On Thu, 9 Sep 2010, in the Usenet newsgroup comp.os.linux.networking, in article
>Thanks! That works.
Quote:>> 188.8.131.52 through 184.108.40.206 were Class B's or /16s
>Which means even without specifying /16 arpwatch would know the right
>width for my 172.16.0.0 I guess?
Assume != know. Arpwatch will assume a /16, but whether it's right or
not is pure random chance. What you do on RFC1918 address ranges is
your own choice, but I don't know of many non-RFC1918 networks using
a mask that wide. While the original DIX Ethernet (normally called
10Base5 or thicknet) did physically allow 10,000 systems on a network
(100 hosts/segment, maximum of two repeaters between any two hosts),
practical matters (allowing all hosts a "chance" to send a packet)
and the Ethernet specification itself imposed much smaller limits
(spec limit was 1024 hosts) - see RFCs 0917, 0925, 0932, 0936, 0940
and 0950 for much discussion. Our original layout used a /22 mask
(255.255.252.0) allowing 1022 hosts per subnet. Having more than
about 300 caused substantial collisions and low through-put, and we
wound up installing the original Kalpana EtherSwitches in ~1990 to
break the subnets into sections with no more than 70 systems per wire.
Modern network design uses the concept of one system per switch port,
and the limitations become the aggregate bandwidth capabilities of the
switches (see the spec sheets for the switches for details).
Quote:>> What problem are you attempting to solve by having two networks on
>> the same wire? It's probably not doing so.
>Just trying to keep my data and management (IPMI) networks logically
>seperate. Bad idea?
Traffic is already separated by tuple (addresses/ports/sequence
numbers), so all you are doing is increasing the size of your ARP
caches and the size or complexity of your network stacks. I don't
see it buying anything.