And see Venti and Fossil for clever archiving solutions and distribution of the back end of a disk server system. These run on Plan 9, but can be used with v9fs from Linux.
"UNIX is obviously the mother of Linux in that sense, in that we stole all the basics. But a lot of ideas come from a small operating system called Plan 9 that was done by a lot of the same guys who did UNIX in the first place." -- Linus
"I don't follow other operating system development too much, except Plan 9 and Inferno.." -- Linus
Yeah, but it escapes me, why the people who design the kernels, drivers, desktops, "system applications" and all that other open source stuff do not need to "make a living". Why the game designers are the only exception?
It was really called "ingress filtering", not egress, see http://community.roxen.com/developers/idocs/rfc/rf c2827.html.
Ingress in IETF parlance means "coming in to the big Internet", and that's where spoofed source needs to be checked, by end-user routers and various levels of transit traffic operator routers connecting to ever bigger networks.
It seems though, that language is eroding, and the concepts of ingress and egress are now confusingly relative, to judge by the One Truth of Google.
I can't find a link to verify it anymore, but the young terrorist guerilla who is talked about here
was somewhere else said to have been playing in an Internet Cafe not just any video game but -- Counter-Strike.
Yes, it is interesting to note James Gosling's software history. I was a HUGE fan of NeWS at the time. (That has been one of the three elegant things in computers I ever loved passionately, others being Modula-2 and Plan 9/Inferno.
First he and others at Sun thought PostScript is an elegant interpretable general purpose language, why not send that to GUI clients to offload arbitrary GUI widgets -- or any code -- from servers? Just add input handling and host communication to PS. You also get better vector/pixel/font/colour model than the hackish X. Vector graphics NeWS windows and fonts could be arbitrarily scaled and still use all resolution you have, i.e. on printers.
They even built a nice event model as PS widget libraries, where stacked dictionaries were an elegant inheritance mechanism.
When NeWS failed as a distributed window system (I think because Adobe destroyed that clone of PS with anti-Sun politics and their own inferior and competing DisplayPS), Gosling became soft for complaints about non-conventionality of PS stack syntax. He threw away the PS syntax and ran away from the GUI wars dropping the graphics part, but kept the idea of distributing interpretable system modules as NeWS intermediate code. NeWS did PS -> pseudocode, now his new language, first called Oak and then Java, compiled to a similar pseudocode.
So sad. NeWS lost so much possibilities and elegance when it became Java, and gained nothing excpet a new language syntax (that I hate, like I hate C++).
Note that NeWS server application were language neutral so that no matter what language you wanted to use on the server, you just wrote your widgets as PS text (or precompiled pseudocode) to the clients at initialization and later on demand by the client when first needed.
Propbably BitTorrent conflicts with GetRight an other similar software, too. Those "multiple-TCP-streams to one HTTP/FTP-server, use mirrors too" programs work fine for me, and I get always very good browser download speeds in Scandinavia, mostly good in (50-150 kilobytes/second) from all over Europe and reasonable speeds from the USA. BitTorrent did not seem to do any better.
And I think building algorithmic ad hoc virtual tree and mesh topologies for distribution is an overkill. Simple strategy like Xolox and EDonkey use is good enough: hash the file blocks, index the file-block-hash-chains by the file names, keep searching for blocks by the hashes in the chain, then swarm download the blocks. This automatically swarms incomplete files too.
Hints for intelligents ways to implement hash based content block distribution can be found in the still unannounced "free-as-in-common-sense source" Plan 9 Venti system:
It is somewhat difficult to port existing Unix applications to Plan 9. There is a POSIX compliance APE environment, but its use id discouraged in the Plan 9 cimmunity. And the environment is full of diffrent "/services" that you should use instead of POSIX system calls to integrate well.
Inferno VM is currently heavily oriented to one programming language, Limbo. There are projects to run Java on the virtual machine, but they are not exactly production quality or marketable. And the philosophies again clash: you should use the existing "/service" components, not the extensive Java environment libraries. If you are a customer of Vita Nuova, you can get the C source to the Inferno environment, and program in C, too.
