I tried going to Oracle's web site and typing "guarantee" into their search box. I got back:
Database Log In Failed
TNS is unable to connect to destination. Invalid TNS address supplied or destination is not listening. This error
can also occur because of underlying network transport problems.
Verify that the TNS name in the connectstring entry of the DAD for this URL is valid and the database listener
is running.
Kinda puts the whole reliability thing in perspective...
I have a Pacific Bell DSL line, running my own mail server with my own domain name (actually a subdomain of stanford.edu). The problem isn't on Pac Bell's side; my parents use Earthlink, and my email to them was bouncing. Some investigation showed that they had configured their mail servers to reject any mail traffic from Pac Bell IP addresses other than the Pac Bell mail servers. This was an explicit decision on their part, again with the motivation of "reducing spam."
Fortunately, I was able to relay my SMTP traffic through Stanford's mail server (since I'm using a valid *.stanford.edu address) for each set of mail destinations that does this access control.
I think it's pretty stupid to assume that a DSL line is going to be using the ISP's email services as well--- especially since Earthlink has no problem _delivering_ mail to that account.
Re:The question is not whether there is a problem
on
Information Poisoning
·
· Score: 1
I have no doubt that it _can_ be solved using the government--- just that it shouldn't be. However, the author of the article seems to think it will be easy; he doesn't seem to comprehend the nature and scope of the Web.
Good point. My wife expresses a similar opinion in her article "DARE to Admit It Doesn't Work". DARE is a joke in many schools. It focuses student's lives around being "drug-free"--- and then misinforms them about the effects of drugs. When students finally realize they've been treated like idiots, there is bound to be some backlash.
Hmph. My CS BA got me into the Stanford PhD program. And from my experience on the teaching and TAing end of the undergrad program here, I think I got a better CS education, too.
I think the reputation of the school matters a lot more than the difference between a BA and a BS--- most in-state employers are perfectly happy to take liberal-arts graduates; there's not a problem competing with BSes from the state university.
I believe another aspect of the problem is that the Intel x86 is not quite virtualizable, so you need to patch code that uses non-virtualizable intstructions at runtime. These are generally operations that would be used in kernel mode only (or are only non-virtualizable in this case) so I could see there being different requirements for "sufficient" virtualization of different OSes.
Why do you think switching (or routing) IPv6 is easier than switching IPv4? The only improvement at this level claimed by IPv6 is a stricter address allocation hierarchy. So routing tables might be smaller in the long run. As far as I can see, it's always going to be easier to route IPv4 packets than IPv6 packets, given an equal number of routing entries, simply because of the smaller address size.
Excellent point, thank you for making it. The deployment of an IPv6-aware DNS server is just one small step. It doesn't address the larger issues involved in deploying IPv6. And I'm somewhat annoyed at CmdrTaco for implying that it does. If all people want is DNSSEC, then that's all they're going to install and configure--- the fact that the software can handle IPv6 is going to be of very limited interest to them.
Or possibly even a source of annoyance if their software starts sending out v6 address requests before looking for the v4 address. I know somebody who has gotten burned by this--- he upgraded his system to support IPv6. The name lookup tries AAAA first, then A. Well, Stanford's load-balancing DNS server returns the wrong thing to the first request ("name not present", basically, rather than "that name exists but we don't have any v6 addresses"), so the nameserver caches the negative answer and returns it in response to the 'A' query as well. Oops, suddenly he can't log into the computer cluster using the normal domain name. It's true that this is a bug with the load-balancing software, not IPv6. It's just yet another hurdle to overcome.
To talk to an IPv4 network interface you need an IPv4 address. If you don't have one (i.e., you're a v6-only host), the only way to get one is to go through a protocol-translator. Basically, a NAT box that multiplexes the same v4 machine to multiple v6 addresses. Which makes the decision to go to v6 rather than using NAT (especially if you already are using private addresses) somewhat less appealing, in my opinion.
Mandatory IP security
Which could be one of the things holding v6 deployment back... If end-to-end IPSEC hasn't been deployed in IPv4, simply "mandating" it in IPv6 doesn't make it easier to do. It just ensures that any IPv6 host might accept IPSEC packets from you--- about what you can assume today. (The mere presence of IPSEC capability on a host says nothing about your ability to use IPSEC to communicate with them.)
