The point isn't to have enough IPs for every user, but to have enough IPs such that the chance of collision is low. The other thing you miss is the ability to merge large corp networks. If you have to merge two companies with two datacenters with 100,000 machines each, the chance of a colliding IP address with only 4 bytes is quite large.
Since my work laptop isn't allowed to join my "home" workgroup, there is no DNS which will work between by laptop and my machine.
At least assign a static IPv6 address to your laptop then add it to your HOSTS file. The biggest issue people have moving from IPv4 to v6 is they're not used to have many IP addresses per machine. This will be the standard for IPv6. Create static addresses.
The only way to do file/printer sharing is by IP address.
On my Win7 network, my $60 HP printer can be addressed via name because of P2P name resolution protocols. Should work if you're in the same broadcast domain and same subnet assuming your systems and devices support the protocols. Even the PS3 resolves and my router sees all the names also. Seems to be a very standard protocol.
For internal to my own network, I'm not sure what IPV6 offers *me*
Probably nothing. Most small internal networks won't benefit. I can think of a lot of benefit it will give me on the internet. tons of IP addresses to allocate to each FreeBSD jail, no NAT issues, and multicast will be f'n awesome once apps start to use it. I see P2P VoIP being very easy with asymmetrical internet connections.
You can't "extend" IPv4 without breaking it. 48 bits is not a natural amount when it comes to aligned data. Minor issue, but most CPUs work on 32/64 bits at a time, so at least 64. Also, if you knew anything about the issues of routing, the length of the IP address isn't about having more devices as much as being able to hand out HUGE blocks of IP address, not care about the waste, while benefiting from not having fragmented routing.
Fragmented routing tables is a BIG issue. Route table fragmentation is inversely related to the amount of wasted IP addresses.
Another large benefit of HUGE address ranges is you don't need to have thousands of devices requesting IP addresses from a central broker and causing massive contention. Instead you assume your chance of obtaining an already in-use IP address is almost none. This gives you optimistic IP address allocation, which scales MUCH MUCH better.
The same goes for internet traffic. My individual usage doesn't make a huge difference but my ISP has to keep the equivalent of spinning reserve in upstream capacity to cover all the users. It isn't cheap to buy this bandwidth.
I forgot to add:
Level3 and Hurricane Electric charge $1/mbit/month for dedicated Teir1 back-bone connection... In 10Gb increments(these bulk rates only apply to the big boys).
I pay $75 for a 30Mb/s over-subscribed data-capped connection. I wouldn't mind paying $30 for 30Mb of dedicated bandwidth. Obviously there are other reoccurring costs like engineers to monitor/repair and leasing dark-fiber to connect to L3/HE, but you get the point. Once you purchase/install the fiber, you can just buy faster optics and pump 10s to 100s of Gb/s over a single fiber. I think the new record was almost 1Tb/s over 1200 miles of a single strand of standard bulk fiber.
The cost of that leased fiber is a fixed cost. You can easily increase the Mb/s over that same fiber while only having to pay $1/mbit. It boggles me that a large ISP can't offer cheap/faster internet when these small ISPs are actually doing it, but they're having trouble expanding because of the monopolistic hold these large ISPs have.
My ISP is rolling out 200/200 FTTP with an average cost of $900/living unit for the infrastructure. Not all units will use the service, so the effective cost per unit is higher, but stuff.
Anyway, at least one company I know is rolling out a nice chassis next year that can handle 20,000 end-points with 1Gb per end-point with 32 customers per 10Gb fiber(32 customers sharing 10Gb of bandwidth, each with a 1Gb connection). The chassis back-plane can support 2Tb/s and has 800Gb/s of uplink. We're talking about a device 1/2 the size of a refrigerator offering 800Gb of bandwidth to 20,000 houses with an average of ~320Mb/s of dedicated bandwidth to the chassis and an overall average of 40Mb of dedicated bandwidth per house via the uplink. The chance of all of that bandwidth being used a once? Slim to none. The next bottleneck is the trunk.
These are passive optical networks. Strait from the ISP to the customers, no switches between. Repeaters and powered equipment can be used, but these systems have 40-60KM ranges, so that covers a lot of land. Nothing but un-powered fiber between the ISP and the customer.