Lack of applications is obvious. There are development tools of course, and a rudimentary Web browser, but not much else.
I would like to highlight some of the good and bad points of Plan 9 and Inferno that were not mentioned in the interview.
The soul of these systems are the protocols 9P, (the new version will be renamed 2000P) and Styx, even more than the actual OS implementations. The protocol is a bit like raising the abstraction level from TCP "transport" layer to somewhere closer to the "session" layer, although the OSI terminology does not fit very well.
First important idea of the protocol is, that all functionality or "objects" is mounted remotedly and bound locally as directories, called "file systems" in Plan 9 parlance.
This means, that naming, user rights management, authentication, encryption and all that which f.ex. CORBA2 provides as complex badly interoperable abstract extensions are there with strict binary interoperability for all heterogenous environments. Of course multiplexing and streaming is there, because you have a set of bidirectional files or "named pipes", if you will.
Note that all this is independent of the programming language. There are C and Java libraries for accessing 9P or Styx objects.
An example: the access to TCP/IP functionality is a/net directory (http://www.vitanuova.com/inferno/man/3/ip.html). You open sockets by writing strings to control files. Sockets are created as dynamic subdirectories in the/net and controlled by writing to additional control files.
The second major point is process security. The file system name spaces are per process. If you only give a process the/net directory, then it absolutely can not access your disk or any other functionality. If you hide the/net/udp subdirectory, the process will not be able to use UDP, never.
The third point is related to second: inheritance or "stack directories" or "union directories". You can have a base file system like/net, then you can have any number or restrictions or augmentations to this in the form other file systems. You can just bind them as a stack, where the upper directories selectively hide or create new file names to the hierarchy.
You can give the stack to the name space of any process. Now some of the original names are visible and data to them goes transparently to the original implementation process. Some names are new, and data is routed to the modification implementation. Some of that may be redirected to the original names after checks or modifications.
And the iplementations can be mounted from anywhere on the network. You can have several machines running several OS' and programming languages with 9P/Styx, and they all are mounted, bound and stacked to one directory, say "/service", for your chosen client process, which does not see the configuration of the system.
For example low level "device" file systems, see
http://www.vitanuova.com/inferno/man/3/INDEX.htm l
and for higher level file example systems
http://www.vitanuova.com/inferno/man/4/INDEX.htm l
Note, that in my scheme above everybody should allow sub-domain delegated with others, so that if I run s.s.i. but no machine *.n.s.s.i., then anybody could register the n.s.s.i. sub-domain with me.
If there would be a conflict of two organisations wanting to register x.any. with the a.n.y. server, the owner of a.n.y. would decide. The would be so much freedom in DNS, that conflicts would not arise.
Why is it so difficult to de-centralize the top level domain administration? There must be some research proposals? I have one idea:
Let's set up 26 new TLDs:
a. b. c. . . . x. y. z.
After that, if I want to have a name www.anssi for my server, I will go and talk to the administrator of domain "i." And register there a domain www.a.n.s.s.i.
Depending on if the domains "s.i.", "s.s.i.", etc. already are registered, I need to go to talk to the highest order domain administration that is already registered.
After that we need to standardize a convention, that a resolver breaks down the last part after the last dot in a name into letters (www.anssi -> www.a.n.s.s.i), before it forwards the query to a DNS server.
Additionally the old TLDs will also be served in new "synonymous domains"
c.o.m. f.i. u.k. . . . etc.
Of course the old TLDs com., fi. and uk. still exist and work, if older resolvers query them.
Now happens something nice:
a) www.anssi works with the updated resolvers
b) *.com and *.fi work with all resolvers, both with the updated and old ones
c) manually written www.a.n.s.s.i. and www.yahoo.c.o.m. work with all resolvers
My ISP just called. A Hollywood detective agency had contacted them and informed, that a dynamic IP address once given to my cable modem had had Gnutella running at some point in time and it had shared some episodes of Futurama. And now they are after me.
