If the announcement is based on actual technical calculations (as opposed to getting-a-government-grant calculations or news-reporter-misinterpreted calculations), they're expecting the system will produce power for $21/watt (including research & development costs) - so $21K/kwh. At $0.10/kwh, that means they'd need to run for 210k hours to break even, which is unlikely; even at $0.30/kWh that'd be 70K hours, or a bit under 9 years. On the other hand, if that's $16B for R&D and $5B for production of a reproducible solution (unlikely?) that's closer to viable.
I think the big question is the power transmission part; solar panel technology is improving, and presumably part of the research is how to develop cheaper launch capabilities.
I don't have mod points today, but I did follow the poster's link. I think the proper moderation would be -1 Crackpot as opposed to -1 Troll, but as the poster says, "Soon, though, all that will change!"
Sure, spam is 97% of email, but email's a small fraction of the bits on the Internet, most of which is the web. It doesn't consume anywhere near the resources of Youtube. If you're in the mail-handling business, it's one of your largest problems, along with storage, reliability, etc. and burns most of your internet bits, but if you're in the general ISP bits, the spam's still not much of your bandwidth compared to the regular web traffic.
What spam really consumes is the attention span of its recipients, and therefore the resources and attention of mail handling providers who want to keep their users from getting spammed. But it's the attention that costs the money.
Practically all the illegal drugs used to be legal; the exceptions are new drugs developed by the pharmaceutical industry that would need FDA approval before selling them, and new drugs that fail the "similar to anything already illegal" test that the US Feds put out a few years ago to deal with designer drugs (which weren't illegal when developed, and the bureaucrats got tired of having to make new rules or get new laws passed to ban them.) Other than that, alcohol's the only major exception, since it used to be legal, got banned in the US, and then got re-legalized. But nobody's been willing to mess with the big killer drug, tobacco, or my favorite dangerous addictive drug, caffeine.
By the way, does anybody know if the LSD limit that the story reported is correct? It's way too low - 0.015mg is 15 micrograms; a typical street dose in Timothy Leary days was 250mcg and today is more like 100mcg, and the threshold for having any effects at all is around 25mcg for most people. If they were actually talking 15 milligrams (0.015 grams), that'd be 150 doses, so a small-time dealer, or if they meant 0.150mg, that'd be a medium-sized single dose, which is somewhat believable for a policing limit.
They won't be growing cocaine in their back yards - we don't really have the climate for it. But cannabis and opium poppies can grow well just about anywhere; the only reason people grow them in remote mountains is to avoid getting caught.
Drug addicts don't steal because they're drug users, they steal because they can't get enough money to feed their habits. You don't see that happening with cheap drugs - tobacco addicts don't go out robbing people to get more cigarettes, and drunks don't go commit burglary so they can afford a $10 liter bottle of cheap gin, even though tobacco's more addictive than heroin and booze is more destructive.
Heroin and meth addicts steal because of the artificially high costs of black-market drugs, but the drugs themselves don't cost much to produce if you don't have to avoid the law. Medical opiates are cheap - a $5 over-the-counter bottle of codeine would be enough to keep Rush Limbaugh happy for the day, except that it's mixed with acetaminophen (and in Canada, caffeine) to keep you from overdosing; it'd be just as cheap to make without the additives. And making meth in a pharmaceutical factory doesn't cost significantly more than making Sudafed.
All this crime and violence is because the government has convinced us there's some difference between a junkie and a drunk. What a waste.
There are a few big chunks that are still being hoarded, but not much; universities have mostly returned their early allocations, and some of them now belong to big ISPs which have legitimate use for them. (For instance, Bell Labs used to have a Class A to support their Cray in Murray Hill, but it's now used for AT&T's ISP customers.)
There's a lot more space out there for small-medium businesses that have Class C/24s that could probably get away with/28s or/29s now, but many of them need to have publicly-routable address blocks since they're connected to multiple ISPs, so they can't use anything smaller. (In the IPv6 world, they could use a/48 just fine.)
But no, it wouldn't save us much time - Asia's using multiple/8s per year for growth.
Yeah, I dealt with that market back in the 1980s, as well as the POSIX efforts, and worked with people who'd dealt with the US Auto Industry's attempts at standardizing on OSI as well. Five years ago, that was pretty much what government support for IPv6 looked like also, but it's different now, because the world really will run out of IPv4 addresses by about 2010-2012, so they really do need to do more than just hide behind NAT. And over the last few years their efforts to get testbeds run and actual operational experience have meant that ISPs and developers have some experience dealing with IPv6 now, and are starting to get ready to have to convert.
I went through this in the 1980s with the government trying to push their users to OSI and POSIX - every RFP had a check box saying whether you supported it, and sometimes you had to write reams of explanation about how you were going to do it, but for the most part the end users got waivers and ran MS-DOS on their underpowered desktops and deployed TCP/IP in any networks that weren't running SNA.
But it's different this time, really, trust us! Ok, 5 years ago it wasn't, but by now IPv4 space is close enough to running out, even for agencies that can live behind firewalls like the military does, that they're starting to get serious about using this stuff, and by requiring it they're forcing development of the tools it takes to use IPv6 in the real world and not just on paper. And they helped get Microsoft to build IPv6 support in Windows to sell to them as well as to the real market.
