your own authentication and encryption on top.
Multicast has been a dream since the 1990s but has been a nightmare in practice.
I am talking about multicast for having applications talk to each other in a datacenter, not to have that traffic exposed to office clients.
I understand most office networks are broken and if I send multicast I would just be causing untold mayhem.
I have already resigned myself to the fact that the only way to have reliable communication with a client is to use HTTP.
But in a DC where you control all your networking devices, configuration and people who maintain it, it should be possible to find a configuration that works reliably and allows devices to talk to each other without too much hassle.
> it should be possible to find a configuration that works reliably and allows devices to talk to each other without too much hassle.
The number of multicast bugs in Cisco/Juniper/whatever enterprise gear is astonishing. It’s basically a DoS waiting to happen. And if you have multiple network vendors in your shop as all real networks do… forget it.
As I said, my long experience is that multicast just doesn’t work reliably in practice except maybe for small layer-2-only networks. The same places where broadcast storms aren’t noticeable.
Unicast with source replication is fast, cheap, and reliable.
Multicast has been a dream since the 1990s but has been a nightmare in practice.