Note that I live in Finland, so I guess somebody has decided to mount a large scale attack against global peer-to-peer piracy.
I see only two choices. Either we recognize, that copyright at the network age can no longer be enforced, and decide to live without it. The historic age has lasted thousands of years, yet we didn't need or have copyright laws until some 200 years ago.
Or we decide to enforce it. The simplest and most straightforward way to do this is to attack the actual violators: individual distributors. Everybody who published copyrighted material on the net could be sued, and compensations big enough to make this a business for lawyers could be ordered for the individual to pay. Many people will not like this, though.
Enforcing this globally is like controlling the drug business: it is hard, but the life for drug dealers can be made somewhat difficult. Besides: drugs bring money to criminals, and no money to the society, so financing the war against drugs is difficult. Most copyright violations don't bring money to the criminals, but enforcing the law brings profits for those who are willing to pay for tracking down the violators.
Technology can't solve the dilemma. Note that we can't make it illegal to only _sell_ pirated material, so that free distribution would be allowed. It won't solve anything, and it will open up all kinds of businessess which offer "free" files for download but charge for some kind of "not-related" service, like indexing, searching, disk space or bandwith.
And for the entertainment industry to set up its own network distribution system won't solve anything. It will not have a market, as long as individual violators are not punished hard enough. It will only feed material for illegal distribution. We know that copy protection cryptography is doomed to fail. Whatever people can see, they can transform to a not-copy-protected digital form. Cryptography can help though, if we decide to enforce copyright. In that case distribution of files encrypted with reasonable quality crypto may be allowed for free distribution through Napster, FT, Gnutella etc... After that only the _keys_ to open the encryption will be the copyrighted material, which you will have to obtain from legal, payable sources. But this only works if an army of detectives and lawyers will hunt down every individual who distributes copyrighted keyes illegally.
Yes. The phone is built running Inferno. This hack replaces that with Linux. I wonder is there is any rational reason to do so, though. Inferno is much better solution for an embedded OS. Notice how the nifty Inferno interface let's you easily replace the whole bootstrap code via the serial port.
In the beginning of my reply above, it should read:
Note, that if you have registered "a.b.c." with "b.c.",
and not like I wrote:
...with "a.b."
Re:A letter hierarchy based scheme for arbitrary T
on
Legitimacy Of ICANN?
·
· Score: 1
Note, that if you have registered "a.b.c." with "a.b.", and somebody is resolving a name "www.anssi.foobarabc", a query can be immediately satisfied without delegation when it arrives to the DNS server at "a.b.c.". If all subdomains below "a.b.c." referenced in "www.anssi.f.o.o.b.a.r.a.b.c." are local to "a.b.c.", then they can be specified in the one and only zone file for "a.b.c.".
Additionally, note that although there will now be much more domain components in the names than before, it will probably not add much to the number of delegated subqueries in global statistics. The more queries your name server answers, the bigger part of the single letter hierarchy it will learn to cache. The single letter domains should have a long TTL. When your name server has answered order of alphabet_size^2 queries, it has probably cached most of the second level domains (a.a., a.b., a.c.,...z.z.). When it has answered alphabet_size^3 queries, it has cached most of the third level domain DNS IP addresses.
To make sure that this learning process is not interrupted too often, it would probably make sense to keep a disk based cache of the letter based root delta domains. And have a guideline specifying a minimun TTL value that the single letter domains must advertise.
A letter hierarchy based scheme for arbitrary TLDs
on
Legitimacy Of ICANN?
·
· Score: 1
How about this, I just posted this to the IETF dnsops wg list:
Let's SET UP NEW TOP LEVEL DOMAINS FOR ALL SINGLE LETTERS a., b.,...z. So the top domains will be the classical ones + these new one letter TLDs.
Now I want to have a name "anssi.abc." for my machine. I rewrite it (first in my head) to be "anssi.a.b.c."