Don't know what town you live in, but any ISP that doesn't have a really solid IPv6 plan by now is going to be toast by 2012 if not before. Either they won't be able to get enough IP addresses, or they're going to have to set up big tunnel servers and learn how to use them in a hurry, or start carrier-NATing their customers which starts to break applications. If IPv6 were something you could simply enable overnight it'd be different, but it's not - you're going to have a significant learning curve as you retrain all your people and rebuild all your management tools and replace any equipment that's not fast enough to run IPv6.
I'd expect that it would get even more exciting for people in the hosting business - not only do you have to keep getting more IP addresses as people virtualize their hosts, but as there start to be IPv6-only end users who want to reach websites, your hosting customers are going to want to have IPv6 support, and if you can't at least give them dual-stack, they're going to become somebody else's web hosting customer instead. You don't happen to have any Layer 3 switching gear in your hosting centers or manage load-balancers for customers do you? Good luck!
Disclaimer: This is my own personal commentary, not the opinion of my employer, other companies they own, or my other corporate overlords. But it's still correct.
All major ISPs are thinking about IPv6, most minor ISPs are thinking about it, and any that aren't thinking about it are going to be in serious trouble by 2012 or before. A few years ago, what they were thinking was mostly [Expletive deleted!], but most of them are a bit more focused by now and the rest won't outlive the Mayan calendar rollover. Sure, some of us own large netblocks that will it may be possible to monetize, but just about everybody who does also owns large Cisco routers that will need to be replaced, which costs a lot more money, and in many cases it's the whole collection of operation support software that needs to be updated, which is even harder than just upgrading all your capital equipment. This is basically about as big a problem for a typical ISP as Y2K was - there are 32-bit address fields embedded in all sorts of things, so it's not as simple as just putting new values into existing tables.
Sometimes you can get around it by shoving your dynamically-addressed users behind another layer of NAT, and adding some IPv6 tunnel devices, but if your DSL or cable modem users require a hardware upgrade to support IPv6, it's potentially going to cost more in support to handle the transition than you'd gain by charging them extra to remain on IPv4, even if you make them buy the box. And if you're a business ISP, and your customer with a T1 line has been running fine on a Cisco 2500-series router for the last decade, well, time's up on that hardware. Most newer CPE routers can handle IPv6 fine, though if you've got a user with a T3 or E3 line (45 or 34 Mbps), you may find that it can't go full wire speed with IPv6.
There are a very few universities that still have big netblocks they haven't given back, but there's been enough market for them for the last decade that anybody who didn't need theirs has had plenty of incentive to sell them already. A much larger impact is the number of businesses who have a/24 address block when NAT and firewalls mean that they could get by just fine with a/28 or/29 to handle an inbound web server and VPN or two. In many cases, they can't do that, because they need their own address block that's large enough to be routable, since they're connected to multiple ISPs for reliability and business reasons. But even if we salvaged all of that, it would only gain us a year or two - there's so much growth in Asia that it'd get used up anyway.
I give Microsoft a lot of credit here - they started relatively early working on how to support IPv6 in their operating systems, and while it's not a vanilla feature on XP, you can install it, and they've done enough work on eating their own dogfood that it's as usable as anything they do, especially on Vista, which they'd hoped would be widely deployed by now.
IPv6 has a lot of problems, but the IPv4 address space is going to fall off the edge of the world by 2012 or earlier, so you're either stuck with IPv6 or multiple-NAT, which is at least as ugly.
Nmap fails entertainingly when the average target subnet has 2**64 addresses instead of 1 or 256. There are ways to cheat on that, e.g. exploiting the MAC-based autoaddressing structure (guess that the target has a Dell or HP PC first), but it's still basically difficult. On the other hand, the average user is still going to have a firewall at home, and they don't really care how the little box they had to buy protects them; the real question is whether they had to buy their own or had it built in to their ISP's DSL/cable-modem.
If you're running a small business, 32 externally-visible IPv4 addresses is probably enough, but for a home network, by about 2012 it's going to start costing a good bit more money to have multiple IPv4 addresses, since the world will have run out of them, and ISPs are likely to move toward another layer of NAT for their dynamic-address customers if they're not using IPv6.
For your business environment, I agree with you about DHCPv6 vs. autoconfig and testing a lot of stuff in a virtual environment before running much of it for the real network. And yeah, older equipment is probably going to stay IPv4 for a while. I'm looking forward to seeing how my Corporate Desktop Support Overlords deploy IPv6 for us...
We didn't have firewalls when I was at university either, and for the most part 4-letter passwords were good enough for punchcard accounts:-). I started seeing university firewalls in the mid-late 90s, and the initial ones were more concerned about keeping students from attacking the outside world than keeping the outside world from attacking students, since the traditional firewall models, which assumed that all the dangerous stuff was on the *outside* simply didn't apply.
