Big Picture (was: Class 'E' - Extending the IP address)
Pierre (P.)Fortin <FORTINP@bnr.ca> Mon, 06 July 1992 15:04 UTC
Received: from NRI.NRI.Reston.Va.US by IETF.NRI.Reston.VA.US id aa08141; 6 Jul 92 11:04 EDT
Received: from munnari.OZ.AU by NRI.Reston.VA.US id aa05492; 6 Jul 92 11:05 EDT
Received: by munnari.oz.au (5.83--+1.3.1+0.50) id AA14678; Tue, 7 Jul 1992 00:09:44 +1000 (from owner-Big-Internet)
Received: from gateway.bnr.ca by munnari.oz.au with SMTP (5.83--+1.3.1+0.50) id AA14674; Tue, 7 Jul 1992 00:09:35 +1000 (from FORTINP@BNR.CA)
Received: from BNRECADA.BNR.CA by gateway.bnr.ca (5.61/1.34) id AA05643; Mon, 6 Jul 92 10:07:52 -0400
Message-Id: <9207061407.AA05643@gateway.bnr.ca>
Received: from BNRECADA.BNR.CA by BNRECADA.BNR.CA (IBM VM SMTP R1.2.1) with BSMTP id 4915; Mon, 06 Jul 92 10:06:02 EDT
Date: Mon, 06 Jul 1992 10:08:00 -0400
To: Big-Internet@munnari.oz.au
MMDF-Warning: Parse error in original version of preceding line at NRI.Reston.VA.US
From: Pierre Fortin <FORTINP@bnr.ca>
Subject: Big Picture (was: Class 'E' - Extending the IP address)
Sender: Pierre Fortin <FORTINP@bnr.ca>
Let me see if I've got this straight... [In all the following figures, ABCDE = assigned and abcde = unassigned network addresses.] (Assuming various proposals were adopted;) The Class structure is taking us recursively though a n/2 process of assigning addresses resulting in the following picture (from RFC1166): 0 1 2 3 01234567890123456789012345678901 000x 0aAAaAAAAA$AAAAAAaAAAAAAaAAAAAAA [$ = 10.0.0.0] 001x aaaAAaaaaAAAAaA@aaaaaaaaaaaaaaaa [@ = my Class A (47)] 010x aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa 011x aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaL [L = loopback] (mainly 127.0.0.1) 100x BBBBBBBBBBB<BBBBBBBBBBBBBB>bbbbb [<...> gleaned from RFC1335] 101x bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb 110x Cccccccccccccccc 1101 ################ C# 1110 Dddddddddddddddd 11110 xxxxxxxx length = x[64|92|128|160|...] 11111 eeeeeeee Simplified further: AAAaaaaaBBbbc#de address space allocated ??? =- . address space used It's bad enough the bulk of the 2**32 addresses have been wasted; now I'm hearing (er... reading) proposals to "punch a hole in the bottom of the barrel"... C'mon!!! Let's get real! :^\ Frankly, I feel very much like an address "hog" when I look at this picture. Especially when I consider that I have ~30,000 hosts spread over ~600 subnets for an average of 50 per... (less than 0.2% used!!!) If the Internet has ~900,000 connected hosts, that's only 0.02% used. WHOA! I'm already making 10x better use of my Class A address space than the rest of the Internet... that makes me feel [a little] better. Seriously, while some of my previous postings were obviously premature, I *still* have a problem buying into the premise that we are running out of address space. The real problem, as I see it, is that the address space has been **squandered**... In my case, I followed RFC950 to subnet network 47 at 12 & 12 bits. Given what we [at BNR/NT] now know about squeezing blood out of a stone, it *is* possible to have 600-800+ high-performance/high-traffic devices on a single subnet while still operating relatively safely in terms of broad-/multi-cast issues. [Done here using Multi-Port Bridges (MPB) with FDDI backbones. I have not yet made use of un-numbered serial links which would reduce our subnet "waste" by ~150. So, assuming: a. <1022 devices per subnet b. <510 subnets corporately we COULD have gotten along with 2**19 addresses (524,288) for a 5.7% address utilization. Then again, with aggregation (through variable- length subnet masks and RFC1219), we could have done just fine with 2**18 (9 & 9 split) addresses (262,144) for 11.4% utilization and room to grow; but this is HINDSIGHT... Looking at the picture differently: actual Class A: 16,777,216 or 0.3906% or 1/256 of the Internet classless 2**19: 524,288 or 0.0122% or 1/8192 classless 2**18: 262,144 or 0.0061% or 1/16384 Differently again (using 2**15 as "in-use" addresses on my network 47): actual Class A: 99.8% WASTED!! classless 2**19: 93.8% WASTED!! classless 2**18: 87.5% WASTED. Remember from above that my address utilization is 10x BETTER than the current Internet... and I _still_ feel TERRIBLE about this situation. With the current state of "secure gateway" technology, there is NO WAY that my Class A network will be visible to the Internet; and this will likely be so beyond any actual deployment of "solutions" being discussed here. So, feel free to consider network 47 "free space" for the purposes of researching new solutions. I wonder how many other Class A networks are similarly hidden? Any Class A address owners care to give their views? Especially those who are not visible to the Internet... With my [hidden] Class A network, regardless of how many gateways I have to the real Internet, I feel is is *MY* responsibility to get traffic from *anywhere* on my network to an Internet gateway for external access. As a result, I find the use of several existing Class C addresses MORE than adequate for Internet access. Of course, this is possible through our requirement that internal users logon to a secure gateway to then have access [via a single Class C address] to the Internet. In other words: other than horsepower limits (CPU and links), our users appear to the Internet as one of two hosts at a couple of gateway locations (FOUR addresses on TWO Class C nets to the Internet). Each user only really adds to the "port" numbers used. Now I really feel like I'm wasting Internet address space... Heck, with this scheme, I would think that a *portion* of my Class C gateway networks would be sufficient. They are subnetted 4 and 4 with a couple of gateway hosts each... I suppose that having a more transparent gateway would be desired by my users; but having an internal-Class-A to external-Class-C-plus-port translation gateway _might_ be satisfactory (gotta *think* about this one some more). Random thoughts: --------------- Looking at the 2**32 address space re-divided into two 2**31 spaces and overlapped; like this: 0 1 2 3 01234567890123456789012345678901 x00x <--------------A&B-------------> [assuming light Class A use] x01x <--------------A&b-------------> x10x <-----a&C------><-----a&C#-----> x11x <-----a&d------><-----a&e------> I might find it [a little] easier to accept all "the sky is falling" arguments. If I next [possibly naively] believe that all the visible Class A networks *could* individually fit into Class B spaces using *existing* subnetting schemes, the ENTIRE (less 0 & 127) existing Class A space would be free... at a cost of very few Class B addresses. Hmmm... I wonder how everyone would feel about 2**31 (2,147,483,648) addresses within which we could investigate new allocations, addresses, IDs and routing mechanisms? ...without otherwise destroying the current [bad as it is perceived] infrastructure. Just one bit to check: 0 = research traffic 1 = "normal" user traffic Let your mind w[a/o]nder... :^) -EndOfRandomThoughts- Just trying to help. Really... Cheers, Pierre PS: Regardless of what the outcome is within these discussions, there are still the issues of WHAT we are telling new users to do via existing RFCs. It's one thing to have a solution, but quite another to make sure it is not similarly scr*wed by current RFCs/thinking...
- Big Picture (was: Class 'E' - Extending the IP ad… Pierre Fortin (P.)