I go to the "c." TLD server administration. I ask them, if anyone has registered a "b.c." first level name with them. If such domain exists, I go to its administrator, and ask, if they have a domain "a.b.c." already. And so on, up to the letter, that is not yet served by anyone. If the domain "a.b.c." already does exist, I can't register a new one. First come first served.
In this case, let's say the domain "b.c." did exist, but the domain "a.b.c." did not. Now I register my "a.b.c." with the domain "b.c." administration. There should be a rule, that single letter domains allow registration for any unregistered single letter subdomains!
So now I can set up names like "anssi.a.b.c." in my own name server, serving the domain "a.b.c.", registered at "b.c.".
NOW I MODIFY MY RESOLVER and ask everybody to do the same, if they please! The new versions of resolvers are standardized to TRANSLATE ALL UNRECOGNISED many-letter TLDs into a sequence of one letter domains. So "anssi.abc." will be resolved as "anssi.a.b.c." domain.
Additionally, domains "c.o.m.", "o.r.g." etc. and corresponding country domains like "f.i." and "u.k." can be set up as peered, synonymous domains for "com.", "org.", "fi., "uk." and such. So that the new resolvers would not even have to know any list of special "recognizable" classic TLD domains.
Now anybody is free to set up whatever TLDs they want to!!!
Old resolvers will work with the old TLDs.
Old resolvers will kind of work with the new names, but only if you translate manually, using "anssi.a.b.c" instead of "anssi.abc".
New resolvers will work with all names, old and new, whether written as "anssi.a.b.c." or "anssi.abc."
Anssi.Porttikivi@teleware.fi
Data communications trainer/consultant
Teleware, Helsinki
Slashdot Mirror
And see Venti and Fossil for clever archiving solutions and distribution of the back end of a disk server system. These run on Plan 9, but can be used with v9fs from Linux.
"UNIX is obviously the mother of Linux in that sense, in that we stole all the basics. But a lot of ideas come from a small operating system called Plan 9 that was done by a lot of the same guys who did UNIX in the first place." -- Linus
"I don't follow other operating system development too much, except Plan 9 and Inferno.." -- Linus
There exists v9fs for Linux. See Google! Secure and high performace file sharing the Plan 9 way!
Yeah, but it escapes me, why the people who design the kernels, drivers, desktops, "system applications" and all that other open source stuff do not need to "make a living". Why the game designers are the only exception?
It was really called "ingress filtering", not egress, see http://community.roxen.com/developers/idocs/rfc/rf c2827.html.
Ingress in IETF parlance means "coming in to the big Internet", and that's where spoofed source needs to be checked, by end-user routers and various levels of transit traffic operator routers connecting to ever bigger networks.
It seems though, that language is eroding, and the concepts of ingress and egress are now confusingly relative, to judge by the One Truth of Google.
I can't find a link to verify it anymore, but the young terrorist guerilla who is talked about here was somewhere else said to have been playing in an Internet Cafe not just any video game but -- Counter-Strike.
Yes, it is interesting to note James Gosling's software history. I was a HUGE fan of NeWS at the time. (That has been one of the three elegant things in computers I ever loved passionately, others being Modula-2 and Plan 9 /Inferno.
First he and others at Sun thought PostScript is an elegant interpretable general purpose language, why not send that to GUI clients to offload arbitrary GUI widgets -- or any code -- from servers? Just add input handling and host communication to PS. You also get better vector/pixel/font/colour model than the hackish X. Vector graphics NeWS windows and fonts could be arbitrarily scaled and still use all resolution you have, i.e. on printers.
They even built a nice event model as PS widget libraries, where stacked dictionaries were an elegant inheritance mechanism.
When NeWS failed as a distributed window system (I think because Adobe destroyed that clone of PS with anti-Sun politics and their own inferior and competing DisplayPS), Gosling became soft for complaints about non-conventionality of PS stack syntax. He threw away the PS syntax and ran away from the GUI wars dropping the graphics part, but kept the idea of distributing interpretable system modules as NeWS intermediate code. NeWS did PS -> pseudocode, now his new language, first called Oak and then Java, compiled to a similar pseudocode.
