Routers have buffers measured in milliseconds. Everything else gets dropped. Even at a slow 10gb rate, 1 seconds of buffering is over 1GB of memory. Once you start getting into the terabit ranges, the bandwidth coming off those links is faster than the L1 cache on your 5ghz OC'd Intel CPU. You couldn't buffer that to memory if you wanted to.
The points of congestion are typically the hardest to QoS. Once you start getting into the 400gb+ link speeds, you can't do real time QoS anymore. Enabling QoS slows you down to ~150gb/s. I assume it'll be much worse at terabit speeds. When you're talking about moving one bit every 1/1000th of a nanosecond, you don't have time to QoS. Eventually we'll be doing photonic routing. It will be interesting trying to "buffer" photons.
We're already reaching the point of physics where the only way to route data is first-in-first-out. Anything else is nearly impossible or comes at a huge performance hit.
Most people use DNS instead of IP directly. I can ping by name every device on my network, and it worked like that out of the box. That includes my wireless printer and wife's android cell phone. My 7 year old Netgear router even shows the name of each device next to their MAC addresses when looking in the config. My PFSense box that I later setup automatically detected every device's name and added it to the Unbound DNS server.
HPD110.localdomain. 3600 IN A 192.168.1.9
C:\Windows\system32>ping HPD110
Pinging HPD110.localdomain [192.168.1.9] with 32 bytes of data:
Reply from 192.168.1.9: bytes=32 time=3ms TTL=255
This DNS entry automatically got created. But I can ping "HPD110" even without a DNS entry, because of whatever common protocol is used.
Here's what my wireless router shows for "HPD110".
# IP Address MAC Address Device Name
1 192.168.1.9 68:B5:99:3E:03:4B HP68B5993E034B
C:\Windows\system32>ping HP68B5993E034B
Pinging HP68B5993E034B [192.168.1.9] with 32 bytes of data:
Reply from 192.168.1.9: bytes=32 time=2ms TTL=255
I did not set up any of these names, they all automatically configured themselves.
In the USA, new rules are going into affect now that we on the last block. One of those is transferring a block has a fixed cost of $10/ip. Can get pricey for a large blocks. You soon must also show they you actually need all the IPs you have, otherwise they can forcefully take them back. And the proof of "needing" will get progressively harder.
You can't change "change" IPv4, it would require an entirely new protocol, even if everything else was treated the same. Large address spaces allow for wasteful use to reduce fragmentation. Fragmentation is the bane of core routers.
Just have more IPs that you could ever want in 100 years, and 100 years from now, you'll need a new protocol. That's IPv6.
And you electricity, for free! They also increase the amount of interference going on and they are getting a service for free that was not agreed upon. They are getting access to not only your electricity, but also your location.
Imagine if your cell phone company enable wifi hotspot for free on your cell phone for whomever was near by and had an account with your cell company. What if Microsoft started implementing CDN services into Windows to reduce load for Windows Update? Hey, Microsoft owns the OS, so no problem, right?
What if your car dealership allowed other people to use your leased car while you were at work? Hey, you're not using it and you don't own it.
All DOCSIS networks are like this, they just hope that most users don't actually make use of their connection. Adding the hot spot increases the number of customers making use of a given node, causing more spikes of traffic.
But you share that 1gb with up to 500 other customers and your 1gb COAX connection shares a 10gb uplink with up to 5,000 customers. With COAX, there is a lot of sharing going on. This is another way to increase the over-subscription even more.
While it's possible that your local link to the node is fine, a node can have upwards of 5,000 customers all sharing a single or teamed 10gb uplink. With 100mb internet becoming common, that's 500,000mb of customer bandwidth being forced into a 10gb-20gb uplink. The last thing you need is a bunch of freeloaders putting more stress on those poor nodes.
If Comcast was doing the Google Fiber setup, it would be more like 4tb of customer bandwidth sharing 4tb of "node" bandwidth.
Buffer bloat exasperates congestion by causing more extreme swings between over-saturated and under-saturated, but there shouldn't be congestion in the first place. Netflix is not attempting to send data at your full connection speed, except for the fraction of a second for initial buffering. Most of the time, they're only attempting to send 2mb/s
Some people have already asked Netflix what they mean and Netflix has stated that their own network and servers are not over-loaded, so the issue must be on the other side.
Here's an analogy for you. I'm wondering why my mail is getting lost or delayed. So I watch each step of the process between my business and the post office. I see that the mail is always getting to the post office on the same day, never delayed or lost. Then I find that my customers are getting mail late or lost.