I think you under-estimate the density modern telecom equipment can handle.
Around here, the government WANTS faster internet, but the Cable company keeps complaining when competition moves in or the state tries to lay its own. The people want it, the people are willing to pay a fair price for it, the Cable companies are not willing to do it.
Actually, this was more like the past decade. The people finally won and we're getting fiber state wide. State is laying the infrastructure and leasing and/or selling at whole sale, while also providing grants/loans for local ISPs to upgrade/expand their infrastructure.
Fiber Testbed in my City: 50/50 uncapped $60/month - get speed at all times and low latency(rollout going live soon, expecting prices to drop)
California fiber 1Gb/1Gb $70/month no cap - get speed at all times and low latency
If their freemarket, the lower the cost, they more they must charge. The price of bandwidth has been plummeting by 50% year-over-year for the past decade. This means it has devalued by 99.9%(1-0.5^10) in 10 years. How else are they supposed to make money with those cheap costs?
There is a reason you pay a connection fee on top of a usage fee for power. Maybe ISPs should start doing this. $15/month connection fee and $2/mbit-avg during peak hours and $0/mbit-avg during off-peak.
I'm down with this. Even with "heavy" usage, my bill would be much cheaper. I bet my bill would be under $20/month even using BitTorrent.
This is the best argument yet as it shows the opposite of what they want to happen. I love how you point out that fixed rates that ignore actual costs and peak usage patterns actually increases inefficient use by actively discouraging off-peak usage.
Same thing could be said about using a GUI. Why waste cheap resources when everything could be small and compact when written in ASM? Why do people graphics anyway? Ohh wait, because technology can provide it at an insanely cheap price. Ever notice the only people complaining about bandwidth is major ISPs who have monopolies, wireless services with actual limited resources, or New Zealand and it's small customer base with an expensive oceanaic-backhaul.
The smart people are saying the bandwidth is damn near free. Level3 being one and some other company that runs trans-Atlantic lines both claim that trunk-line growth is out-pacing demand by quite a bit. So much so that bandwidth prices are dropping faster than supply. Whole-sale costs are reduced by 50% while available bandwidth is increased by only 50% every year.
Level3 ran a story a few years back talking about how a HUGE portion of their back-bone is dark-fiber because technology is out-pacing demand. Hurricane electric offers 1Mb/month for $1 of DEDICATED bandwidth. Peering is even cheaper if not free in many cases. An ISP should be able to make a decent profit on $2/mbit with no data cap, and probably a monthly connection fee for the infrastructure.
Nearly all of the current infrastructure for the last-mile has been paid for by tax-payers. Not customers, tax-payers. Hundreds of billions in grants have been given to the major ISPs over the past 15 years, with more than enough to cover the value of their current infrastructure. The only costs to these large ISPs is maintenance and operational costs. They should have little to no debt from infrastructure investments because of government hand-outs.
Very recently the government asked the Major cable companies how much to upgrade 80% of their customers to 100Mb internet. they responded. $8-$10/customer. fk'n A. $10 for the infrastructure?! Remember, infrastructure is the biggest cost by leaps and bounds.
Another consultant for large ISPs said that major cable's average costs for bandwidth+maintenance+upgrades comes out to ~$6/month averages over the typical upgrade cycle. Everything above that is profit.
While the idea of QoS is nice because of local choke points, Level3 had a nice blog about how it's actually cheaper and more effective to just upgrade bandwidth for trunk lines. QoS has a lot of processing over-head and it gets complex with corner cases. Just brute-forcing the issue is cheaper and makes everyone happier.
QoS is more of an issue of the end-points than that of main trunks. The easiest way to fairly manage end-points is via TDMA. As a group of end-points over-saturate a node/link/etc, the controller unit just evenly distributes communication time.
My example was an extreme case to prove a point; but like you implied, it would be hard for a person to get around not using their internet at all during peak hours without being inconvenienced.
I guess my main point wasn't that the average person could practically time their usage to reduce the burst, but that in all of this discussion of being "fair" by charging per data unit, it is completely unfair to ignore how costs in the system actually work. If I want to seed a FreeBSD/Linux ISO during off-peak, why should I be charged near infinite above the cost? The price should in some way reflect the cost.
