Fwd: Re: Reed's views, was [e2e] Cannara's views

John Day day at std.com
Mon Apr 16 19:00:16 PDT 2001

Sorry, did a "Reply" instead of "Reply All"

>Date: Mon, 16 Apr 2001 20:57:23 -0400
>To: jtk at aharp.is-net.depaul.edu
>From: John Day <day at std.com>
>Subject: Re: Reed's views, was [e2e] Cannara's views
>>John Day wrote:
>>  > In fact, even Ethernet reflects the state of our understanding at the
>>  > time.  Then we didn't understand that addresses only had to have the
>>  > scope of the layer in which they were used.  I really doubt that
>>  > there will ever be an Ethernet segment or even a bridged ethernet
>>  > with 2**48 devices or anything remotely close to it.  16 bits would
>>  > have been more than enough.  What is interesting is that IEEE 802
>>  > still hasn't figured it out.  Look at firewire addresses.
>>I was still learning how to dress myself while all this stuff was going
>>on, but reading into what Seifert in his "The Switch Book" says, the
>>thought about the 48-bit address space was done more for administrative
>>purposes than for the purpose of actually needing that many addressable
>>stations on a single data link.  An excerpt:
>>"... the Ethernet designers consciously took a different approach to
>>Data Link layer addressing.  Rather than trying to save transmission
>>overhead by conserving bits, we instead opted to create a huge address
>>space capable of providing a globally-unique Data-Link address to every
>>device for all time.  The Ethernet address space was designed to allow a
>>unique address to be permanently assigned to every device that would
>>ever attach to a LAN."
>Which in fact supports my point. All of that reflects our 
>understanding at the time, that Ethernet addresses would have to be 
>assigned and be globally unique, just as IMP addresses were and 
>later IP addresses.  It was only later that we realized that data 
>link (and Ethernet addresses) only have to be unambiguous within the 
>scope of the layer they are in. Link layers do not normally have the 
>same scope as Network layers except in very simple networks.  We 
>didn't really understand that when Ethernet was being created. 
>Consequently, not only did they not 48 bits, they didn't need to be 
>centrally administered either.  In fact, they only needed to be 
>administered and assigned by the owner of that particular Ethernet 
>segment.  To some extent, the success of Ethernet has made it very 
>difficult to explain to people what addresses are all about since 
>really Ethernet "addresses" aren't addresses at all, but serial 
>numbers of NICs applied at the factory.
>Was the address of your house assigned to it at the factory?  Okay, 
>so your house wasn't built in a factory, suppose they did it for 
>mobile homes.  Pretty cumbersome, eh?
>The nature of an Ethernet segment is such that a flat address space, 
>as opposed to one that is topologically dependent (the property that 
>makes a name an address)  for a LOCAL area network doesn't really 
>cause problems.  The only problem is that there are too many people 
>who think because it works for Ethernets it should work for 
>And by the way, there was concern that there would be enough.  The 
>argument for 48 bits at the time, as I remember it, was that if each 
>device weighed 1kg, then 2**48kg was more than the mass of the 
>universe.  We all got a little chuckle out of it as I remember.
>Take care,

More information about the end2end-interest mailing list