Mandatory Quality of Service
Lots of IPv6 marketing claims "improved QoS". Most of these claims seem to be based on the presence of a "flow identifier" field in the IPv6 header. A flow identifier alone doesn't do any good without some system to identify meaning to it. Again, QoS doesn't become easier to implement and deploy just because IPv6 requires it. Just because the feature's there doesn't mean the router does anything meaningful with it. (I would welcome any correction--- perhaps I just haven't seen the relevant specification?)
Guaranteed mobile IP addressing
Sure, IPv6 mobility is a better design than current IPv4 mobility. (Now, I happen to think that mobility needs to be handled above the IP layer, but that's an argument for another time...) But there's nothing "guaranteed" about it. In fact, IPv6 mobility depends on having a "AAA" structure, the design of which is still being worked out. Even with the architecture there, "guaranteed" is too strong a word--- only a network's willingness to provide the service gives you the ability to use it.
Automatic Network Configuration
"for Hierarchies?" I don't understand that. Many people view autoconfiguration of devices as a useful thing. But there's no direct support for ad-hoc networking, which is what I tend to think of as "automatica network configuration".
Simpler Headers == Quicker Routing
Software implementation == magnitudes slower routing. So you see a big disadvantage to going to "native" IPv6 until people start creating forwarding hardware for it. Which is expensive, since now you have to have a 128-bit data path (or 256-bit, if you route on source and destination--- as you do for multicast) rather than a 32-bit or 64-bit data path. Perhaps you're referring to the simpler option design? I admit it's worlds better. But in the real world, most packets don't come with options (and those that do get punted up to software), so the real cost is routing lookups. IPv6 claims to make routing tables smaller (using the strict addressing hierarchy)--- we'll see--- but even if they stay the same size, the prefix match gets longer--- which requires either more memory accesses or bigger ternary CAMS. No guarantee of quicker routing in any way.
Mandatory Anycasting
I don't like anycast. It's generally not responsive to higher-level failure, but since it's at the network level, you might be stuck with an unresponsive server for a while. Multicast is a better design decision in the local area.
Mandatory Multicasting
We've had (multi-source) multicast for longer than the Web. It hasn't really been deployed worldwide for a variety of reasons. (Hard to route, hard to bill, hard to debug...) Making it "mandatory" only increases wariness about deploying IPv6. Also, single-source multicast (SSM) looks like it may actually go somewhere, has no address shortage, and is much easier to route and debug. But you don't need to go to v6 to use it.
Mandatory Connection Fail-over Support
I must plead ignorance to this one, too. However, IPv6 can make multihoming your network a much more difficult problem, since you receive different address ranges for your machines from each of your ISPs. Yet, the entire IPv6 address is the endpoint identifier. So, esentially, your choice of address locks you into a particular ISP. Various tunnelling designs have been suggested to improve this, but they increase the complexity of the network. (To be fair, it's not too much worse than multihoming in IPv4--- unless, like Stanford, you already have an AS number for BGP and are not likely to get a TLA in IPv6. Why upgrade?)
IDRP Routing Protocol
Again, I must plead ignorance. But why can't this routing protocol (if it's a good idea) be done with v4?
In many states, the deadline for registering to vote has already (or will soon) passed. Unless there's been a surge in voter registration lately (which I haven't heard anything about), the "54%" number is completely bogus. Of course, even many registered voters don't go to the polls... maybe they just see it as a way of signing up for annoygin jury duty?;)
I'm surprised so many in the media ignore this. A vote for Nader is a vote for federal campaign funding for the Greens! Hardly a wasted vote in the long term... that is, if you want the Green party to succeed.
My wife has a simila r argument to yours--- voting for one of the big parties is not going to change their policies. They already have your vote. Voting in large numbers for somebody else is the only way to attract attention.
Well... some Libertarian theories do increase the power of corporations by allowing practically anything to be done with your "consent" (e.g., contractual agreement.) Such a situation tends to reward those with superior bargaining power and legal know-how: i.e., your employer, not you.
The sweatshop example is relevant here. True, people would probably be worse off if the sweatshop closed. Would the sweatshop be worse off if an individual refused to work there? Not likely--- only when large numbers refuse is there an incentive to raise wages or improve conditions. This is, in my opinion, a fundamental imbalance in an employment relationship. (It can be corrected by unionization--- which many conservatives don't seem to like. It seems to me that unionized workers are a necessary dual to non-compete agreements: if my employer can demand I not seek work elsewhere in my field, surely I as a worker can demand that employees only come from my union.)