So, if I'm a lucky person that gets an uncapped internet connection, then I pick a random Comcast address and start sending a 1Mbit UDP stream at them, by the end of the month, they will have gone over their data cap and be charged?
With Comcast's new data plan, people will get a 300GB cap with $10/50GB after. If I send 2Mb/s throughout the month, I will use ~618GB. This means they will be 318GB over their cap. That would be $70(rounded up to 350GB) added to their bill, plus whatever they use. A 2Mb wouldn't even be enough for them to notice and they would have no idea. Technically you're not flooding them, so it's not a DoS and it doesn't disrupt any services.
The different is that SSH typically doesn't compete with you for bandwidth, but for latency. High bandwidth streams ruin the experience for everyone not because of bandwidth but because of buffer bloat. The ideal situation is a "fair" TDMA style distribution of network traffic using relatively small quantum. The problem is trying to implement that at a core router is damn near impossible.
GPON, for example, uses TDMA to distribute and share bandwidth on a given "node"(technically not a node). This allows a saturated node to degrade in a graceful and fair way. Once you start moving upstream to other choke points, routers don't care about streams/end-points and just switch based on packets. This means a single end-point could potentially take an unfair amount bandwidth by not backing off. Lots of TCP streams from a single end-point tend to have this problem.
Obviously this is mostly a congestion control issue, but congestion control is actually a cooperative thing, not enforced.
Energy isn't free, they have to fuel those power plants. Heavy internet users don't really increase electrical costs in a meaningful way outside of the base load to power the equipment but that doesn't care about network traffic.
Except off-peak data transfers are "free". Nearly of the cost of being an ISP is infrastructure. ISPs have to size their back-haul connections to peak usage which is almost entirely determined by the average user.
Someone transfer 20Mb/s 12 hours per day but during off-peak is going to cost less than someone transferring 2Mb/s for 1 hour during peak.
The cost to the ISP isn't determined by volume, but by burst. Volume can influence the burst, but it is only loosely coupled. I could easily time and packet shape my traffic to the point where it's cheaper for the ISP to handle 10TB of data than it is 10GB of data. In other words, I could manage a heavy BitTorrent seed in such a way that it costs less than your mom updating her FB status and watching a few funny cat videos on Youtube.
Sounds like....hehehe... pun... Sounds like your *nix setups have a more consistent latency. Thanks for the response. It's hard to find info on the topic without delving into pro discussions that go over my head or make certain assumptions about their audience.
"If you give an amphetamine to a kid who does not have a serious issue they are not going to be sitting still and studying.
My doctor said the same thing. He said the amount of amphetamine I was taking would normally make a person very hyper, but it settled me down. He even recommended small dosages of caffeine, which also settled me. Not so much large dosages.
There's a weird background processing thing that goes on when you stop thinking about stuff.
I love taking advantage of this by looking at a problem, identifying the issue, then I go to browse the web, like/., techdirt, ars, etc. Then I come back after 5-10 minutes and attempt to create a solution. I find this process tends to help me best when solving new problems. Really tough problems(starting a new project) can take a few days of this back-and-forth, which also includes researching the subject, identifying possible issues, and going home to "sleep on it".
I find when I talk to other people, I try to mirror what they're thinking about what I'm saying. By having someone to listen, just me talking at them allows me to activate this part of my brain and help me criticize my own logic as I attempt to explain it. Thinking or talking to myself does not activate this brain pathway for me.
One of my favorite things to do is analyze what I think and why I thought it.
Another way you could use a hint is that I don't care if 1% of these pixels are completely wrong.
More like 100% of your pixels are wrong, but within 1% of their intended value. Get 60fps of those pixels and suddenly it all averages out and it looks the same.
The biggest issue would be geometric calculation where sharp lines could jitter. Pixel shaders would be less of an issue because minor tone differences won't be noticed.
Re:Most programs don't need a 64-bit address space
on
Linux 3.4 Released
·
· Score: 1
16GB is pretty cheap, beyond that it starts to get expensive
Depends on what you mean by "expensive". I remember 16MB of ram going for $800. Now I can get 32GB(4x8GB) ECC 1333(1600 is about $100 more) 1.35v namebrand for $290. DDR4 will drop those prices down a lot with newer die-shrinks.
