h a l f b a k e r yThe mutter of invention.
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
Please log in.
Before you can vote, you need to register.
Please log in or create an account.
|
EDITED: I originally started with trying to map something
like Microsoft:Domain Controller to an IPv6 address, but
as [pmarks] pointed out, the network segment is used to
implement hierarchical routing. But there's still hope for
this dough....
The key problem I'm trying to solve is memorability
of
the IPv6 addresses. I believe any standard to make an arbitrary IPV6 address more memorable is helpful (as
long as it compliments the IPv6 standard).
Consider
the following random address:
862A:7373:3386:BF1F:8D77:D3D2:220F:D7E0 for
SomeDomain.com
DNS is down and you need to ping it to see if it's
up. In IT we need to memorize IPv4 addresses all
the time. As pointed out by [pmarks], we can't let
a word determine the address, but we do need to
be able to extract a more memorable
representation.
The best way I see this to happen is similar to "scheme1"
to have a one-to-many relationship between characters
and 4-bit values. With every base16 value having 3
possible characters an algorithm can find a more
memorable
representation. The algorithm which suggests a
memorable representation would use a dictionary
of English words to help.
The result may be (only
network segment shown - and indicative only):
FEET3CATFFDOOR90:.....
{Remembered as FEET 3 CAT FF DOOR 90}
Still some randomness in there, but heaps easier to
remember.
So hopefully, we're still baking - I'll try and update my
application to suggest memorable strings.
Demo Tool
https://sourceforge...projects/ipv6namer/
An open source tool which suggests words to use to aid in remembering IPv6 addresses. [toadth, Nov 27 2009, last modified Dec 17 2009]
[link]
|
| |
Interesting. (But I'm sticking with DNS) |
|
| |
problem is if you limit the addresses to letter binhex and then limit the letters to sequences that make sense then the number of available addresses falls very sharply, if you then allow that people will elect to set "wordy" things in there then suddenly we have a problem of duplication. |
|
| |
Comparison with DNS and IPv4:
* DNS: 1 record mapping IP to Name syncronised
across thousands of routers worldwide.
* IPv4: Not very feasible, can't get enough
characters, or have to use Scheme4, which has too
high of probability for collisions.
* IPv6: Numbers too hard to memorise with current
scheme. Plenty of space for worded scheme
without collisions. Is sort of a built in DNS system,
but it would be best to register your allocation
with an authority to avoid any disputes. |
|
| |
[WcW] I believe that with Scheme1 (my preferred
method), duplication is not a problem -
probabilistically. If you just use 26 letters and 10
digits. You have 36 characters binding to 16 values.
Ie. At most you have a digit and two letters bound
to the same value. 0,G and X all map to 0. F and Y
map to 16. |
|
| |
More research into English words - more
specifically likely business nouns, would be able to
determine the best mapping. |
|
| |
But even this shouldn't be a problem. |
|
| |
Consider, the words "Apple" and "Microsoft".
Unless {A and M}, {p and i}, {p and c}, {l and r}...
etc.. all mapped to the same values AND the
hashed padding equaled the same value, there
would be no collision. |
|
| |
It's possible that a word like "Potato" could have
the exact mapping to values as "Corned", but even
if that was to happen, the hashed padding of both
values is based of the ASCII values and therefore
they would both have different padding. |
|
| |
Being "wordy" or even short words have little if
any impact on the chance of collisions... |
|
| |
Shorter words have better chances of having the
same Base16 values, but much less chances of
having the same padding. |
|
| |
Longer words have much less chances of having
the same Base16 values, and better chances of
having the same padding. |
|
| |
So hopefully, you now better understand the
mechanics of the idea and the extremely rare
probabilities involved regarding collisions. |
|
| |
[UnaBubba] ICANN don't have to. Like I said you
might just use it for the host segment (of which
you control). And in an IPv6 network I'd imagine
you wouldn't have to type in the network segment
when pinging a LAN computer by IPv6 for instance. |
|
| |
Eg. Ping LOCAL:DomainController -t |
|
| |
The OS would resolve LOCAL to be your local
network segment, and resolve DomainController to
a host segment (faster than a network DNS query). |
|
| |
[bobofthefuture] I don't think so. A network
segment name wouldn't represent a "single" host
(like DNS does), it represents the hole network -
of which there can be 18 billion billion hosts. |
|
| |
The DNS query of microsoft.com may return an
IPv6 Readable Address of [Microsoft:Web].