Don't get me wrong--- I'm very sympathetic to many Libertarian positions, and do feel that the role of government needs to be lessened. But history shows us that corporations are very clever about finding ways to exploit people--- debt slavery, company towns, patents;), intellectual property agreements, mandatory unpaid overtime, etc. Corporations have real force on their side, even if it doesn't take the form of "men with guns".
Corporations are artificial people, not natural ones. The rights they have need not necessarily be the same ones that actual human beings get. I personally wouldn't mind if people (acting as individuals) were unhindered by government regulations, but limited-liability corporations (and thus the people acting as agents of the same) had a decent amount of government oversight.
My primary story is a little more amusing. I and my wife went to the polling place, handed over the cards we had been sent and stated our party affiliations --- Independent and Libertarian --- which were also indicated on the cards.
The guy running the place ignored us and said "two Democrats" to his coworker.
We ended up with the right ballots eventually. But I wonder how many people at that polling place didn't. Ah, the wonders of volunteer labor.
(BTW, California's open primary is a farce, since they still have to print a different ballot for each party--- so that they can report the count on a per-party basis.)
Re:I don't want the archives
on
Deja For Sale
·
· Score: 1
I don't agree that there's any fundamental human right to have your posts forgotten.:) One might argue that your posts belong to you (via copyright law), but the public nature of their distribution would seem to argue against unlimited control by you.
What sort of regulation are you suggesting? If I, personally, keep an archive of every Usenet post you send, would you come sue me? Or would the Internet Police beat down my door?
Like sending an email, posting on Usenet is implicitly giving others the right to view your copyrighted material. While there are (admittedly fuzzy) limits to what I can do with a post, I'd strongly oppose any prior restrction on my rights to archive communication sent to me--- whether individually or as part of a group. So while I can see an argument that my posts can't be republished, the act of archiving itself shouldn't be prohibited. Doing so leads to a view of copyright I (and much of the Slashdot "community") don't particularly like--- the abolition of fair use, tilting the balance too far toward content owners.
On a more practical note, it's essential to preserve electronic communications because we increasingly don't have a paper history. Future historians need to have something to work with--- although they probably won't be interested in any of my e-mails or Usenet posts, they definitely care about President Clinton's. Where in between would the need for privacy require the erasure of history?
I found that interview very enlightening. What, Salon shouldn't publish the interviews they do with stupid people? Exposing the thoughts of the people behind SDMI is useful--- Salon certainly can't be held responsible for those stupid opinions.
Would you raise the same objections to an interview with Jack Valenti because he also said things you know to be untrue and misleading?
Really? I had W2K running on VMware without any problem. Not that I actually did anything useful with it, but I didn't encounter any problem with my RH-6ish, custom kernel build system.
VMWare _is_ an excellent gift idea for geeks, though. I have a friend who uses it for kernel hacking, although unfortunately he hasn't figured out a way to make remote debugging work over the emulated serial console.
Feynman wrote about reversible computing, too. (I really enjoyed reading a collection of his work called "The Physics of Computation.") From what I remember, reversible computing has no energy requirements as long as you're willing to wait arbitrarily long for the result. Since the computation is reversible, the computing process is as likely to go backwards as forwards without any driving force. So, if you actually want an answer within the lifetime of a universe, you do in fact need to use some energy. But IANA physicist, just married to one.
There is a company, Digital Fountain, that has a system for doing bulk software distribution over multicast. (It's pretty cool--- they use an encoding that allows you to collect _any_ N distinct packets from the datastream and recreate the original file.) I don't know whether their software could be used over the MBONE, but hopefully the deployment of single-source multicast (which a lot of people are excited about) will help get things moving toward large-scale availability of multicast. Interested parties should probably check out the SIGCOMM '98 paper.
There are ways to deal with that, though. Multi-layered encodings let you control the bandwidth coming toward you, and there are a multitude of schemes for reliable multicast.
The Digital Fountain approach is one example of a system designed for large-scale software distribution over multicast.
Well, actually DNS wasn't invented at the same time as "the net". The early days of the internet used a shared hosts.txt file that everybody downloaded periodically.
Slashdot Mirror
I tried going to Oracle's web site and typing "guarantee" into their search box. I got back:
Database Log In Failed
TNS is unable to connect to destination. Invalid TNS address supplied or destination is not listening. This error can also occur because of underlying network transport problems.