Slashdot Mirror
The point isn't to have enough IPs for every user, but to have enough IPs such that the chance of collision is low. The other thing you miss is the ability to merge large corp networks. If you have to merge two companies with two datacenters with 100,000 machines each, the chance of a colliding IP address with only 4 bytes is quite large.
Since my work laptop isn't allowed to join my "home" workgroup, there is no DNS which will work between by laptop and my machine.
At least assign a static IPv6 address to your laptop then add it to your HOSTS file. The biggest issue people have moving from IPv4 to v6 is they're not used to have many IP addresses per machine. This will be the standard for IPv6. Create static addresses.
The only way to do file/printer sharing is by IP address.
On my Win7 network, my $60 HP printer can be addressed via name because of P2P name resolution protocols. Should work if you're in the same broadcast domain and same subnet assuming your systems and devices support the protocols. Even the PS3 resolves and my router sees all the names also. Seems to be a very standard protocol.
For internal to my own network, I'm not sure what IPV6 offers *me*
Probably nothing. Most small internal networks won't benefit. I can think of a lot of benefit it will give me on the internet. tons of IP addresses to allocate to each FreeBSD jail, no NAT issues, and multicast will be f'n awesome once apps start to use it. I see P2P VoIP being very easy with asymmetrical internet connections.
You can't "extend" IPv4 without breaking it. 48 bits is not a natural amount when it comes to aligned data. Minor issue, but most CPUs work on 32/64 bits at a time, so at least 64. Also, if you knew anything about the issues of routing, the length of the IP address isn't about having more devices as much as being able to hand out HUGE blocks of IP address, not care about the waste, while benefiting from not having fragmented routing.
Fragmented routing tables is a BIG issue. Route table fragmentation is inversely related to the amount of wasted IP addresses.
Another large benefit of HUGE address ranges is you don't need to have thousands of devices requesting IP addresses from a central broker and causing massive contention. Instead you assume your chance of obtaining an already in-use IP address is almost none. This gives you optimistic IP address allocation, which scales MUCH MUCH better.
The same goes for internet traffic. My individual usage doesn't make a huge difference but my ISP has to keep the equivalent of spinning reserve in upstream capacity to cover all the users. It isn't cheap to buy this bandwidth.
I forgot to add:
Level3 and Hurricane Electric charge $1/mbit/month for dedicated Teir1 back-bone connection... In 10Gb increments(these bulk rates only apply to the big boys).
I pay $75 for a 30Mb/s over-subscribed data-capped connection. I wouldn't mind paying $30 for 30Mb of dedicated bandwidth. Obviously there are other reoccurring costs like engineers to monitor/repair and leasing dark-fiber to connect to L3/HE, but you get the point. Once you purchase/install the fiber, you can just buy faster optics and pump 10s to 100s of Gb/s over a single fiber. I think the new record was almost 1Tb/s over 1200 miles of a single strand of standard bulk fiber.
The cost of that leased fiber is a fixed cost. You can easily increase the Mb/s over that same fiber while only having to pay $1/mbit. It boggles me that a large ISP can't offer cheap/faster internet when these small ISPs are actually doing it, but they're having trouble expanding because of the monopolistic hold these large ISPs have.
My ISP is rolling out 200/200 FTTP with an average cost of $900/living unit for the infrastructure. Not all units will use the service, so the effective cost per unit is higher, but stuff.
Anyway, at least one company I know is rolling out a nice chassis next year that can handle 20,000 end-points with 1Gb per end-point with 32 customers per 10Gb fiber(32 customers sharing 10Gb of bandwidth, each with a 1Gb connection). The chassis back-plane can support 2Tb/s and has 800Gb/s of uplink. We're talking about a device 1/2 the size of a refrigerator offering 800Gb of bandwidth to 20,000 houses with an average of ~320Mb/s of dedicated bandwidth to the chassis and an overall average of 40Mb of dedicated bandwidth per house via the uplink. The chance of all of that bandwidth being used a once? Slim to none. The next bottleneck is the trunk.
These are passive optical networks. Strait from the ISP to the customers, no switches between. Repeaters and powered equipment can be used, but these systems have 40-60KM ranges, so that covers a lot of land. Nothing but un-powered fiber between the ISP and the customer.
