For anyone looking for even more depth, Tanenbaum's chapter on Media Access Control also talks about ALOHA in great detail before moving on to Ethernet, which is based on the work done by the University of Hawaii's packet radio experiments.
Interestingly enough, like in Antarctica, we can also provide these pioneers with vast amounts of porn. That's pretty easy to do. OTOH, I'm not sure if that would make things better or worse in the long run.
I always took that phrase to be two mutually-exclusive categories; not one statement qualifiying the other. E.g., "News for nerds AND stuff that matters", not "News for nerds BUT JUST stuff that matters".
Not true. There's a one-to-one correspondence between the width of the band and the symbol rate. One of your symbols could be "The Declaration of Independence in its entirety". Whatever meaning you ascribe to the symbol is up to you. The relationship between the bitrate and the bandwidth is more complex, because it is the result of many functions. If people want to conflate "bandwidth" and "baud rate", fine (it's sort of like conflating "wavelength" and "frequency"), but "bitrate" and "bandwidth" are only related insofar as one depends on the characteristics of the other.
Technical conversations are the exception to your "language drift" rule, by necessity. The common meaning must be precise, otherwise important subtleties are lost. Anyone who has done at least a semester of college physics will remember having to relearn words like "speed" and "weight", because regular people do not use those words the same way that technicians do.
You make fun, but I have a friend who actually was a firefighter in Antarctica, at McMurdo. It's not an easy job, especially when you consider that all of your normal methods for putting a fire out won't work there. We humans have the darndest ability to make fire wherever we go. This particular fellow is now a firefighter in Iraq. Apparently Antarctica was too easy for him.
The guy who works in the office next to my brother figures out how to put fires out in zero-g. Fortunately (although not so fortunate for the three people who died) NASA had an early experience with fire that made them realize they would need to address the problem before anyone went into space.
We've already done a great job at butchering the term anyway. Most usage doesn't correspond with reality.
"bandwidth" is a function of the physical characteristcs of the medium. I.e., with wires, the impedance goes up as the wire gets longer. This changes the wire's ability to transmit high frequencies (it "attenuates" the signal). Even dictionaries get this wrong.
"baud" or "symbol rate" is a function of bandwidth and your modulation scheme.
"bitrate" is a function of the symbol rate, and also depends on things like your packet structures and encoding schemes (actually, it depends on a LOT of things).
"speed" is not a technical term in this context, but most "technical" people equate it with "bitrate". For most other people, "speed" means "how long do I have to wait?"
:end rant
(if you can't complain about this kind of stuff on a website billing itself as "news for nerds", where can you complain?)
But it just goes to show that carriers feel no need to compete. Most of us have no ability to choose the products we want from them, and with Uncle Sam's help, they can keep us from seeing how lame they really are.
But suddenly I say: and some people want this same government in charge of our military and now I'll be modded troll into oblivion.
Fixed that for ya. Oh wait, still a dumb thing to say...?
The government is a very large and diverse group of people. Some of those people do legitimately deserve to be criticized, but many, many, many of them do not. They do their jobs daily and with excellence, often for little compensation.
To infer that the government would be bad at managing health care because of a single instance of idiotic training materials is an example of woefully poor logic...
The problem with cellphone audio is not that it isn't properly compressed. It is. The problem is that telephone connections are extremely sensitive to jitter (or more precisely, humans are very sensitive to jitter). In fact, cellphone use the very same technique that MP3 uses, which is a perceptual codec. It is true that they also attenuate the signal somewhat, but this is mainly because designing A-D hardware that works over a wider range is more expensive, and for the most part, pointless. Who cares if you have booming bass on your telephone? MP3 is actually worse when it comes to jitter, because MPEG is VERY dependent on the last frame having come through correctly, until you hit a keyframe.
In some ways, the distinction between VoIP and traditional business phone systems has been blurring anyway. For the last 20 years or so businesses have gotten their phone lines via T1/E1 trunks which then go into a computer switching system (PBX). In the past, those switching systems ran custom/proprietary software, but now, you're starting to see them run things like Asterisk. Nowadays, people will either push the POTS service down to the ISP and receive SIP trunks (i.e., VoIP) in return, or else add SIP trunks via their existing Internet connection and run the gateway service themselves (that's what we're doing). Business phone connections haven't really been "plain old copper" for a long time.
I don't think you're going to see people give up NATs easily. NAT is a bona fide security feature, not just a consequence of having a LAN. This is the same thing that makes detecting bad segmentation faults possible in an operating system, and from that perspective, a separate address space is very desirable.
Any kind of 'fundamental change' that happens on the Internet needs to accept that NATs part of good architecture. You really want your toaster on the same address space as your Cray?
Maybe that's true in other places, but in MA, the last time I was in the market for a car, I found that the reason I couldn't get a diesel is because very few of them pass MA's SO2 emissions. That's for new cars, and so I found that there was a bit of a cottage industry for few-weeks-old diesels that had been bought out-of-state and transported in. I was specifically looking for a VW Golf TDI, and this was around 2005, so things may have changed since then. I didn't end up getting one because financing through a bank for such a transaction was... complicated.
If we're to believe the copyright lobby's claims that downloading music without paying is 'stealing', then I think we can also conclude that 'stealing' , in some cases, hurts no one.
Murder is defined as being an "unlawful killing". If a court rules that a person is not guilty of murder, then that person is not a "murderer", by definition. If the person's role in the killing is unambiguous, that person is a "killer", and their action may even be morally unjustified. But don't conflate the two terms: murder is a legal concept. Unfortunately, the terms "guilty" and "innocent" are also used in legal proceedings but their common meanings are not the same. In law, "guilty" and "not guilty" are logical negatives; in common speech, they are not.
Contrast this with soldiers, whose job it is to kill people. They are also not murderers, because their actions are legally justified by the state.
The really interesting question is when the state fails, what is the status of legal killers? If we use Nuremberg as an example, we can see that the answer depends on who's running the trial.
This makes me wonder why we're putting all kinds of information in the passport itself anyway. What kind of US border control personnel does not have direct access to that same passport data in a database? Putting the information in the passport itself makes it susceptible to tampering (e.g., putting a new picture in; modifying the ID to make it look like it's for someone else). It seems to me that all your passport should have is an encrypted ID which keys to a record in the database. Now, if you're truly worried about your border guards being too lazy to key in or scan (barcode, anyone?) the ID, fine, go ahead and use RFID tech, because now you're not exposing anything but an encrypted identifier.
The only other reason I can think of is information-sharing with foreign nations, who will not (and should not) have access to our passport database. But clearly, the authoritative information source should not be a paper booklet.
You're either being willfully ignorant or stupid. Let's do a simple calculation, shall we?
Assume, for a moment, that Comcast has 5 million subscribers. For simplicity's sake, let's say all of those five million people have "guaranteed" 5Mbit data plans. Also, for simplicity, let's say that Comcast has a single backbone trunk so that we essentially have a star network.
(5*10^6 bits/sec) * (5*10^6 subscribers) = 2.5*10^13 bits/sec required bandwidth at the trunk.
That's 25 terabit/sec. Under your plan, Comcast would need to lay enough fibre (and switches-- don't forget about those!) to handle 25 terabit/sec connections. At the moment, the state of the art in optical communication is OC-768, which can do roughly 40 Gbit/sec (IIRC, the theoretical ceiling on fibre bandwidth is 50 terabit/sec, but we're nowhere near being able to achieve that at the moment). In order to provide for 25 terabits/sec, you're going to need 625 of those OC-768 concatenated into a single trunk.
Now, OC-768 is basically so expensive that if you have to ask what it costs, you're not a customer, but we can try to extrapolate the cost. A leased OC-3 will typically run you $12K USD/month. Let's say that the actual cost is half that, so $6K USD/month since Comcast clearly will own their own lines. Let's naively say, too, that since OC-768 is 256 times faster than OC-3, that it also costs 256 times more (it's probably more, since it's state of the art). And we need 625 of them:
$6000 * 256 * 625 * 12 months = $11.5 trillion dollars annually. For comparison, the GDP of the United States is roughly $13 trillion. And this is only for the cable-- you also need equipment, cable-layers, engineers, and lawyers to negotiate right-of-way.
Now, somewhere along the way, some engineer's going to say, "hey, most of the time, we only need 1/100000th of that bandwidth to satisfy 99.9% of our users." What do you think they're going to do?
For the record, I am not at all opposed to Net Neutrality. In fact, Ed Markey was my Congressman until very recently, and I've written him a number of times to give him my opinion on the matter.
You're right about discriminating on the basis of source, too. That is the key part of the Net Neutrality debate. Unfortunately, the uninformed information-wants-to-be-free file-sharers jump in there and confuse the issue a bit.
Unfortunately, this part:
Honestly, just implementing a multi-level feedback queue like they've had in OS schedulers for decades
is spot-on, but also not possible with our current Internet architecture. TCP has a feedback mechanism, but it's in the wrong spot in the network stack. The problem isn't at the two ends of the connection, it's in between. By the time the TCP state machine has been notified of a problem, it's way too late to do anything subtle-- instead we backoff exponentially. But more than that-- TCP doesn't know anything about channel capacity. It would be nice if the network could be a little smarter about informing the sender what it could use.
Absent good mechanisms for managing these things, the ISPs are always going to fall back on the "this is my network and I can run it any way I want" argument, so your two-step process isn't so simple...
Network management is a fact of life. Many automatic mechanisms (e.g., load-balancing of circuits) need to know the state (e.g., load) of a particular circuit in order to balance traffic across another one. This is 'reasonable'. Other measures are 'reasonable' too.
People may disagree with me, but I also think that it is reasonable to make sure that jitter-sensitive data (like VoIP) is treated differently than Bittorrent traffic, which is not at all sensitive to jitter. The IP suite of protocols are extremely limited when it comes to flow control-- they can only do congestion prevention or egress rate limiting. If you're at the point where congestion is a problem, everyone is going to suffer. If that means that someone's Bittorrent traffic needs to be capped, I'm OK with that.
So the only other solution to 'reasonable' traffic management is overprovisioning. I know that your average Slashdotter thinks that ISPs should not 'oversubscribe' their lines, but saying this reveals a fundamental misunderstanding of the way packet switched networks work. I run networks. They're always oversubscribed. That's what makes them better than the POTS network-- the realization that most of the time, you don't need all that bandwidth for everybody. This is why packet-switched networks are cheaper, and counter-intuitively, more reliable (there was a paper in the '70's that showed that pooling memory resources vs statically allocating resources made out-of-memory errors orders of magnitude less likely; sorry don't remember the cite of the top of my head, but, same idea). Overprovisioning to the fantasy-level of a Slashdotter is very expensive because you're not just talking about extra bandwidth in the endpoints-- you're talking about bandwidth at the core.
The Internet always has had, and probably always will have growing pains. Right now VoIP, video-on-demand, and Bittorrent are competing for scarce resources. Until then, operators need to manage traffic. I will leave how as a discussion point for every else.
Yeah, sysadmins get shit on. But hey, we all knew it came with the territory before we got into this job. Part of this is due to the fact that people seem to think that computer run on magic or something, so they have no clue what we do.
As an aside, the IT department at work has kept a running tally on how long it's been since we've been thanked for our work at the company picnic. I've been here for 6 years... nothing yet. Meanwhile, the lowliest assistant gets a mention (and even sometimes [some assistant's] husband or wife, "for moral support").
Ah, good insight. That makes a lot of sense. And by "graphics drivers", they probably don't mean "any ol' driver that will work" but "accelerated drivers".
I never really demand much from my graphics hardware (I'd say that 95% of the OpenBSD machines I set up don't run X... some don't even have graphics capability), so I would have never thought of this.
You also got modded insightful for your meta-whining. I suspect it's because lots of us have had your same experience. For me, the scales are still weighted toward "worth it", because even though one is often instantly labeled an asshole for having an opinion around here, every now and then someone says something really great. I know that it's popular among certain geeks to pooh pooh the Slashdot crowd, but I've really learned some life-changing things around here. I know, it sounds cheesy, but I'm not kidding.
Unfortunately, the habitual trolls are probably more likely to stick around (because they're ill or something) than the rest of us. Maybe some smart Slashdotter will tell me what that phenomenon is called.
Slashdot Mirror
For anyone looking for even more depth, Tanenbaum's chapter on Media Access Control also talks about ALOHA in great detail before moving on to Ethernet, which is based on the work done by the University of Hawaii's packet radio experiments.
Interestingly enough, like in Antarctica, we can also provide these pioneers with vast amounts of porn. That's pretty easy to do. OTOH, I'm not sure if that would make things better or worse in the long run.
I always took that phrase to be two mutually-exclusive categories; not one statement qualifiying the other. E.g., "News for nerds AND stuff that matters", not "News for nerds BUT JUST stuff that matters".
Look, you made me go and be a nerd again.
Not true. There's a one-to-one correspondence between the width of the band and the symbol rate. One of your symbols could be "The Declaration of Independence in its entirety". Whatever meaning you ascribe to the symbol is up to you. The relationship between the bitrate and the bandwidth is more complex, because it is the result of many functions. If people want to conflate "bandwidth" and "baud rate", fine (it's sort of like conflating "wavelength" and "frequency"), but "bitrate" and "bandwidth" are only related insofar as one depends on the characteristics of the other.
Technical conversations are the exception to your "language drift" rule, by necessity. The common meaning must be precise, otherwise important subtleties are lost. Anyone who has done at least a semester of college physics will remember having to relearn words like "speed" and "weight", because regular people do not use those words the same way that technicians do.
You make fun, but I have a friend who actually was a firefighter in Antarctica, at McMurdo. It's not an easy job, especially when you consider that all of your normal methods for putting a fire out won't work there. We humans have the darndest ability to make fire wherever we go. This particular fellow is now a firefighter in Iraq. Apparently Antarctica was too easy for him.
The guy who works in the office next to my brother figures out how to put fires out in zero-g. Fortunately (although not so fortunate for the three people who died) NASA had an early experience with fire that made them realize they would need to address the problem before anyone went into space.
You're expected to mix metaphors here. It's surfing the Information Superhighway. Which, incidentally, is a series of tubes.
We've already done a great job at butchering the term anyway. Most usage doesn't correspond with reality.
(if you can't complain about this kind of stuff on a website billing itself as "news for nerds", where can you complain?)
But it just goes to show that carriers feel no need to compete. Most of us have no ability to choose the products we want from them, and with Uncle Sam's help, they can keep us from seeing how lame they really are.
But suddenly I say: and some people want this same government in charge of our military and now I'll be modded troll into oblivion.
Fixed that for ya. Oh wait, still a dumb thing to say...?
The government is a very large and diverse group of people. Some of those people do legitimately deserve to be criticized, but many, many, many of them do not. They do their jobs daily and with excellence, often for little compensation.
To infer that the government would be bad at managing health care because of a single instance of idiotic training materials is an example of woefully poor logic...
The problem with cellphone audio is not that it isn't properly compressed. It is. The problem is that telephone connections are extremely sensitive to jitter (or more precisely, humans are very sensitive to jitter). In fact, cellphone use the very same technique that MP3 uses, which is a perceptual codec. It is true that they also attenuate the signal somewhat, but this is mainly because designing A-D hardware that works over a wider range is more expensive, and for the most part, pointless. Who cares if you have booming bass on your telephone? MP3 is actually worse when it comes to jitter, because MPEG is VERY dependent on the last frame having come through correctly, until you hit a keyframe.
In some ways, the distinction between VoIP and traditional business phone systems has been blurring anyway. For the last 20 years or so businesses have gotten their phone lines via T1/E1 trunks which then go into a computer switching system (PBX). In the past, those switching systems ran custom/proprietary software, but now, you're starting to see them run things like Asterisk. Nowadays, people will either push the POTS service down to the ISP and receive SIP trunks (i.e., VoIP) in return, or else add SIP trunks via their existing Internet connection and run the gateway service themselves (that's what we're doing). Business phone connections haven't really been "plain old copper" for a long time.
I don't think you're going to see people give up NATs easily. NAT is a bona fide security feature, not just a consequence of having a LAN. This is the same thing that makes detecting bad segmentation faults possible in an operating system, and from that perspective, a separate address space is very desirable.
Any kind of 'fundamental change' that happens on the Internet needs to accept that NATs part of good architecture. You really want your toaster on the same address space as your Cray?
Do your Macs have Dell service tags, too?
Maybe because the public has shunned it?
Maybe that's true in other places, but in MA, the last time I was in the market for a car, I found that the reason I couldn't get a diesel is because very few of them pass MA's SO2 emissions. That's for new cars, and so I found that there was a bit of a cottage industry for few-weeks-old diesels that had been bought out-of-state and transported in. I was specifically looking for a VW Golf TDI, and this was around 2005, so things may have changed since then. I didn't end up getting one because financing through a bank for such a transaction was... complicated.
If we're to believe the copyright lobby's claims that downloading music without paying is 'stealing', then I think we can also conclude that 'stealing' , in some cases, hurts no one.
Murder is defined as being an "unlawful killing". If a court rules that a person is not guilty of murder, then that person is not a "murderer", by definition. If the person's role in the killing is unambiguous, that person is a "killer", and their action may even be morally unjustified. But don't conflate the two terms: murder is a legal concept. Unfortunately, the terms "guilty" and "innocent" are also used in legal proceedings but their common meanings are not the same. In law, "guilty" and "not guilty" are logical negatives; in common speech, they are not.
Contrast this with soldiers, whose job it is to kill people. They are also not murderers, because their actions are legally justified by the state.
The really interesting question is when the state fails, what is the status of legal killers? If we use Nuremberg as an example, we can see that the answer depends on who's running the trial.
This makes me wonder why we're putting all kinds of information in the passport itself anyway. What kind of US border control personnel does not have direct access to that same passport data in a database? Putting the information in the passport itself makes it susceptible to tampering (e.g., putting a new picture in; modifying the ID to make it look like it's for someone else). It seems to me that all your passport should have is an encrypted ID which keys to a record in the database. Now, if you're truly worried about your border guards being too lazy to key in or scan (barcode, anyone?) the ID, fine, go ahead and use RFID tech, because now you're not exposing anything but an encrypted identifier.
The only other reason I can think of is information-sharing with foreign nations, who will not (and should not) have access to our passport database. But clearly, the authoritative information source should not be a paper booklet.
What does "incorporated against the states" mean?
Your calculator has an 'S' key?
You're either being willfully ignorant or stupid. Let's do a simple calculation, shall we?
Assume, for a moment, that Comcast has 5 million subscribers. For simplicity's sake, let's say all of those five million people have "guaranteed" 5Mbit data plans. Also, for simplicity, let's say that Comcast has a single backbone trunk so that we essentially have a star network.
(5*10^6 bits/sec) * (5*10^6 subscribers) = 2.5*10^13 bits/sec required bandwidth at the trunk.
That's 25 terabit/sec. Under your plan, Comcast would need to lay enough fibre (and switches-- don't forget about those!) to handle 25 terabit/sec connections. At the moment, the state of the art in optical communication is OC-768, which can do roughly 40 Gbit/sec (IIRC, the theoretical ceiling on fibre bandwidth is 50 terabit/sec, but we're nowhere near being able to achieve that at the moment). In order to provide for 25 terabits/sec, you're going to need 625 of those OC-768 concatenated into a single trunk.
Now, OC-768 is basically so expensive that if you have to ask what it costs, you're not a customer, but we can try to extrapolate the cost. A leased OC-3 will typically run you $12K USD/month. Let's say that the actual cost is half that, so $6K USD/month since Comcast clearly will own their own lines. Let's naively say, too, that since OC-768 is 256 times faster than OC-3, that it also costs 256 times more (it's probably more, since it's state of the art). And we need 625 of them:
$6000 * 256 * 625 * 12 months = $11.5 trillion dollars annually. For comparison, the GDP of the United States is roughly $13 trillion. And this is only for the cable-- you also need equipment, cable-layers, engineers, and lawyers to negotiate right-of-way.
Now, somewhere along the way, some engineer's going to say, "hey, most of the time, we only need 1/100000th of that bandwidth to satisfy 99.9% of our users." What do you think they're going to do?
You're right about discriminating on the basis of source, too. That is the key part of the Net Neutrality debate. Unfortunately, the uninformed information-wants-to-be-free file-sharers jump in there and confuse the issue a bit.
Unfortunately, this part:
Honestly, just implementing a multi-level feedback queue like they've had in OS schedulers for decades
is spot-on, but also not possible with our current Internet architecture. TCP has a feedback mechanism, but it's in the wrong spot in the network stack. The problem isn't at the two ends of the connection, it's in between. By the time the TCP state machine has been notified of a problem, it's way too late to do anything subtle-- instead we backoff exponentially. But more than that-- TCP doesn't know anything about channel capacity. It would be nice if the network could be a little smarter about informing the sender what it could use.
Absent good mechanisms for managing these things, the ISPs are always going to fall back on the "this is my network and I can run it any way I want" argument, so your two-step process isn't so simple...
Network management is a fact of life. Many automatic mechanisms (e.g., load-balancing of circuits) need to know the state (e.g., load) of a particular circuit in order to balance traffic across another one. This is 'reasonable'. Other measures are 'reasonable' too.
People may disagree with me, but I also think that it is reasonable to make sure that jitter-sensitive data (like VoIP) is treated differently than Bittorrent traffic, which is not at all sensitive to jitter. The IP suite of protocols are extremely limited when it comes to flow control-- they can only do congestion prevention or egress rate limiting. If you're at the point where congestion is a problem, everyone is going to suffer. If that means that someone's Bittorrent traffic needs to be capped, I'm OK with that.
So the only other solution to 'reasonable' traffic management is overprovisioning. I know that your average Slashdotter thinks that ISPs should not 'oversubscribe' their lines, but saying this reveals a fundamental misunderstanding of the way packet switched networks work. I run networks. They're always oversubscribed. That's what makes them better than the POTS network-- the realization that most of the time, you don't need all that bandwidth for everybody. This is why packet-switched networks are cheaper, and counter-intuitively, more reliable (there was a paper in the '70's that showed that pooling memory resources vs statically allocating resources made out-of-memory errors orders of magnitude less likely; sorry don't remember the cite of the top of my head, but, same idea). Overprovisioning to the fantasy-level of a Slashdotter is very expensive because you're not just talking about extra bandwidth in the endpoints-- you're talking about bandwidth at the core.
The Internet always has had, and probably always will have growing pains. Right now VoIP, video-on-demand, and Bittorrent are competing for scarce resources. Until then, operators need to manage traffic. I will leave how as a discussion point for every else.
Yeah, sysadmins get shit on. But hey, we all knew it came with the territory before we got into this job. Part of this is due to the fact that people seem to think that computer run on magic or something, so they have no clue what we do.
As an aside, the IT department at work has kept a running tally on how long it's been since we've been thanked for our work at the company picnic. I've been here for 6 years... nothing yet. Meanwhile, the lowliest assistant gets a mention (and even sometimes [some assistant's] husband or wife, "for moral support").
Ah, good insight. That makes a lot of sense. And by "graphics drivers", they probably don't mean "any ol' driver that will work" but "accelerated drivers".
I never really demand much from my graphics hardware (I'd say that 95% of the OpenBSD machines I set up don't run X... some don't even have graphics capability), so I would have never thought of this.
You also got modded insightful for your meta-whining. I suspect it's because lots of us have had your same experience. For me, the scales are still weighted toward "worth it", because even though one is often instantly labeled an asshole for having an opinion around here, every now and then someone says something really great. I know that it's popular among certain geeks to pooh pooh the Slashdot crowd, but I've really learned some life-changing things around here. I know, it sounds cheesy, but I'm not kidding.
Unfortunately, the habitual trolls are probably more likely to stick around (because they're ill or something) than the rest of us. Maybe some smart Slashdotter will tell me what that phenomenon is called.