Verify that the TNS name in the connectstring entry of the DAD for this URL is valid and the database listener is running.
Kinda puts the whole reliability thing in perspective...
I don't want to use a @pacbell.net address. Period. ;) Neither do I feel like paying outrageous fees for a business-class DSL connection.
I have a Pacific Bell DSL line, running my own mail server with my own domain name (actually a subdomain of stanford.edu). The problem isn't on Pac Bell's side; my parents use Earthlink, and my email to them was bouncing. Some investigation showed that they had configured their mail servers to reject any mail traffic from Pac Bell IP addresses other than the Pac Bell mail servers. This was an explicit decision on their part, again with the motivation of "reducing spam."
Fortunately, I was able to relay my SMTP traffic through Stanford's mail server (since I'm using a valid *.stanford.edu address) for each set of mail destinations that does this access control.
I think it's pretty stupid to assume that a DSL line is going to be using the ISP's email services as well--- especially since Earthlink has no problem _delivering_ mail to that account.
I have no doubt that it _can_ be solved using the government--- just that it shouldn't be. However, the author of the article seems to think it will be easy; he doesn't seem to comprehend the nature and scope of the Web.
Good point. My wife expresses a similar opinion in her article "DARE to Admit It Doesn't Work". DARE is a joke in many schools. It focuses student's lives around being "drug-free"--- and then misinforms them about the effects of drugs. When students finally realize they've been treated like idiots, there is bound to be some backlash.
Hmph. My CS BA got me into the Stanford PhD program. And from my experience on the teaching and TAing end of the undergrad program here, I think I got a better CS education, too.
I think the reputation of the school matters a lot more than the difference between a BA and a BS--- most in-state employers are perfectly happy to take liberal-arts graduates; there's not a problem competing with BSes from the state university.
Nerds come in all flavors. :)
I believe another aspect of the problem is that the Intel x86 is not quite virtualizable, so you need to patch code that uses non-virtualizable intstructions at runtime. These are generally operations that would be used in kernel mode only (or are only non-virtualizable in this case) so I could see there being different requirements for "sufficient" virtualization of different OSes.
Why do you think switching (or routing) IPv6 is easier than switching IPv4? The only improvement at this level claimed by IPv6 is a stricter address allocation hierarchy. So routing tables might be smaller in the long run. As far as I can see, it's always going to be easier to route IPv4 packets than IPv6 packets, given an equal number of routing entries, simply because of the smaller address size.
Excellent point, thank you for making it. The deployment of an IPv6-aware DNS server is just one small step. It doesn't address the larger issues involved in deploying IPv6. And I'm somewhat annoyed at CmdrTaco for implying that it does. If all people want is DNSSEC, then that's all they're going to install and configure--- the fact that the software can handle IPv6 is going to be of very limited interest to them.
Or possibly even a source of annoyance if their software starts sending out v6 address requests before looking for the v4 address. I know somebody who has gotten burned by this--- he upgraded his system to support IPv6. The name lookup tries AAAA first, then A. Well, Stanford's load-balancing DNS server returns the wrong thing to the first request ("name not present", basically, rather than "that name exists but we don't have any v6 addresses"), so the nameserver caches the negative answer and returns it in response to the 'A' query as well. Oops, suddenly he can't log into the computer cluster using the normal domain name. It's true that this is a bug with the load-balancing software, not IPv6. It's just yet another hurdle to overcome.
To talk to an IPv4 network interface you need an IPv4 address. If you don't have one (i.e., you're a v6-only host), the only way to get one is to go through a protocol-translator. Basically, a NAT box that multiplexes the same v4 machine to multiple v6 addresses. Which makes the decision to go to v6 rather than using NAT (especially if you already are using private addresses) somewhat less appealing, in my opinion.
Which could be one of the things holding v6 deployment back... If end-to-end IPSEC hasn't been deployed in IPv4, simply "mandating" it in IPv6 doesn't make it easier to do. It just ensures that any IPv6 host might accept IPSEC packets from you--- about what you can assume today. (The mere presence of IPSEC capability on a host says nothing about your ability to use IPSEC to communicate with them.)