I think you under-estimate the density modern telecom equipment can handle.
Around here, the government WANTS faster internet, but the Cable company keeps complaining when competition moves in or the state tries to lay its own. The people want it, the people are willing to pay a fair price for it, the Cable companies are not willing to do it.
Actually, this was more like the past decade. The people finally won and we're getting fiber state wide. State is laying the infrastructure and leasing and/or selling at whole sale, while also providing grants/loans for local ISPs to upgrade/expand their infrastructure.
fk you cable!
Fiber Testbed in my City: 50/50 uncapped $60/month - get speed at all times and low latency(rollout going live soon, expecting prices to drop)
California fiber 1Gb/1Gb $70/month no cap - get speed at all times and low latency
If their freemarket, the lower the cost, they more they must charge. The price of bandwidth has been plummeting by 50% year-over-year for the past decade. This means it has devalued by 99.9%(1-0.5^10) in 10 years. How else are they supposed to make money with those cheap costs?
There is a reason you pay a connection fee on top of a usage fee for power. Maybe ISPs should start doing this. $15/month connection fee and $2/mbit-avg during peak hours and $0/mbit-avg during off-peak.
I'm down with this. Even with "heavy" usage, my bill would be much cheaper. I bet my bill would be under $20/month even using BitTorrent.
This is the best argument yet as it shows the opposite of what they want to happen. I love how you point out that fixed rates that ignore actual costs and peak usage patterns actually increases inefficient use by actively discouraging off-peak usage.
Same thing could be said about using a GUI. Why waste cheap resources when everything could be small and compact when written in ASM? Why do people graphics anyway? Ohh wait, because technology can provide it at an insanely cheap price. Ever notice the only people complaining about bandwidth is major ISPs who have monopolies, wireless services with actual limited resources, or New Zealand and it's small customer base with an expensive oceanaic-backhaul.
One of those is not like the others.
The smart people are saying the bandwidth is damn near free. Level3 being one and some other company that runs trans-Atlantic lines both claim that trunk-line growth is out-pacing demand by quite a bit. So much so that bandwidth prices are dropping faster than supply. Whole-sale costs are reduced by 50% while available bandwidth is increased by only 50% every year.
Level3 ran a story a few years back talking about how a HUGE portion of their back-bone is dark-fiber because technology is out-pacing demand. Hurricane electric offers 1Mb/month for $1 of DEDICATED bandwidth. Peering is even cheaper if not free in many cases. An ISP should be able to make a decent profit on $2/mbit with no data cap, and probably a monthly connection fee for the infrastructure.
Nearly all of the current infrastructure for the last-mile has been paid for by tax-payers. Not customers, tax-payers. Hundreds of billions in grants have been given to the major ISPs over the past 15 years, with more than enough to cover the value of their current infrastructure. The only costs to these large ISPs is maintenance and operational costs. They should have little to no debt from infrastructure investments because of government hand-outs.
Very recently the government asked the Major cable companies how much to upgrade 80% of their customers to 100Mb internet. they responded. $8-$10/customer. fk'n A. $10 for the infrastructure?! Remember, infrastructure is the biggest cost by leaps and bounds.
Another consultant for large ISPs said that major cable's average costs for bandwidth+maintenance+upgrades comes out to ~$6/month averages over the typical upgrade cycle. Everything above that is profit.
While the idea of QoS is nice because of local choke points, Level3 had a nice blog about how it's actually cheaper and more effective to just upgrade bandwidth for trunk lines. QoS has a lot of processing over-head and it gets complex with corner cases. Just brute-forcing the issue is cheaper and makes everyone happier.
QoS is more of an issue of the end-points than that of main trunks. The easiest way to fairly manage end-points is via TDMA. As a group of end-points over-saturate a node/link/etc, the controller unit just evenly distributes communication time.
My example was an extreme case to prove a point; but like you implied, it would be hard for a person to get around not using their internet at all during peak hours without being inconvenienced.
I guess my main point wasn't that the average person could practically time their usage to reduce the burst, but that in all of this discussion of being "fair" by charging per data unit, it is completely unfair to ignore how costs in the system actually work. If I want to seed a FreeBSD/Linux ISO during off-peak, why should I be charged near infinite above the cost? The price should in some way reflect the cost.
