On Aug 22, 2008, at 11:06 AM, Herbert Graf wrote:

> In fact (and I might be completely wrong here), I'd personally find  
> that
> if a stack sent out an ARP request for a NON local address (an address
> that doesn't match the netmask) that would be a sign of a broken  
> stack.

It's somewhat debated just how much "hosts" are supposed to know  
about routing, and it's not THAT uncommon for a host to just ARP for  
everything.  A well-behaved router will answer appropriately (ie it  
lies!) if it has a route to the ARPed host...

Prior to well-working DHCP (Dynamic Host Configuration Protocol),  
this was one of the few ways to get a host working on a network  
without having to know lots of little details that end users weren't  
likely to know (What's my subnet mask?  What IS the default router?)   
A host can also guess wrong about which destinations are "local" due  
to thinks like variable length subnet masks, issues of network class,  
multiple subnets on one cable and so on.  For example, cisco terminal  
servers (when they're not also routers) have also been willing to  
utilize ARP for ALL destinations since forever, and cisco routers  
will happily answer such ARP requests.  It's also nice for minimal IP  
implementations on very small hosts.

ARP started out as a way to resolve a local IP address to a physical  
ethernet address, but it works quite well as a generalized "What is  
the physical destination for this high-level address" protocol.  I  
like it when a protocol shows that much flexibility.  It's a MUCH  
better idea than having hosts try to listen to routing protocol  
messages.  I have particularly fond memories of "ARP Storms", and the  
one well-meaning host/router that responded an arp for an IP  
broadcast address with the HW broadcast address...
(Host that thinks the IP broadcast address is 0.0.0.0 gets a packet  
(via broadcast) for 255.255.255.255.  "Hah!", it thinks.  "I'm a  
router and someone wants me to forward this packet, or surely I would  
not have received it.  I do not have this particular route, so I will  
ARP for it!" (Note that ALL similarly configured (early desktop unix)  
workstations get the broadcast and do the same thing; that's a  
"broadcast storm.")  Now if some somewhat differently mis-configured  
says "idiot. 255.255.255.255 maps to FFFF:FFFF:FFFF", and all the  
first set happily forward their packet there... Instant meltdown!   
Whee!)

BillW


-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist