Who runs their own radio network with multiple towers instead of cell phones?
I cannot think of anyone. Even police and emergency services have switched to cell phones, albeit on a dedicated network. A network where 1.5Mbps per tower is woefully insufficient, of course.
You expect Cogent to source-route Netflix packets in their backbone? Good luck with that. Besides, where should Cogent send the traffic? They could send it through a paid transit to L3, of course...
The reason why most people don't like bandwidth metering
...is that variable costs suck. You cannot budget for them, and it means you always have to watch out for whether something inadvertently used up a lot of bandwidth without you noticing. I would happily pay a bit more for unmetered gas and electricity, but usage would likely go up a lot, and I am not willing to pay 10x as much.
I will only go metered for Internet if I can save at least $50 per month, which is exceedingly unlikely. I am basically paying tens of dollars per month in insurance against "excessive bandwidth", and the ISP is able to deliver that insurance practically for free. It would be rather stupid for ISPs to say no to that kind of free money.
Comcast was dropping traffic on the floor. Traffic that their customers had paid to receive. If Comcast is unhappy with the traffic patterns generated by their customers, they should not do business with those customers, or they should raise the price for them. Deliberately dropping paid-for traffic on the floor is exactly what a decent ISP never does.
Often last-mile lines are less than 0.1 percent utilized, measured as aggregate 95% peak. No ISP sets up their network to handle a thousand times more traffic than actually exists. That would be entirely uneconomical.
Proper providers make sure that lines get upgraded when there is a risk of congestion. If traffic patterns change significantly, such as with the advent of Netflix, backbone links must be upgraded. Luckily Netflix also made a number of people upgrade their last mile, so the 0.1 percent figure did not really budge all that much.
There IS an expectation that users do not use 100% of their bandwidth to get to the most expensive transit partner all the time. Hopefully some of that traffic stays local or goes via unpaid peering or at least through one of the dirt cheap transits like Cogent.
Just announce your BGP and automatically get peering. You don't even need to contact anyone about it.
You will only get access to smaller providers + some of he.net that way. Akamai or Google will not, AFAIK, talk to the IX route servers. I am not sure about Netflix; they offer a caching box for free anyway if you receive a reasonable amount of traffic from them.
Single mode fibre is for Internet what copper is for phone lines. It works even if it is 50 years old, and it provides enough bandwidth for the foreseeable future. Phone wire increased the useful bandwidth by 3 orders of magnitude through its history without physical changes to the lines. It is reasonable to believe that single mode fibre will do the same (particularly because we can deliver beyond 1Tbps per fibre today, just not economically).
Let the monopoly handle the physical wires. Require that all fibre goes uninterrupted back to somewhere with power, space and cooling enough for equipment from multiple providers -- no passive GPON splitters in inaccessible locations. Anyone with $1000 to spend on a Routerboard CCR can start an ISP.
Just isolate the pins involved and cut them physically. That's not so hard.
In a few HDMI generations, they will move to a unified packet-based format for everything so it all runs on the same wires. No, I have no insider information, but dedicated wires are a waste of bandwidth, and display bandwidth is precious.
CRT picture quality is crap. The colors are good and the black level too, but the geometry is completely off and the resolution is crap. For small displays you can get a half-decent resolution at perhaps 80Hz refresh, but as soon as you want something TV-size you have lost -- and even on the small displays the pixels are fuzzy and never in the right place.
If you wait longer to replace the tires, leaving them with a dangerously small amount of tread, you save a significant amount of tax... You would need very efficient enforcement.
Tax the criminals by the "service" of capture them.
Putting criminals into unserviceable debt so they get zero extra income from making money legally does not sound like a good strategy for fighting crime.
I should know, I was born in Denmark where prisoners until 1994 had to pay for the privilege of being in jail. They are still liable for lawyer fees, unfortunately, so the problem has only been solved partially.
There is absolutely nothing simplified about the Itanium. When it comes to branch prediction, it has both a traditional hardware branch predictor, but also predicate bits, branch hinting and 8 branch registers.
Yes, out-of-order execution was cut, but that is not really an ISA feature but more of an implementation detail. The Itanium ISA is almost impossible to execute out-of-order due to its enormous internal state among other misfeatures. The difficulty of going out-of-order was just another nail in the coffin for the slow and expensive Itanium.
Anyway, a minimal Itanium core is enormous compared to a "small" out-of-order core like the Pentium Pro, so simple is not an appropriate word for it.
Atom was just the Pentium MMX reheated. For integer workloads, it runs at pretty much original Pentium MMX instructions-per-clock, but at much higher frequencies of course. This year Intel finally made the Atom out-of-order, probably due to increasing pressure from out-of-order implementations of ARM.
No. Itanium was completely insane. Itanium took everything ever invented in computer architecture and tried to fit it onto one chip. At the same time it added every feature from x86 and PA-RISC, apart from the actual ISA of course, in order to simplify porting operating systems and other software. Making e.g. 10W Itanium chips is out of the question unless you software-emulate the whole thing at 10MHz on a sane architecture.