Mandatory Quality of Service
Lots of IPv6 marketing claims "improved QoS". Most of these claims seem to be based on the presence of a "flow identifier" field in the IPv6 header. A flow identifier alone doesn't do any good without some system to identify meaning to it. Again, QoS doesn't become easier to implement and deploy just because IPv6 requires it. Just because the feature's there doesn't mean the router does anything meaningful with it. (I would welcome any correction--- perhaps I just haven't seen the relevant specification?)
Guaranteed mobile IP addressing
Sure, IPv6 mobility is a better design than current IPv4 mobility. (Now, I happen to think that mobility needs to be handled above the IP layer, but that's an argument for another time...) But there's nothing "guaranteed" about it. In fact, IPv6 mobility depends on having a "AAA" structure, the design of which is still being worked out. Even with the architecture there, "guaranteed" is too strong a word--- only a network's willingness to provide the service gives you the ability to use it.
Automatic Network Configuration
"for Hierarchies?" I don't understand that. Many people view autoconfiguration of devices as a useful thing. But there's no direct support for ad-hoc networking, which is what I tend to think of as "automatica network configuration".
Simpler Headers == Quicker Routing
Software implementation == magnitudes slower routing. So you see a big disadvantage to going to "native" IPv6 until people start creating forwarding hardware for it. Which is expensive, since now you have to have a 128-bit data path (or 256-bit, if you route on source and destination--- as you do for multicast) rather than a 32-bit or 64-bit data path. Perhaps you're referring to the simpler option design? I admit it's worlds better. But in the real world, most packets don't come with options (and those that do get punted up to software), so the real cost is routing lookups. IPv6 claims to make routing tables smaller (using the strict addressing hierarchy)--- we'll see--- but even if they stay the same size, the prefix match gets longer--- which requires either more memory accesses or bigger ternary CAMS. No guarantee of quicker routing in any way.
Mandatory Anycasting
I don't like anycast. It's generally not responsive to higher-level failure, but since it's at the network level, you might be stuck with an unresponsive server for a while. Multicast is a better design decision in the local area.
Mandatory Multicasting
We've had (multi-source) multicast for longer than the Web. It hasn't really been deployed worldwide for a variety of reasons. (Hard to route, hard to bill, hard to debug...) Making it "mandatory" only increases wariness about deploying IPv6. Also, single-source multicast (SSM) looks like it may actually go somewhere, has no address shortage, and is much easier to route and debug. But you don't need to go to v6 to use it.
Mandatory Connection Fail-over Support
I must plead ignorance to this one, too. However, IPv6 can make multihoming your network a much more difficult problem, since you receive different address ranges for your machines from each of your ISPs. Yet, the entire IPv6 address is the endpoint identifier. So, esentially, your choice of address locks you into a particular ISP. Various tunnelling designs have been suggested to improve this, but they increase the complexity of the network. (To be fair, it's not too much worse than multihoming in IPv4--- unless, like Stanford, you already have an AS number for BGP and are not likely to get a TLA in IPv6. Why upgrade?)
IDRP Routing Protocol
Again, I must plead ignorance. But why can't this routing protocol (if it's a good idea) be done with v4?
In many states, the deadline for registering to vote has already (or will soon) passed. Unless there's been a surge in voter registration lately (which I haven't heard anything about), the "54%" number is completely bogus. Of course, even many registered voters don't go to the polls... maybe they just see it as a way of signing up for annoygin jury duty? ;)
(Sorry, posted anonymously by mistake.)
I'm surprised so many in the media ignore this. A vote for Nader is a vote for federal campaign funding for the Greens! Hardly a wasted vote in the long term... that is, if you want the Green party to succeed.
My wife has a simila r argument to yours--- voting for one of the big parties is not going to change their policies. They already have your vote. Voting in large numbers for somebody else is the only way to attract attention.
Well... some Libertarian theories do increase the power of corporations by allowing practically anything to be done with your "consent" (e.g., contractual agreement.) Such a situation tends to reward those with superior bargaining power and legal know-how: i.e., your employer, not you.
The sweatshop example is relevant here. True, people would probably be worse off if the sweatshop closed. Would the sweatshop be worse off if an individual refused to work there? Not likely--- only when large numbers refuse is there an incentive to raise wages or improve conditions. This is, in my opinion, a fundamental imbalance in an employment relationship. (It can be corrected by unionization--- which many conservatives don't seem to like. It seems to me that unionized workers are a necessary dual to non-compete agreements: if my employer can demand I not seek work elsewhere in my field, surely I as a worker can demand that employees only come from my union.)