But no, 99% of the world isn't protected by NAT. 99% of the world is protected by little boxes that people bought when they got their network connection or second computer, and the firewalling functions those boxes provide happen to be implemented using NAT. And anybody who's running services at home (which due to brain-damaged ISP policies usually means gamers rather than home web servers) has to haggle with the little box to open holes in it, because NAT breaks the end-to-end model that the Internet is based on. (Why is Skype so popular? Good NAT evasion...) It's good enough for Anonymous Clients, but for IPv6 the world will be much better off if most of the firewall boxes are doing something other than NAT. (On the other hand, most people will still need little boxes to handle ISP IPv6 services, doing stuff like autoaddressing and 6to4 NATs of various flavors, though in some cases the ISP will be supporting them instead of just running dumb DSL-modem services.)
Anonymous Clients don't need names. But if anything, the fact that IPv6 addresses are often hard to remember is a *feature*, because it forces people to deploy name resolution for anything that does need names.
Having an ISP go to all the work of accounting for every device you need an IPv6 address for, all for free, would be charity; ISPs can't afford to do that. They're going to hand you a block of addresses and make you deal with address assignment inside them. One reason IPv6 address space is so big is so that we never have to expand it again (which is what killed the 64-bit addressing proposals), but another reason is so that we can afford a clean separation between the bits that the ISP assigns and the bits that the customer manages, because managing things costs money. The reason many ISPs only give you one IPv4 address today is because IPv4 addresses are scarce and expensive, and you use NAT to put multiple machines behind it (if you want multiple machines) because you can do that without dealing with the ISP, which costs you both money. IPv6 addresses are designed so they'll never be scarce - your ISP gives you one block of addresses and you do whatever you want inside that block. It does mean that your ISP-connection-box needs to be a router, since it's handling a whole block of addresses on your side and not just one, but these days that's cheap.
In early IPv6 planning, the block size was/64 - it was a nice clean round number, and was big enough for a Netware-style autoaddressing which gave the user 16 bits of subnet numbers if they wanted them and 48 bits of MAC hardware address, though it was later decided that autoaddressing should use EUI-64 hardware addresses instead of 48-bit MACs, which means that the boundary needed to shift, since/64 is now only one subnet. The consensus among ISP operations folks these days is that the boundary will probably be to assign/48s to businesses and either/56s or maybe/48s to homes. There may be some ISPs that only want to give you a/64, but there's a high enough fraction of the market that needs multiple subnets because of wireless and other applications that it's an unpopular position. And there are some ISPs that are talking about/56 for smaller businesses and/48 for larger ones, since at that point the differences between what consumers need and what businesses need are more about reliability, billing options, and value-added services such as television or managed PBX services.
You referred to Joe Sysadmin wanting to assign addresses, so you're apparently thinking about a business context - a few devices like routers may need manually assigned addresses, but humans are going to deal with DNS addresses, and the IPv6 applications folks have been working on different tools for managing that for the last decade and a half. The worldview has changed a bit, from the original MAC-based autoaddressing to a more DHCP-centric view, but either way widespread manual addressing is non-scalable and usually silly in the IPv4 world and worse in the IPv6 world. (Doesn't mean that it doesn't happen; I'm currently working on a customer firewall project where we're doing lots of it, but if I were the architect we'd have done it differently...)
And as far as whether you've got a global address space or can assign your own IP addresses that overlap with the outside world, it's global, get used to it, and if you try assigning local addresses that overlap with the rest of the world's addresses, you're going to get spanked. Back in the mid-90s, when IP applications weren't universal (e.g. businesses still ran SNA and NetBEUI and Novell Netware IPX and didn't always have Internet connections on their internal networks), and RFC1918 hadn't been invented, I did have computer-consulting customers who had done internal projects using addresses they'd Just Made Up. "We're a Bank - This'll never connect to the Outside World! Who cares if our address space overlaps with University of Toronto." That was before DHCP, so it really *hurt* when they had to renumber their network:-)
One of the pipe dreams of early-years IPv6 planning was that by handling IPv6 address allocations cleanly and hierarchically, without the leftover IPv4 swamp space, routers could all have a nice clean shiny view of the rest of the world, with far fewer routes and less routing table churn than IPv4. But it ain't happening, folks, because the user requirements that led to much of the IPv4 routing complexity are still there.
Businesses that want their own provider-independent IPv4 space so they don't have to renumber their network or public face when they change ISPs have no incentive to accept provider-allocated space in IPv6, and no expectation that they'll want to do that even after the IPv6 market matures.
Dual-Homing for Reliability is a much more serious technical objection - if you're a business with one ISP, even if you've got physically diverse feeds to protect you from backhoes, if that ISP has routing problems you're toast. It's less common than it used to be in the early 90s, but individual ISPs do still flake occasionally, especially if they do something like install a new router software version without testing it thoroughly enough, something that's going to happen very frequently in the next five years as ISPs deploy IPv6 for real and find all the subtle problems that Cisco and Juniper haven't discovered yet (especially the problems in older hardware...) And dual-homing means multiple routing table entries, so the routing tables keep growing as inbound IP becomes more and more critical for businesses..
The main alternative solution to multiple homing that I've seen is "shim6" - it's a hopeless ugly mess, trying to insert a protocol layer in between IP and TCP/UDP to maintain sessions across multiple IP addresses. It only works well if everybody you're trying to talk to supports it, which will be a long time even if Microsoft, Apple, and Linus all adopt it.
