[CentOS] Multicast versus broadcast network load (was:Re: How to make nodes in my local LAN see each other's names)

[Lamar Owen]

On Friday, December 02, 2011 12:40:32 PM Les Mikesell wrote:
> On Fri, Dec 2, 2011 at 11:16 AM, Lamar Owen <lowen at pari.edu> wrote:
> > But, lacking metrics, it's somewhat of a moot point.
 
> My point is that every device on your network has to process every
> broadcast packet.  Maybe you have CPU overkill on all your computers,
> but you might also have some dumb controllers too.  And they have to
> go out the wifi too.

Ethernet multicast frames, depending on switch implementation, may go to every device as well.  How the NIC responds to multicast ethernet frames is likely implementation-dependent, but, again, I don't have metrics on that, except for a failure mode on a couple of controller-type devices we experienced once. 

This failure appeared, on first blush, to be a broadcast storm, but ended up being a misconfigured IP webcam which was configured to send its mpeg4 stream via multicast as well; it loaded every port on every switch on one subnet's VLAN; our LAN is using a mix of cisco catalyst 6500, 5500, and 2900XL switches; some 3Com superstack II switches, and a handful of Extreme Networks Summit 1i switches; not low-end unmanaged stuff.  This stream was ~5Mb/s (it is a relatively high-resolution color IP camera).

Now, we have a number of SitePlayer Telnet devices (really nice and inexpensive ethernet-to-serial boxes that have great use in remote serial consoles).  These SitePlayer Telnet boxes have 10Base-T ports; the 5Mb/s mpeg4 stream overwhelmed them and dropped them off the LAN completely.  Several of them hard-crashed and went completely offline, requiring a power cycle to get back to operation.

This was not broadcast traffic; it was multicast traffic, but the switches flooded those frames to every port in that VLAN anyway, and a controller that was not a member of that multicast group got flooded.  Now, to be fair, the SitePlayer Telnet does Bonjour, and thus does respond to mDNS, but that is supposed to be on a different multicast group.  Whether that was a factor or not I don't know; but I do know that the Davis Instruments ethernet devices we have, and that don't do mDNS, also went offline during the multicast event.

So, no numerical metrics, but anecdotal evidence that multicast can be just as bad as broadcast to controllers with insufficient bandwidth or CPU power.

(And it pointed to the fact that those SitePlayer Telnet boxes really should have been on a different VLAN and thus in a different broadcast domain.......)