Don't get me wrong--- I'm very sympathetic to many Libertarian positions, and do feel that the role of government needs to be lessened. But history shows us that corporations are very clever about finding ways to exploit people--- debt slavery, company towns, patents ;), intellectual property agreements, mandatory unpaid overtime, etc. Corporations have real force on their side, even if it doesn't take the form of "men with guns".
Corporations are artificial people, not natural ones. The rights they have need not necessarily be the same ones that actual human beings get. I personally wouldn't mind if people (acting as individuals) were unhindered by government regulations, but limited-liability corporations (and thus the people acting as agents of the same) had a decent amount of government oversight.
My primary story is a little more amusing. I and my wife went to the polling place, handed over the cards we had been sent and stated our party affiliations --- Independent and Libertarian --- which were also indicated on the cards.
The guy running the place ignored us and said "two Democrats" to his coworker.
We ended up with the right ballots eventually. But I wonder how many people at that polling place didn't. Ah, the wonders of volunteer labor.
(BTW, California's open primary is a farce, since they still have to print a different ballot for each party--- so that they can report the count on a per-party basis.)
I don't agree that there's any fundamental human right to have your posts forgotten. :) One might argue that your posts belong to you (via copyright law), but the public nature of their distribution would seem to argue against unlimited control by you.
What sort of regulation are you suggesting? If I, personally, keep an archive of every Usenet post you send, would you come sue me? Or would the Internet Police beat down my door?
Like sending an email, posting on Usenet is implicitly giving others the right to view your copyrighted material. While there are (admittedly fuzzy) limits to what I can do with a post, I'd strongly oppose any prior restrction on my rights to archive communication sent to me--- whether individually or as part of a group. So while I can see an argument that my posts can't be republished, the act of archiving itself shouldn't be prohibited. Doing so leads to a view of copyright I (and much of the Slashdot "community") don't particularly like--- the abolition of fair use, tilting the balance too far toward content owners.
On a more practical note, it's essential to preserve electronic communications because we increasingly don't have a paper history. Future historians need to have something to work with--- although they probably won't be interested in any of my e-mails or Usenet posts, they definitely care about President Clinton's. Where in between would the need for privacy require the erasure of history?
I found that interview very enlightening. What, Salon shouldn't publish the interviews they do with stupid people? Exposing the thoughts of the people behind SDMI is useful--- Salon certainly can't be held responsible for those stupid opinions.
Would you raise the same objections to an interview with Jack Valenti because he also said things you know to be untrue and misleading?
Really? I had W2K running on VMware without any problem. Not that I actually did anything useful with it, but I didn't encounter any problem with my RH-6ish, custom kernel build system.
VMWare _is_ an excellent gift idea for geeks, though. I have a friend who uses it for kernel hacking, although unfortunately he hasn't figured out a way to make remote debugging work over the emulated serial console.
Feynman wrote about reversible computing, too. (I really enjoyed reading a collection of his work called "The Physics of Computation.") From what I remember, reversible computing has no energy requirements as long as you're willing to wait arbitrarily long for the result. Since the computation is reversible, the computing process is as likely to go backwards as forwards without any driving force. So, if you actually want an answer within the lifetime of a universe, you do in fact need to use some energy. But IANA physicist, just married to one.
There is a company, Digital Fountain, that has a system for doing bulk software distribution over multicast. (It's pretty cool--- they use an encoding that allows you to collect _any_ N distinct packets from the datastream and recreate the original file.) I don't know whether their software could be used over the MBONE, but hopefully the deployment of single-source multicast (which a lot of people are excited about) will help get things moving toward large-scale availability of multicast. Interested parties should probably check out the SIGCOMM '98 paper.
There are ways to deal with that, though. Multi-layered encodings let you control the bandwidth coming toward you, and there are a multitude of schemes for reliable multicast.
The Digital Fountain approach is one example of a system designed for large-scale software distribution over multicast.
Holy Brooks' Law, Batman! Or maybe it's Alioth's Law: adding more hours to a project just makes the demand for labor greater.
Well, actually DNS wasn't invented at the same time as "the net". The early days of the internet used a shared hosts.txt file that everybody downloaded periodically.
Hehe... yeah, even my numbers are probably an underestimate.
Oh--- and a single platform also makes it easier to have lab hours. Otherwise the TA has to split his time among several buildings...