And Hierarchical Routing really only makes sense if the network topology is hierarchical, which it isn't. Not only do ISPs overlap in multiple cities, but large and medium business customers also have locations in multiple cities and often multiple continents, and the routing tables have to cope with that. Joe's Bar can do fine with provider-assigned or geographically-assigned address space, because it's all in one city, but Joe's Multinational can't easily do that - if it gets an address block from its San Francisco ISP, it needs to also have routes so that traffic to its New York office doesn't get shipped to San Francisco first, and especially so routes to its London, Mumbai, and Tokyo offices don't go to SF first. So you still end up with messy routing tables. Some of that complexity has limits to its growth and some doesn't - most of it belongs to maybe the US Fortune 10,000 and the equivalents in Europe, Japan, Southeast Asia, and India - but some of it's likely to keep growing as the world economy expands.
And if you've got AT&T, you've probably also got a bunch of national DSL providers like Speakeasy who can provide you different policies and services on top of the AT&T DSLAMs. It tends to cost a bit more, but you get more choices about static addresses, email/web/etc. service, bandwidth caps, etc. I'm currently using sonic.net on top of AT&T DSL, and get my static addresses and a shell account which I mainly use to run procmail and webmail, and even N years ago when the telco's policies were very restrictive (no web or mail servers at home, no sharing wireless with neighbors, no more than N computers on your line, etc.), sonic's policies were "You're buying service from us because you want a Real Internet Connection - do whatever you want (except spam, of course)." The telco's relaxed a lot since then, but I've still been happy with value-added service.
It is slower than Comcast, but more than fast enough for me, and I'd prefer almost *any* ISP's policies to living with the typical cable modem company's policies.
Just about anywhere in the US that has DSL has far more than two choices of ISP from a price/policy standpoint. Unlike cable modems, where the right technical choice is to do routing all the way down, and where the main technical approach to sharing with multiple ISPs is PPPoE or even uglier things, DSL is fundamentally a Layer 2 protocol that makes it easy to share between different sets of router operators, so issues like address space, bandwidth caps, and port 25 blocking are per-ISP, and there are lots of national ISPs like Speakeasy that can sell you service on top of the telco DSL wire. That doesn't get you competition from a highest-available-speed perspective; you're still limited by the DSLAM hardware and your distance from the central office, but at least you're not stuck with your telco's idea of forward-thinking Internet service policy.
In bigger cities in the US there's also Layer 1 competition, with Covad and occasional other CLEC DSLAM providers who use the telco copper wire, so they may have higher or lower speeds than what the telco offers. And for both cable and DSL, it's possible to have shared services at Layer 8, i.e. wholesale billing arrangements so you're getting your service from Example.Net instead of the telco/cableco, so you may not have the same price caps or policies about sharing or limits on static addresses, but any routing and port-25 blocking gets done by the infrastructure provider. (For Layer 9 reasons, cable companies usually don't offer this, but some telcos are ok with it.)
And of course if you live too far from your telco office, so DSL doesn't reach you, then you're stuck with either cable modem, satellite, or wireless. You don't even need to be near big cities - a friend of mine runs wireless ISP service in rural Wyoming, putting antennas on top of silos and such, though he *really* doesn't want any P2P running on his network because of bandwidth costs and performance impacts.
Because the 128-bit IPv6 address space is so excessively large, there's really no value in hoarding the stuff - an ISP who has a/32 isn't going to try charging you even $1 for each single IPv6 address because they know the market isn't going to fork over $2**96 for it. If you go read the what the network operators are saying when they talk to each other, the original designs ~15 years ago assumed 64 bits for the operator to play with and 64 for the end user, which gave the end user 16 bits for subnets and 48 bits for MAC-address-based auto-addressing, letting you run your network like Netware, but when that got replaced with EUI-64 64-bit-MAC auto-addressing (which is uglier but should never need changing), operators moved to giving users larger blocks.
The general view is that corporate customers get/48s, and home users probably get/56 but some ISPs lean toward/48, though a few ISPs seem to like/64 as a default for home users and/56 or/48 as an extra-price upgrade. (/64 means that either you only get one subnet at home or that you do something fancy with the addressing behind your home firewall, but realistically, as home wireless is becoming near-universal, almost everybody ends up with multiple subnets so working with/64 is a pain.) Operationally, for an ISP, it's a lot simpler and cheaper if you can treat all your users the same way, so you can put one set of instructions on the web site and have one set of scripts for the help desk folks to work from and have troubleshooting processes that work. There are a few people advocating/60 for home networks, but the value of aligning addresses on byte boundaries vs. the added complexity and minimal savings means that they get shouted down rapidly.
Ok, you probably need some kind of reverse-DNS tag or MS Domain tag if you're using those, but what function are you trying to achieve with the names? Debugging? Boring names are fine. It's especially strange if your users have laptops, so the workstations are moving around, or if your client machines are virtual.
Naming servers makes sense, though once you've got enough of those, the names start to get boring (my print server is mo3980; it's somewhere in Missouri...) If you've got roughly one client machine per user, give them a user-centric name (mine's usually either bstewart or my email id, depending on which of our corporate underlords is running the desktop support this year.) If you've got lots of machines per user, whether real or virtual, make them username1, username2, etc. or else give the user a subdomain and let them name them, so it's fnord.username.engineering.example.com.