Considering you don't need billion billion addresses
at home, your home IP may be in the form
{ISP:Account} [iiNet Australia:John Smith] |
|
| |
I would think that all network segments will be
dispensed by IANA. And the cost of buying a
network segment would have a cost - much higher
than buying a domain name I would think. |
|
| |
The issue here is, that even if my scheme is not
used for IPv6 wording, the fact is there are
enough bytes to make words and therefore there
can be added value. And as you pointed out there
may be auctions for words which describe
subjects and topics (with Nouns being reserved
for use by Companies owning the trademarks). |
|
| |
This idea shows a lack of understanding of what IPv6
addresses actually are. An address is not simply a random
string of bits; address blocks are delegated to ISPs and
their customers in a hierarchical manner, so that routes
can be aggregated to prevent the size of the global routing
table from exploding. For that reason, deriving the first
half of the address from words is not feasible. |
|
| |
Typically with IPv6, the last 64 bits of the address are free
to be assigned within a LAN segment, so word-derived
suffixes could be somewhat interesting. |
|
| |
However, this idea suggests a non-reversible mapping
which "probably won't collide". If you want a one-way
mapping with collision resistance, you should be using a
real hash function. |
|
| |
But, I think this idea would only remotely make sense if
the mapping were bidirectional, so you could look at an
IPv6 address and read the text from the suffix. You can
pack A-Z and some punctuation into 6 bits, so a
bidirectional mapping could be constructed that lets you
store 10 characters in a 64-bit suffix. |
|
| |
However, the whole idea is stupid because IPv6 addresses
are low-level constructs for communicating between
machines, not for looking pretty to humans. If you want
human-readability, we've had the DNS system in place for
25 years now. |
|
| |
[pmarks]
Yer you're right - i'm going to change the original
post. I did know there were provisions
for hierarchical routing, but didn't think this
meant
building the hierarchy into the addresses. See
rfc2374. The first 48 bits of the network segment
are for public routing, the last 16 bits are for
private routing leaving the last 64 bits for the host
interface. |
|
| |
Now that you mention it, implementing an 8 byte
hash would've been easier, however more
destructive and more non-reversible. |
|
| |
My scheme is still recoverable though. The
purpose
of the scheme wasn't for aesthetic human-
readability - it's all about memorability. |
|
| |
You can still use a user defined word for specifying
the host interfaces segment. But the network
segment side cannot be derived from a word. |
|
| |
Go the IPv6 namer link in SourceForge. I've
updated the code to work the other way - helping
you use english words to make IPv6 addresses
easier to remember. |
|
| |
Examples: A362D63D907327A2 =
A3MIDMUD9GNU27A2
5E8630C9659BC9B1 = LEOMUGS9MLPRCPR1 |
|
| |
I think it would be better to suggest words based
on English phonetics, looking for common sounds
{CH, ES, ED, ING, PH, BO, BI, BA, TT} to make up
words such as CHEDING, BATTING, PHED. Etc..