So, if I'm a lucky person that gets an uncapped internet connection, then I pick a random Comcast address and start sending a 1Mbit UDP stream at them, by the end of the month, they will have gone over their data cap and be charged?
With Comcast's new data plan, people will get a 300GB cap with $10/50GB after. If I send 2Mb/s throughout the month, I will use ~618GB. This means they will be 318GB over their cap. That would be $70(rounded up to 350GB) added to their bill, plus whatever they use. A 2Mb wouldn't even be enough for them to notice and they would have no idea. Technically you're not flooding them, so it's not a DoS and it doesn't disrupt any services.
The different is that SSH typically doesn't compete with you for bandwidth, but for latency. High bandwidth streams ruin the experience for everyone not because of bandwidth but because of buffer bloat. The ideal situation is a "fair" TDMA style distribution of network traffic using relatively small quantum. The problem is trying to implement that at a core router is damn near impossible.
GPON, for example, uses TDMA to distribute and share bandwidth on a given "node"(technically not a node). This allows a saturated node to degrade in a graceful and fair way. Once you start moving upstream to other choke points, routers don't care about streams/end-points and just switch based on packets. This means a single end-point could potentially take an unfair amount bandwidth by not backing off. Lots of TCP streams from a single end-point tend to have this problem.
Obviously this is mostly a congestion control issue, but congestion control is actually a cooperative thing, not enforced.
Energy isn't free, they have to fuel those power plants. Heavy internet users don't really increase electrical costs in a meaningful way outside of the base load to power the equipment but that doesn't care about network traffic.
Except off-peak data transfers are "free". Nearly of the cost of being an ISP is infrastructure. ISPs have to size their back-haul connections to peak usage which is almost entirely determined by the average user.
Someone transfer 20Mb/s 12 hours per day but during off-peak is going to cost less than someone transferring 2Mb/s for 1 hour during peak.
The cost to the ISP isn't determined by volume, but by burst. Volume can influence the burst, but it is only loosely coupled. I could easily time and packet shape my traffic to the point where it's cheaper for the ISP to handle 10TB of data than it is 10GB of data. In other words, I could manage a heavy BitTorrent seed in such a way that it costs less than your mom updating her FB status and watching a few funny cat videos on Youtube.
Sounds like....hehehe... pun... Sounds like your *nix setups have a more consistent latency. Thanks for the response. It's hard to find info on the topic without delving into pro discussions that go over my head or make certain assumptions about their audience.
"If you give an amphetamine to a kid who does not have a serious issue they are not going to be sitting still and studying.
My doctor said the same thing. He said the amount of amphetamine I was taking would normally make a person very hyper, but it settled me down. He even recommended small dosages of caffeine, which also settled me. Not so much large dosages.
There's a weird background processing thing that goes on when you stop thinking about stuff.
I love taking advantage of this by looking at a problem, identifying the issue, then I go to browse the web, like /., techdirt, ars, etc. Then I come back after 5-10 minutes and attempt to create a solution. I find this process tends to help me best when solving new problems. Really tough problems(starting a new project) can take a few days of this back-and-forth, which also includes researching the subject, identifying possible issues, and going home to "sleep on it".
I find when I talk to other people, I try to mirror what they're thinking about what I'm saying. By having someone to listen, just me talking at them allows me to activate this part of my brain and help me criticize my own logic as I attempt to explain it. Thinking or talking to myself does not activate this brain pathway for me.
One of my favorite things to do is analyze what I think and why I thought it.
Unit test in accurate mode, then release-compile for fast mode.
Another way you could use a hint is that I don't care if 1% of these pixels are completely wrong.
More like 100% of your pixels are wrong, but within 1% of their intended value. Get 60fps of those pixels and suddenly it all averages out and it looks the same.
The biggest issue would be geometric calculation where sharp lines could jitter. Pixel shaders would be less of an issue because minor tone differences won't be noticed.
16GB is pretty cheap, beyond that it starts to get expensive
Depends on what you mean by "expensive". I remember 16MB of ram going for $800. Now I can get 32GB(4x8GB) ECC 1333(1600 is about $100 more) 1.35v namebrand for $290. DDR4 will drop those prices down a lot with newer die-shrinks.