Naming machines in a shared lab sometimes makes sense, if different people need a bunch of clients at different times. My current network lab has some routers named after cities, some after baseball teams, etc., but mostly they don't have interesting names.
"Up is down today" "Charm is acting as Strange as last week"
Back during the pre-internet UUCP days, and even for a while after that, we had lots of series of machine names - mountains, beers, dwarves, dwarfs (i.e. Tolkien vs. Disney), composers. My wife's testing department named machines after psychoactive substances, anything from Ritalin to speed to Prozac to coke. At one point I was considering naming machines after common appliances (toaster, mrcoffee, xerox, etc.), but aside from the Trademark Police, that became less practical once some of those companies started networking their hardware.
Slashdot Mirror
If the announcement is based on actual technical calculations (as opposed to getting-a-government-grant calculations or news-reporter-misinterpreted calculations), they're expecting the system will produce power for $21/watt (including research & development costs) - so $21K/kwh. At $0.10/kwh, that means they'd need to run for 210k hours to break even, which is unlikely; even at $0.30/kWh that'd be 70K hours, or a bit under 9 years.
On the other hand, if that's $16B for R&D and $5B for production of a reproducible solution (unlikely?) that's closer to viable.
I think the big question is the power transmission part; solar panel technology is improving, and presumably part of the research is how to develop cheaper launch capabilities.
I don't have mod points today, but I did follow the poster's link. I think the proper moderation would be -1 Crackpot as opposed to -1 Troll, but as the poster says, "Soon, though, all that will change!"
Sure, spam is 97% of email, but email's a small fraction of the bits on the Internet, most of which is the web. It doesn't consume anywhere near the resources of Youtube.
If you're in the mail-handling business, it's one of your largest problems, along with storage, reliability, etc. and burns most of your internet bits, but if you're in the general ISP bits, the spam's still not much of your bandwidth compared to the regular web traffic.
What spam really consumes is the attention span of its recipients, and therefore the resources and attention of mail handling providers who want to keep their users from getting spammed. But it's the attention that costs the money.
Practically all the illegal drugs used to be legal; the exceptions are new drugs developed by the pharmaceutical industry that would need FDA approval before selling them, and new drugs that fail the "similar to anything already illegal" test that the US Feds put out a few years ago to deal with designer drugs (which weren't illegal when developed, and the bureaucrats got tired of having to make new rules or get new laws passed to ban them.) Other than that, alcohol's the only major exception, since it used to be legal, got banned in the US, and then got re-legalized. But nobody's been willing to mess with the big killer drug, tobacco, or my favorite dangerous addictive drug, caffeine.
By the way, does anybody know if the LSD limit that the story reported is correct? It's way too low - 0.015mg is 15 micrograms; a typical street dose in Timothy Leary days was 250mcg and today is more like 100mcg, and the threshold for having any effects at all is around 25mcg for most people.
If they were actually talking 15 milligrams (0.015 grams), that'd be 150 doses, so a small-time dealer, or if they meant 0.150mg, that'd be a medium-sized single dose, which is somewhat believable for a policing limit.
They won't be growing cocaine in their back yards - we don't really have the climate for it. But cannabis and opium poppies can grow well just about anywhere; the only reason people grow them in remote mountains is to avoid getting caught.
Drug addicts don't steal because they're drug users, they steal because they can't get enough money to feed their habits.
You don't see that happening with cheap drugs - tobacco addicts don't go out robbing people to get more cigarettes,
and drunks don't go commit burglary so they can afford a $10 liter bottle of cheap gin, even though tobacco's more addictive than heroin and booze is more destructive.
Heroin and meth addicts steal because of the artificially high costs of black-market drugs, but the drugs themselves don't cost much to produce if you don't have to avoid the law. Medical opiates are cheap - a $5 over-the-counter bottle of codeine would be enough to keep Rush Limbaugh happy for the day, except that it's mixed with acetaminophen (and in Canada, caffeine) to keep you from overdosing; it'd be just as cheap to make without the additives. And making meth in a pharmaceutical factory doesn't cost significantly more than making Sudafed.
All this crime and violence is because the government has convinced us there's some difference between a junkie and a drunk. What a waste.
Ok, sorry, I shouldn't be snarky about it, but if you go look up IPv8 you'll know what I'm making fun of...
There are a few big chunks that are still being hoarded, but not much; universities have mostly returned their early allocations, and some of them now belong to big ISPs which have legitimate use for them. (For instance, Bell Labs used to have a Class A to support their Cray in Murray Hill, but it's now used for AT&T's ISP customers.)
There's a lot more space out there for small-medium businesses that have Class C /24s that could probably get away with /28s or /29s now, but many of them need to have publicly-routable address blocks since they're connected to multiple ISPs, so they can't use anything smaller. (In the IPv6 world, they could use a /48 just fine.)
But no, it wouldn't save us much time - Asia's using multiple /8s per year for growth.