So sad. NeWS lost so much possibilities and elegance when it became Java, and gained nothing excpet a new language syntax (that I hate, like I hate C++).
Note that NeWS server application were language neutral so that no matter what language you wanted to use on the server, you just wrote your widgets as PS text (or precompiled pseudocode) to the clients at initialization and later on demand by the client when first needed.
It had an extra blank,this works: http://cm.bell-labs.com/cm/cs/who/seanq/venti-fast .pdf
http://cm.bell-labs.com/cm/cs/who/seanq/venti-fast .pdf
Propbably BitTorrent conflicts with GetRight an other similar software, too. Those "multiple-TCP-streams to one HTTP/FTP-server, use mirrors too" programs work fine for me, and I get always very good browser download speeds in Scandinavia, mostly good in (50-150 kilobytes/second) from all over Europe and reasonable speeds from the USA. BitTorrent did not seem to do any better.
5 9- 1&oe=ISO-8859-1&q=venti&meta=group%3Dcomp.os.plan9
And I think building algorithmic ad hoc virtual tree and mesh topologies for distribution is an overkill. Simple strategy like Xolox and EDonkey use is good enough: hash the file blocks, index the file-block-hash-chains by the file names, keep searching for blocks by the hashes in the chain, then swarm download the blocks. This automatically swarms incomplete files too.
Hints for intelligents ways to implement hash based content block distribution can be found in the still unannounced "free-as-in-common-sense source" Plan 9 Venti system:
http://groups.google.com/groups?hl=fi&ie=ISO-88
Private Namespaces for Linux
1 2a .htm
http://www.ddj.com/documents/s=1782/ddj0112a/01
OK, and the bad news:
It is somewhat difficult to port existing Unix applications to Plan 9. There is a POSIX compliance APE environment, but its use id discouraged in the Plan 9 cimmunity. And the environment is full of diffrent "/services" that you should use instead of POSIX system calls to integrate well.
Inferno VM is currently heavily oriented to one programming language, Limbo. There are projects to run Java on the virtual machine, but they are not exactly production quality or marketable. And the philosophies again clash: you should use the existing "/service" components, not the extensive Java environment libraries. If you are a customer of Vita Nuova, you can get the C source to the Inferno environment, and program in C, too.
Lack of applications is obvious. There are development tools of course, and a rudimentary Web browser, but not much else.
I would like to highlight some of the good and bad points of Plan 9 and Inferno that were not mentioned in the interview.
/net directory (http://www.vitanuova.com/inferno/man/3/ip.html). You open sockets by writing strings to control files. Sockets are created as dynamic subdirectories in the /net and controlled by writing to additional control files.
/net directory, then it absolutely can not access your disk or any other functionality. If you hide the /net/udp subdirectory, the process will not be able to use UDP, never.
/net, then you can have any number or restrictions or augmentations to this in the form other file systems. You can just bind them as a stack, where the upper directories selectively hide or create new file names to the hierarchy.
m l
m l
The soul of these systems are the protocols 9P, (the new version will be renamed 2000P) and Styx, even more than the actual OS implementations. The protocol is a bit like raising the abstraction level from TCP "transport" layer to somewhere closer to the "session" layer, although the OSI terminology does not fit very well.
First important idea of the protocol is, that all functionality or "objects" is mounted remotedly and bound locally as directories, called "file systems" in Plan 9 parlance.
This means, that naming, user rights management, authentication, encryption and all that which f.ex. CORBA2 provides as complex badly interoperable abstract extensions are there with strict binary interoperability for all heterogenous environments. Of course multiplexing and streaming is there, because you have a set of bidirectional files or "named pipes", if you will.
Note that all this is independent of the programming language. There are C and Java libraries for accessing 9P or Styx objects.