Anyway, Itanium succeeded in killing off the Unix market for MIPS and PA-RISC and almost killed off SPARC (but Sun veered at the last moment). It was a great success, politically. I do not believe Intel management was smart enough to see that ahead of time though; if they did it was one of the most shrewd business moves in history.
The x86 ISA has a much worse reputation than it deserves, and I say that as a RISC lover. The encoding takes up less space than almost all RISC ISAs, which is helpful if you are power constrained. Saving state is really quick on x86, so context switches and interrupts are fast. In contrast, Itanium encoding is horrendously verbose due to all the NOP instructions, and saving state on Itanium is extremely slow.
You are pretty much limited to EoMPLS-style service with that kind of configuration. As soon as you go VPLS or layer 3 you need rules either per MAC address or per subnet, and that will eat up your flow table.
Most places are OK with 40Gbps at the core and link aggregation, and that is available in dirt cheap Openflow switches. 100Gbps is not quite there yet at the low end, but it cannot be far away now. Alas, long range 40Gbps is still unavailable, but that will probably be fixed within a year.
What makes SDN worth it with a few flows though? Someone must obviously want it, I am just failing to see the use case, apart from easy VM migration (and some cheap switches already have great scripting support).
It will require line-of-sight or at least line-of-sight-including-reflections, but outdoors it should be just fine. Getting it inside buildings will likely require at least passive repeaters.
Mobile data speeds are only available if you are the only active subscriber on that cell and you are standing right next to it. In practice you will see a small fraction of the speed, unless you are one of the lucky field testers using it before normal subscribers are allowed in.
10Gbps per cell should enable a stable 100Mbps connection in practice even at peak hour, and that would be very useful.
People dig up lignite to burn it. Lots of it. A proper carbon fiber landfill would probably be a fairly decent coal mine, at least compared to the quite lousy ones left today.
Slashdot Mirror
Who runs their own radio network with multiple towers instead of cell phones?
I cannot think of anyone. Even police and emergency services have switched to cell phones, albeit on a dedicated network. A network where 1.5Mbps per tower is woefully insufficient, of course.
Modern cell towers use SyncE.
You expect Cogent to source-route Netflix packets in their backbone? Good luck with that. Besides, where should Cogent send the traffic? They could send it through a paid transit to L3, of course...
The reason why most people don't like bandwidth metering
...is that variable costs suck. You cannot budget for them, and it means you always have to watch out for whether something inadvertently used up a lot of bandwidth without you noticing. I would happily pay a bit more for unmetered gas and electricity, but usage would likely go up a lot, and I am not willing to pay 10x as much.
I will only go metered for Internet if I can save at least $50 per month, which is exceedingly unlikely. I am basically paying tens of dollars per month in insurance against "excessive bandwidth", and the ISP is able to deliver that insurance practically for free. It would be rather stupid for ISPs to say no to that kind of free money.
Comcast was dropping traffic on the floor. Traffic that their customers had paid to receive. If Comcast is unhappy with the traffic patterns generated by their customers, they should not do business with those customers, or they should raise the price for them. Deliberately dropping paid-for traffic on the floor is exactly what a decent ISP never does.
Often last-mile lines are less than 0.1 percent utilized, measured as aggregate 95% peak. No ISP sets up their network to handle a thousand times more traffic than actually exists. That would be entirely uneconomical.
Proper providers make sure that lines get upgraded when there is a risk of congestion. If traffic patterns change significantly, such as with the advent of Netflix, backbone links must be upgraded. Luckily Netflix also made a number of people upgrade their last mile, so the 0.1 percent figure did not really budge all that much.
There IS an expectation that users do not use 100% of their bandwidth to get to the most expensive transit partner all the time. Hopefully some of that traffic stays local or goes via unpaid peering or at least through one of the dirt cheap transits like Cogent.
Just announce your BGP and automatically get peering. You don't even need to contact anyone about it.
You will only get access to smaller providers + some of he.net that way. Akamai or Google will not, AFAIK, talk to the IX route servers. I am not sure about Netflix; they offer a caching box for free anyway if you receive a reasonable amount of traffic from them.
Single mode fibre is for Internet what copper is for phone lines. It works even if it is 50 years old, and it provides enough bandwidth for the foreseeable future. Phone wire increased the useful bandwidth by 3 orders of magnitude through its history without physical changes to the lines. It is reasonable to believe that single mode fibre will do the same (particularly because we can deliver beyond 1Tbps per fibre today, just not economically).
Let the monopoly handle the physical wires. Require that all fibre goes uninterrupted back to somewhere with power, space and cooling enough for equipment from multiple providers -- no passive GPON splitters in inaccessible locations. Anyone with $1000 to spend on a Routerboard CCR can start an ISP.
The price of a T1 has not changed because it is entirely obsolete. No one sane would want one, and specialty items are expensive.
In places with competitive markets, you can get the SLA you want with the technology you want.
Just isolate the pins involved and cut them physically. That's not so hard.