Yeah, I dealt with that market back in the 1980s, as well as the POSIX efforts, and worked with people who'd dealt with the US Auto Industry's attempts at standardizing on OSI as well. Five years ago, that was pretty much what government support for IPv6 looked like also, but it's different now, because the world really will run out of IPv4 addresses by about 2010-2012, so they really do need to do more than just hide behind NAT. And over the last few years their efforts to get testbeds run and actual operational experience have meant that ISPs and developers have some experience dealing with IPv6 now, and are starting to get ready to have to convert.
I went through this in the 1980s with the government trying to push their users to OSI and POSIX - every RFP had a check box saying whether you supported it, and sometimes you had to write reams of explanation about how you were going to do it, but for the most part the end users got waivers and ran MS-DOS on their underpowered desktops and deployed TCP/IP in any networks that weren't running SNA.
But it's different this time, really, trust us! Ok, 5 years ago it wasn't, but by now IPv4 space is close enough to running out, even for agencies that can live behind firewalls like the military does, that they're starting to get serious about using this stuff, and by requiring it they're forcing development of the tools it takes to use IPv6 in the real world and not just on paper. And they helped get Microsoft to build IPv6 support in Windows to sell to them as well as to the real market.
Don't know what town you live in, but any ISP that doesn't have a really solid IPv6 plan by now is going to be toast by 2012 if not before. Either they won't be able to get enough IP addresses, or they're going to have to set up big tunnel servers and learn how to use them in a hurry, or start carrier-NATing their customers which starts to break applications. If IPv6 were something you could simply enable overnight it'd be different, but it's not - you're going to have a significant learning curve as you retrain all your people and rebuild all your management tools and replace any equipment that's not fast enough to run IPv6.
I'd expect that it would get even more exciting for people in the hosting business - not only do you have to keep getting more IP addresses as people virtualize their hosts, but as there start to be IPv6-only end users who want to reach websites, your hosting customers are going to want to have IPv6 support, and if you can't at least give them dual-stack, they're going to become somebody else's web hosting customer instead. You don't happen to have any Layer 3 switching gear in your hosting centers or manage load-balancers for customers do you? Good luck!
Umm, no, thanks for playing.
Disclaimer: This is my own personal commentary, not the opinion of my employer, other companies they own, or my other corporate overlords. But it's still correct.
All major ISPs are thinking about IPv6, most minor ISPs are thinking about it, and any that aren't thinking about it are going to be in serious trouble by 2012 or before. A few years ago, what they were thinking was mostly [Expletive deleted!], but most of them are a bit more focused by now and the rest won't outlive the Mayan calendar rollover. Sure, some of us own large netblocks that will it may be possible to monetize, but just about everybody who does also owns large Cisco routers that will need to be replaced, which costs a lot more money, and in many cases it's the whole collection of operation support software that needs to be updated, which is even harder than just upgrading all your capital equipment. This is basically about as big a problem for a typical ISP as Y2K was - there are 32-bit address fields embedded in all sorts of things, so it's not as simple as just putting new values into existing tables.
Sometimes you can get around it by shoving your dynamically-addressed users behind another layer of NAT, and adding some IPv6 tunnel devices, but if your DSL or cable modem users require a hardware upgrade to support IPv6, it's potentially going to cost more in support to handle the transition than you'd gain by charging them extra to remain on IPv4, even if you make them buy the box. And if you're a business ISP, and your customer with a T1 line has been running fine on a Cisco 2500-series router for the last decade, well, time's up on that hardware. Most newer CPE routers can handle IPv6 fine, though if you've got a user with a T3 or E3 line (45 or 34 Mbps), you may find that it can't go full wire speed with IPv6.
There are a very few universities that still have big netblocks they haven't given back, but there's been enough market for them for the last decade that anybody who didn't need theirs has had plenty of incentive to sell them already. A much larger impact is the number of businesses who have a /24 address block when NAT and firewalls mean that they could get by just fine with a /28 or /29 to handle an inbound web server and VPN or two. In many cases, they can't do that, because they need their own address block that's large enough to be routable, since they're connected to multiple ISPs for reliability and business reasons. But even if we salvaged all of that, it would only gain us a year or two - there's so much growth in Asia that it'd get used up anyway.
I give Microsoft a lot of credit here - they started relatively early working on how to support IPv6 in their operating systems, and while it's not a vanilla feature on XP, you can install it, and they've done enough work on eating their own dogfood that it's as usable as anything they do, especially on Vista, which they'd hoped would be widely deployed by now.
Oh, wait, wrong meme...
IPv6 has a lot of problems, but the IPv4 address space is going to fall off the edge of the world by 2012 or earlier, so you're either stuck with IPv6 or multiple-NAT, which is at least as ugly.
Nmap fails entertainingly when the average target subnet has 2**64 addresses instead of 1 or 256. There are ways to cheat on that, e.g. exploiting the MAC-based autoaddressing structure (guess that the target has a Dell or HP PC first), but it's still basically difficult. On the other hand, the average user is still going to have a firewall at home, and they don't really care how the little box they had to buy protects them; the real question is whether they had to buy their own or had it built in to their ISP's DSL/cable-modem.