An example: the access to TCP/IP functionality is a
The second major point is process security. The file system name spaces are per process. If you only give a process the
The third point is related to second: inheritance or "stack directories" or "union directories". You can have a base file system like
You can give the stack to the name space of any process. Now some of the original names are visible and data to them goes transparently to the original implementation process. Some names are new, and data is routed to the modification implementation. Some of that may be redirected to the original names after checks or modifications.
And the iplementations can be mounted from anywhere on the network. You can have several machines running several OS' and programming languages with 9P/Styx, and they all are mounted, bound and stacked to one directory, say "/service", for your chosen client process, which does not see the configuration of the system.
For example low level "device" file systems, see
http://www.vitanuova.com/inferno/man/3/INDEX.ht
and for higher level file example systems
http://www.vitanuova.com/inferno/man/4/INDEX.ht
or in Plan 9
http://plan9.bell-labs.com/sys/man/3/INDEX.html
http://plan9.bell-labs.com/sys/man/4/INDEX.html
Note, that in my scheme above everybody should allow sub-domain delegated with others, so that if I run s.s.i. but no machine *.n.s.s.i., then anybody could register the n.s.s.i. sub-domain with me.
If there would be a conflict of two organisations wanting to register x.any. with the a.n.y. server, the owner of a.n.y. would decide. The would be so much freedom in DNS, that conflicts would not arise.
Why is it so difficult to de-centralize the top level domain administration? There must be some research proposals? I have one idea:
Let's set up 26 new TLDs:
a.
b.
c.
.
.
.
x.
y.
z.
After that, if I want to have a name www.anssi for my server, I will go and talk to the administrator of domain "i." And register there a domain www.a.n.s.s.i.
Depending on if the domains "s.i.", "s.s.i.", etc. already are registered, I need to go to talk to the highest order domain administration that is already registered.
After that we need to standardize a convention, that a resolver breaks down the last part after the last dot in a name into letters (www.anssi -> www.a.n.s.s.i), before it forwards the query to a DNS server.
Additionally the old TLDs will also be served in new "synonymous domains"
c.o.m.
f.i.
u.k.
.
.
.
etc.
Of course the old TLDs com., fi. and uk. still exist and work, if older resolvers query them.
Now happens something nice:
a) www.anssi works with the updated resolvers
b) *.com and *.fi work with all resolvers, both with the updated and old ones
c) manually written www.a.n.s.s.i. and www.yahoo.c.o.m. work with all resolvers
Yours,
Anssi Porttikivi
app@iki.fi
My ISP just called. A Hollywood detective agency had contacted them and informed, that a dynamic IP address once given to my cable modem had had Gnutella running at some point in time and it had shared some episodes of Futurama. And now they are after me.
Note that I live in Finland, so I guess somebody has decided to mount a large scale attack against global peer-to-peer piracy.
I see only two choices. Either we recognize, that copyright at the network age can no longer be enforced, and decide to live without it. The historic age has lasted thousands of years, yet we didn't need or have copyright laws until some 200 years ago.
Or we decide to enforce it. The simplest and most straightforward way to do this is to attack the actual violators: individual distributors. Everybody who published copyrighted material on the net could be sued, and compensations big enough to make this a business for lawyers could be ordered for the individual to pay. Many people will not like this, though.
Enforcing this globally is like controlling the drug business: it is hard, but the life for drug dealers can be made somewhat difficult. Besides: drugs bring money to criminals, and no money to the society, so financing the war against drugs is difficult. Most copyright violations don't bring money to the criminals, but enforcing the law brings profits for those who are willing to pay for tracking down the violators.
Technology can't solve the dilemma. Note that we can't make it illegal to only _sell_ pirated material, so that free distribution would be allowed. It won't solve anything, and it will open up all kinds of businessess which offer "free" files for download but charge for some kind of "not-related" service, like indexing, searching, disk space or bandwith.