In a few HDMI generations, they will move to a unified packet-based format for everything so it all runs on the same wires. No, I have no insider information, but dedicated wires are a waste of bandwidth, and display bandwidth is precious.
CRT picture quality is crap. The colors are good and the black level too, but the geometry is completely off and the resolution is crap. For small displays you can get a half-decent resolution at perhaps 80Hz refresh, but as soon as you want something TV-size you have lost -- and even on the small displays the pixels are fuzzy and never in the right place.
There is more to a good picture than color.
If you wait longer to replace the tires, leaving them with a dangerously small amount of tread, you save a significant amount of tax... You would need very efficient enforcement.
Also, tire smuggling would be a problem.
Tax the criminals by the "service" of capture them.
Putting criminals into unserviceable debt so they get zero extra income from making money legally does not sound like a good strategy for fighting crime.
I should know, I was born in Denmark where prisoners until 1994 had to pay for the privilege of being in jail. They are still liable for lawyer fees, unfortunately, so the problem has only been solved partially.
This could be because the fines in the UK are indistinguishable from 0 for most people. And yes I have had a ticket, unfortunately.
The points on the license and the insurance trouble are an effective deterrent, but those do not generate revenue for the government or the police.
An enemy that may look like normal people, but want to kill or convert all others ...
You are behind, that is definitely about the Commies, the Cold War stuff is back. Islamophobia was last decade.
There is absolutely nothing simplified about the Itanium. When it comes to branch prediction, it has both a traditional hardware branch predictor, but also predicate bits, branch hinting and 8 branch registers.
Yes, out-of-order execution was cut, but that is not really an ISA feature but more of an implementation detail. The Itanium ISA is almost impossible to execute out-of-order due to its enormous internal state among other misfeatures. The difficulty of going out-of-order was just another nail in the coffin for the slow and expensive Itanium.
Anyway, a minimal Itanium core is enormous compared to a "small" out-of-order core like the Pentium Pro, so simple is not an appropriate word for it.
Atom was just the Pentium MMX reheated. For integer workloads, it runs at pretty much original Pentium MMX instructions-per-clock, but at much higher frequencies of course. This year Intel finally made the Atom out-of-order, probably due to increasing pressure from out-of-order implementations of ARM.
No. Itanium was completely insane. Itanium took everything ever invented in computer architecture and tried to fit it onto one chip. At the same time it added every feature from x86 and PA-RISC, apart from the actual ISA of course, in order to simplify porting operating systems and other software. Making e.g. 10W Itanium chips is out of the question unless you software-emulate the whole thing at 10MHz on a sane architecture.
Anyway, Itanium succeeded in killing off the Unix market for MIPS and PA-RISC and almost killed off SPARC (but Sun veered at the last moment). It was a great success, politically. I do not believe Intel management was smart enough to see that ahead of time though; if they did it was one of the most shrewd business moves in history.
The x86 ISA has a much worse reputation than it deserves, and I say that as a RISC lover. The encoding takes up less space than almost all RISC ISAs, which is helpful if you are power constrained. Saving state is really quick on x86, so context switches and interrupts are fast. In contrast, Itanium encoding is horrendously verbose due to all the NOP instructions, and saving state on Itanium is extremely slow.
Even in Chechnya, where bad guys control pretty much all parts of the voting process, it is obvious to an intelligent person that there is fraud.
With electronic voting, the fraud will be much harder to spot.
Some of the same problems. In many cases you can cancel your mail vote by going to a voting booth.
If mail voting was popular, it would need to be made more secure.
I believe Linux Torvalds provided the OS, lets not get that mixed up. Stallman provided the tools.
An OS is more than just the kernel. Run Debian GNU/kFreeBSD and you will hardly notice that anything changed from normal Debian.
You are pretty much limited to EoMPLS-style service with that kind of configuration. As soon as you go VPLS or layer 3 you need rules either per MAC address or per subnet, and that will eat up your flow table.
Most places are OK with 40Gbps at the core and link aggregation, and that is available in dirt cheap Openflow switches. 100Gbps is not quite there yet at the low end, but it cannot be far away now. Alas, long range 40Gbps is still unavailable, but that will probably be fixed within a year.
What makes SDN worth it with a few flows though? Someone must obviously want it, I am just failing to see the use case, apart from easy VM migration (and some cheap switches already have great scripting support).
It will require line-of-sight or at least line-of-sight-including-reflections, but outdoors it should be just fine. Getting it inside buildings will likely require at least passive repeaters.
4G has much lower latency than 3G. That helps a lot when dealing with the various delays, even if you never get anywhere near the maximum bandwidth.
Mobile data speeds are only available if you are the only active subscriber on that cell and you are standing right next to it. In practice you will see a small fraction of the speed, unless you are one of the lucky field testers using it before normal subscribers are allowed in.
10Gbps per cell should enable a stable 100Mbps connection in practice even at peak hour, and that would be very useful.
People dig up lignite to burn it. Lots of it. A proper carbon fiber landfill would probably be a fairly decent coal mine, at least compared to the quite lousy ones left today.