If you're running a small business, 32 externally-visible IPv4 addresses is probably enough, but for a home network, by about 2012 it's going to start costing a good bit more money to have multiple IPv4 addresses, since the world will have run out of them, and ISPs are likely to move toward another layer of NAT for their dynamic-address customers if they're not using IPv6.
For your business environment, I agree with you about DHCPv6 vs. autoconfig and testing a lot of stuff in a virtual environment before running much of it for the real network. And yeah, older equipment is probably going to stay IPv4 for a while. I'm looking forward to seeing how my Corporate Desktop Support Overlords deploy IPv6 for us...
We didn't have firewalls when I was at university either, and for the most part 4-letter passwords were good enough for punchcard accounts :-). I started seeing university firewalls in the mid-late 90s, and the initial ones were more concerned about keeping students from attacking the outside world than keeping the outside world from attacking students, since the traditional firewall models, which assumed that all the dangerous stuff was on the *outside* simply didn't apply.
But no, 99% of the world isn't protected by NAT. 99% of the world is protected by little boxes that people bought when they got their network connection or second computer, and the firewalling functions those boxes provide happen to be implemented using NAT. And anybody who's running services at home (which due to brain-damaged ISP policies usually means gamers rather than home web servers) has to haggle with the little box to open holes in it, because NAT breaks the end-to-end model that the Internet is based on. (Why is Skype so popular? Good NAT evasion...) It's good enough for Anonymous Clients, but for IPv6 the world will be much better off if most of the firewall boxes are doing something other than NAT. (On the other hand, most people will still need little boxes to handle ISP IPv6 services, doing stuff like autoaddressing and 6to4 NATs of various flavors, though in some cases the ISP will be supporting them instead of just running dumb DSL-modem services.)
Anonymous Clients don't need names. But if anything, the fact that IPv6 addresses are often hard to remember is a *feature*, because it forces people to deploy name resolution for anything that does need names.
Having an ISP go to all the work of accounting for every device you need an IPv6 address for, all for free, would be charity; ISPs can't afford to do that. They're going to hand you a block of addresses and make you deal with address assignment inside them. One reason IPv6 address space is so big is so that we never have to expand it again (which is what killed the 64-bit addressing proposals), but another reason is so that we can afford a clean separation between the bits that the ISP assigns and the bits that the customer manages, because managing things costs money. The reason many ISPs only give you one IPv4 address today is because IPv4 addresses are scarce and expensive, and you use NAT to put multiple machines behind it (if you want multiple machines) because you can do that without dealing with the ISP, which costs you both money. IPv6 addresses are designed so they'll never be scarce - your ISP gives you one block of addresses and you do whatever you want inside that block. It does mean that your ISP-connection-box needs to be a router, since it's handling a whole block of addresses on your side and not just one, but these days that's cheap.
In early IPv6 planning, the block size was /64 - it was a nice clean round number, and was big enough for a Netware-style autoaddressing which gave the user 16 bits of subnet numbers if they wanted them and 48 bits of MAC hardware address, though it was later decided that autoaddressing should use EUI-64 hardware addresses instead of 48-bit MACs, which means that the boundary needed to shift, since /64 is now only one subnet. /48s to businesses and either /56s or maybe /48s to homes. There may be some ISPs that only want to give you a /64, but there's a high enough fraction of the market that needs multiple subnets because of wireless and other applications that it's an unpopular position. And there are some ISPs that are talking about /56 for smaller businesses and /48 for larger ones, since at that point the differences between what consumers need and what businesses need are more about reliability, billing options, and value-added services such as television or managed PBX services.
The consensus among ISP operations folks these days is that the boundary will probably be to assign
You referred to Joe Sysadmin wanting to assign addresses, so you're apparently thinking about a business context - a few devices like routers may need manually assigned addresses, but humans are going to deal with DNS addresses, and the IPv6 applications folks have been working on different tools for managing that for the last decade and a half. The worldview has changed a bit, from the original MAC-based autoaddressing to a more DHCP-centric view, but either way widespread manual addressing is non-scalable and usually silly in the IPv4 world and worse in the IPv6 world. (Doesn't mean that it doesn't happen; I'm currently working on a customer firewall project where we're doing lots of it, but if I were the architect we'd have done it differently...)
And as far as whether you've got a global address space or can assign your own IP addresses that overlap with the outside world, it's global, get used to it, and if you try assigning local addresses that overlap with the rest of the world's addresses, you're going to get spanked. Back in the mid-90s, when IP applications weren't universal (e.g. businesses still ran SNA and NetBEUI and Novell Netware IPX and didn't always have Internet connections on their internal networks), and RFC1918 hadn't been invented, I did have computer-consulting customers who had done internal projects using addresses they'd Just Made Up. "We're a Bank - This'll never connect to the Outside World! Who cares if our address space overlaps with University of Toronto." That was before DHCP, so it really *hurt* when they had to renumber their network :-)
One of the pipe dreams of early-years IPv6 planning was that by handling IPv6 address allocations cleanly and hierarchically, without the leftover IPv4 swamp space, routers could all have a nice clean shiny view of the rest of the world, with far fewer routes and less routing table churn than IPv4. But it ain't happening, folks, because the user requirements that led to much of the IPv4 routing complexity are still there.