And for the entertainment industry to set up its own network distribution system won't solve anything. It will not have a market, as long as individual violators are not punished hard enough. It will only feed material for illegal distribution. We know that copy protection cryptography is doomed to fail. Whatever people can see, they can transform to a not-copy-protected digital form. Cryptography can help though, if we decide to enforce copyright. In that case distribution of files encrypted with reasonable quality crypto may be allowed for free distribution through Napster, FT, Gnutella etc... After that only the _keys_ to open the encryption will be the copyrighted material, which you will have to obtain from legal, payable sources. But this only works if an army of detectives and lawyers will hunt down every individual who distributes copyrighted keyes illegally.
Pirated Farscape episodes are easily available on Gnutelle with a good client like Xolox.
Try www.xolox.nl. Arguably the hottest Gnutella client, due to its elegant ease of fetching files with multiple simultaneous segmented downloads.
A little buggy yet, though. And closed source, Microsoft only.
Yes. The phone is built running Inferno. This hack replaces that with Linux. I wonder is there is any rational reason to do so, though. Inferno is much better solution for an embedded OS. Notice how the nifty Inferno interface let's you easily replace the whole bootstrap code via the serial port.
your sorry bastard discussing his stupid ideas with himself
In the beginning of my reply above, it should read:
and not like I wrote:
Note, that if you have registered "a.b.c." with "a.b.", and somebody is resolving a name "www.anssi.foobarabc", a query can be immediately satisfied without delegation when it arrives to the DNS server at "a.b.c.". If all subdomains below "a.b.c." referenced in "www.anssi.f.o.o.b.a.r.a.b.c." are local to "a.b.c.", then they can be specified in the one and only zone file for "a.b.c.".
Additionally, note that although there will now be much more domain components in the names than before, it will probably not add much to the number of delegated subqueries in global statistics. The more queries your name server answers, the bigger part of the single letter hierarchy it will learn to cache. The single letter domains should have a long TTL. When your name server has answered order of alphabet_size^2 queries, it has probably cached most of the second level domains (a.a., a.b., a.c.,...z.z.). When it has answered alphabet_size^3 queries, it has cached most of the third level domain DNS IP addresses.
To make sure that this learning process is not interrupted too often, it would probably make sense to keep a disk based cache of the letter based root delta domains. And have a guideline specifying a minimun TTL value that the single letter domains must advertise.
How about this, I just posted this to the IETF dnsops wg list:
Let's SET UP NEW TOP LEVEL DOMAINS FOR ALL SINGLE LETTERS a., b.,...z. So the top domains will be the classical ones + these new one letter TLDs.
Now I want to have a name "anssi.abc." for my machine. I rewrite it (first in my head) to be "anssi.a.b.c."
I go to the "c." TLD server administration. I ask them, if anyone has registered a "b.c." first level name with them. If such domain exists, I go to its administrator, and ask, if they have a domain "a.b.c." already. And so on, up to the letter, that is not yet served by anyone. If the domain "a.b.c." already does exist, I can't register a new one. First come first served.
In this case, let's say the domain "b.c." did exist, but the domain "a.b.c." did not. Now I register my "a.b.c." with the domain "b.c." administration. There should be a rule, that single letter domains allow registration for any unregistered single letter subdomains!
So now I can set up names like "anssi.a.b.c." in my own name server, serving the domain "a.b.c.", registered at "b.c.".
NOW I MODIFY MY RESOLVER and ask everybody to do the same, if they please! The new versions of resolvers are standardized to TRANSLATE ALL UNRECOGNISED many-letter TLDs into a sequence of one letter domains. So "anssi.abc." will be resolved as "anssi.a.b.c." domain.
Additionally, domains "c.o.m.", "o.r.g." etc. and corresponding country domains like "f.i." and "u.k." can be set up as peered, synonymous domains for "com.", "org.", "fi., "uk." and such. So that the new resolvers would not even have to know any list of special "recognizable" classic TLD domains.
Anssi.Porttikivi@teleware.fi
Data communications trainer/consultant
Teleware, Helsinki
Sorry, these links should work better:
Vita Nuova
Plan 9