they are still memorable and increases the chances
of finding words. |
|
| |
Feel free to contribute. I'm sure there's an Linguist who can help. Or someone who knows of a place
where I can find a list of English word components
(phonetics). |
|
| |
Did some volume testing - here are some
statistics: Probability of Finding an N length word
[N,Chances out of 1mil.] = { [2, 999981], [3,973134],
[4,506495], [5,87047], [6,8804], [7,615], [8,47],
[9,4]} |
|
| |
Coverage of N characters with words of at least X
length [N, X, Chances out of 100k] = {[16,2,82],
[8,2,45249], [16,3,.5], [8,3,9579], [8,4,160], [8,5,5]} |
|
| |
In my testing I found the following HEX: B5ED 80BE
1EB6 C58E which produced a fully covered address,
using words no smaller than 3 characters - BLED OX
BE HERM SLOE, unfortuantely the chances of this
happening are about 1 in 200,000. |
|
| |
There's a 50/50 chance to have 8 characters
covered with words at least 2 characters long. This
needs to be much closer to 100%. |
|
| |
I think the best way to acheive this is to go for
the strategy of using, small phonetics. |
|
| |
Even better! Show the possabilities for each
character and let the User string together
something they think will be most suitable.. I'll do
this first. |
|
| |
Implemented 2 changes:
1. Redistributed letters in mapping to increase
prob. of gettings word matches. First I analysed
the probability of a letter being used based on
Dictionary.txt, then logically adjusted the
mapping. (eg. there are 6 values which could only
have one letter, so I assigned the most common
letters there, then where values had two letters, I
assigned the most common letters to pair with the
least common letters). Here is the updated stats:
a) Probability of Finding an N length word
[N,Chances out of 1mil.] = { [2, 1000000],
[3,982931], [4,527349], [5,89684], [6,8848], [7,704],
[8,40], [9,1]}
b) Coverage of N characters with words of at least
X length [N, X, Chances out of 100k] = {[16,2,102],
[8,2,51631], [16,3,1.5], [8,3,10502], [8,4,178],
[8,5,6]} |
|
| |
2. Implement "Pairs". The probability of having two
specific letters next to each other is quite rare.
But if we take the most common pairs and bind
them to values, we have better chances of making
words. I analysed the probablity of any two letters
appearing based on Dictionary.txt, then
distributed 22 of the most common across the
mapping. Starting at 0 (which is the least common
end - with single character mapping). Here is the
updated stats:
a) { [2, 1000000], [3,999973], [4,841067], [5,185467],
[6,15534], [7,847], [8,35], [9,1]}
b) {[16,2,3780], [8,2,96882], [16,3,40], [8,3,38070],
[8,4,790], [8,5,10]} |
|
| |
I had to make some additional changes to support
string to value mapping. When reading an IPv6
string, while iterating through each letter the
algorithm must: i] look for pair matches first
(looking ahead N characters). ii] if a string
matches, the follogin N characters in the
string/pair is ignored (ie. can't be used as another
pair) iii] if no string matches, then regular
character to value mapping is used. |
|
| |
Because of the pairs, the words can actually be
longer in characters than the original HEX string.
EG. SHORTEN in the old scheme would map to 7
HEX characters. In the new scheme EN is a
common pair, so the word SHORTEN would map to
6 HEX characters. I believe the loss in efficiency is
more than made up, by the improvement in
memorability. Remembering a series of single or
twin characters is a lot harder than remembering
words. The human brain doesn't just remember
the spelling of "SHORTEN", but the action of
SHORTENING. I encourage any academic institution
to perform further research, where human
subjects are given several tests made up of IPv6
Words to see which schemes are the best for
memorability. |
|
| |
Finally, I set out to acheive a much better chance
of having 8 spread with at least 2 character words.
With change 2, we went from 50/50 chances to
97/100 chances! The aim has been to make IPv6
addresses memorable and to acheieve this we
need to ensure we have high chances of providing
words. I've just got to make a few final changes to
the project before updating it on the website. I'll
let you know when it's up. |
|
| |
The source forge project is updated. You can now
demo using "pairs". While doing this and testing, I
found a little set-back. EG. RI is a pair, but it is
possible to select R (for it's value - 14) and I (11)
seperately. I included a check to notify the user
when such "Incident Pairs" occur. The user needs
to simply select other words or change a value
position to rectify such an error. |
|
| |
Due to this, I also need to update the program to
automatically avoid those problems. And also this
will reduce the statistics slightly. |
|
| |