And if you've got AT&T, you've probably also got a bunch of national DSL providers like Speakeasy who can provide you different policies and services on top of the AT&T DSLAMs. It tends to cost a bit more, but you get more choices about static addresses, email/web/etc. service, bandwidth caps, etc. I'm currently using sonic.net on top of AT&T DSL, and get my static addresses and a shell account which I mainly use to run procmail and webmail, and even N years ago when the telco's policies were very restrictive (no web or mail servers at home, no sharing wireless with neighbors, no more than N computers on your line, etc.), sonic's policies were "You're buying service from us because you want a Real Internet Connection - do whatever you want (except spam, of course)." The telco's relaxed a lot since then, but I've still been happy with value-added service.
It is slower than Comcast, but more than fast enough for me, and I'd prefer almost *any* ISP's policies to living with the typical cable modem company's policies.
Just about anywhere in the US that has DSL has far more than two choices of ISP from a price/policy standpoint. Unlike cable modems, where the right technical choice is to do routing all the way down, and where the main technical approach to sharing with multiple ISPs is PPPoE or even uglier things, DSL is fundamentally a Layer 2 protocol that makes it easy to share between different sets of router operators, so issues like address space, bandwidth caps, and port 25 blocking are per-ISP, and there are lots of national ISPs like Speakeasy that can sell you service on top of the telco DSL wire. That doesn't get you competition from a highest-available-speed perspective; you're still limited by the DSLAM hardware and your distance from the central office, but at least you're not stuck with your telco's idea of forward-thinking Internet service policy.
In bigger cities in the US there's also Layer 1 competition, with Covad and occasional other CLEC DSLAM providers who use the telco copper wire, so they may have higher or lower speeds than what the telco offers. And for both cable and DSL, it's possible to have shared services at Layer 8, i.e. wholesale billing arrangements so you're getting your service from Example.Net instead of the telco/cableco, so you may not have the same price caps or policies about sharing or limits on static addresses, but any routing and port-25 blocking gets done by the infrastructure provider. (For Layer 9 reasons, cable companies usually don't offer this, but some telcos are ok with it.)
And of course if you live too far from your telco office, so DSL doesn't reach you, then you're stuck with either cable modem, satellite, or wireless. You don't even need to be near big cities - a friend of mine runs wireless ISP service in rural Wyoming, putting antennas on top of silos and such, though he *really* doesn't want any P2P running on his network because of bandwidth costs and performance impacts.
Because the 128-bit IPv6 address space is so excessively large, there's really no value in hoarding the stuff - an ISP who has a /32 isn't going to try charging you even $1 for each single IPv6 address because they know the market isn't going to fork over $2**96 for it. If you go read the what the network operators are saying when they talk to each other, the original designs ~15 years ago assumed 64 bits for the operator to play with and 64 for the end user, which gave the end user 16 bits for subnets and 48 bits for MAC-address-based auto-addressing, letting you run your network like Netware, but when that got replaced with EUI-64 64-bit-MAC auto-addressing (which is uglier but should never need changing), operators moved to giving users larger blocks.
The general view is that corporate customers get /48s, and home users probably get /56 but some ISPs lean toward /48, though a few ISPs seem to like /64 as a default for home users and /56 or /48 as an extra-price upgrade. ( /64 means that either you only get one subnet at home or that you do something fancy with the addressing behind your home firewall, but realistically, as home wireless is becoming near-universal, almost everybody ends up with multiple subnets so working with /64 is a pain.) Operationally, for an ISP, it's a lot simpler and cheaper if you can treat all your users the same way, so you can put one set of instructions on the web site and have one set of scripts for the help desk folks to work from and have troubleshooting processes that work. There are a few people advocating /60 for home networks, but the value of aligning addresses on byte boundaries vs. the added complexity and minimal savings means that they get shouted down rapidly.
I hope the crackers were polite enough to give it one....
Ok, you probably need some kind of reverse-DNS tag or MS Domain tag if you're using those, but what function are you trying to achieve with the names? Debugging? Boring names are fine. It's especially strange if your users have laptops, so the workstations are moving around, or if your client machines are virtual.
Naming servers makes sense, though once you've got enough of those, the names start to get boring (my print server is mo3980; it's somewhere in Missouri...) If you've got roughly one client machine per user, give them a user-centric name (mine's usually either bstewart or my email id, depending on which of our corporate underlords is running the desktop support this year.) If you've got lots of machines per user, whether real or virtual, make them username1, username2, etc. or else give the user a subdomain and let them name them, so it's fnord.username.engineering.example.com.
Naming machines in a shared lab sometimes makes sense, if different people need a bunch of clients at different times. My current network lab has some routers named after cities, some after baseball teams, etc., but mostly they don't have interesting names.
"Up is down today" "Charm is acting as Strange as last week"
Back during the pre-internet UUCP days, and even for a while after that, we had lots of series of machine names - mountains, beers, dwarves, dwarfs (i.e. Tolkien vs. Disney), composers. My wife's testing department named machines after psychoactive substances, anything from Ritalin to speed to Prozac to coke. At one point I was considering naming machines after common appliances (toaster, mrcoffee, xerox, etc.), but aside from the Trademark Police, that became less practical once some of those companies started networking their hardware.