So you could have a 10gig agreement with AT&T and a 5gig agreement with Level3 and be doing fine. 30% of all peak traffic comes from Netflix. But Netflix has their peering agreement with AT&T so you're all good. Then, suddenly, Netflix switches peering hosts and goes to Level3.
Level 3 has stated that this is common issue across the entire Internet, which is why Level 3 has an average peak port utilization of 37%. Level 3 has designed their network to handle large shifts. You can pay Level 3 to handle peering for you or you can do it yourself, but don't come crying when someone changes routes.
Netflix stated that they had to hire Cogent because Verizon refused to accept Netflix traffic from any other CDN. Netflix stated that they were willing to pay the higher price of Level 3, but Verizon wouldn't accept it.
Maybe Verizon knew that Cogent was bad and wanted to try to cause Netflix into a "guilt by association" situation. Or maybe Verizon finds it easier to flex against Cogent than Level 3, who is many times larger than Verizon when it comes to transit.
Im not sure where people are getting these "facts" about the dismantled middle class but theyre terribly wrong
$18k was middle class black in 1950, but once you adjust for inflation, that means in 2004, they should be making about $54k assuming a low 2% inflation, but they're only making $31k. So the median white man when from middle class in 1950 to "lower class" in 2004. The other demographics have gone up, and that's great, but we have fewer middle class.
OpenSSL design is fundamentally flawed. Bug fixes will probably introduce more bugs in many cases. I mean, OpenSSL will use your actual private key as a source of entropy. How messed up is that?
Netflix tried to acquire access to Comcast from several other CDN providers and Comcast said they will not allow Netflix data from any other CDN, only Cogent. Then Comcast complained about Cogent.
I'm glad people are coming with ideas of how to detect it other than gravitational lensing far far away from us. I can't wait to see what data we get in a few years.
This has no effect on gravitationally bound systems, which confuses me considering the whole universe, i would think, is a gravitationally bound system
Once an object reaches escape velocity, it is no longer gravitationally bound, no matter how close. You can think of the expansion of space time as a form of acceleration, and as long as two objects are gravitationally bounded, the gravity will over-whelm the expansion.
I wonder how that could affect the perceived orbits, if at all. I'm sure someone could contrive an extreme situation where one object is right at the break even point between gravity and space expansion, and instead of the two objects falling towards each other, they just stay motionless.
Slashdot Mirror
Routers have buffers measured in milliseconds. Everything else gets dropped. Even at a slow 10gb rate, 1 seconds of buffering is over 1GB of memory. Once you start getting into the terabit ranges, the bandwidth coming off those links is faster than the L1 cache on your 5ghz OC'd Intel CPU. You couldn't buffer that to memory if you wanted to.
The points of congestion are typically the hardest to QoS. Once you start getting into the 400gb+ link speeds, you can't do real time QoS anymore. Enabling QoS slows you down to ~150gb/s. I assume it'll be much worse at terabit speeds. When you're talking about moving one bit every 1/1000th of a nanosecond, you don't have time to QoS. Eventually we'll be doing photonic routing. It will be interesting trying to "buffer" photons.
We're already reaching the point of physics where the only way to route data is first-in-first-out. Anything else is nearly impossible or comes at a huge performance hit.
Use site local IPv6 addresses or use DNS. You shouldn't rely on your public IPs.
Most people use DNS instead of IP directly. I can ping by name every device on my network, and it worked like that out of the box. That includes my wireless printer and wife's android cell phone. My 7 year old Netgear router even shows the name of each device next to their MAC addresses when looking in the config. My PFSense box that I later setup automatically detected every device's name and added it to the Unbound DNS server.
HPD110.localdomain. 3600 IN A 192.168.1.9
C:\Windows\system32>ping HPD110
Pinging HPD110.localdomain [192.168.1.9] with 32 bytes of data:
Reply from 192.168.1.9: bytes=32 time=3ms TTL=255
This DNS entry automatically got created. But I can ping "HPD110" even without a DNS entry, because of whatever common protocol is used.
Here's what my wireless router shows for "HPD110".
# IP Address MAC Address Device Name
1 192.168.1.9 68:B5:99:3E:03:4B HP68B5993E034B
C:\Windows\system32>ping HP68B5993E034B
Pinging HP68B5993E034B [192.168.1.9] with 32 bytes of data:
Reply from 192.168.1.9: bytes=32 time=2ms TTL=255
I did not set up any of these names, they all automatically configured themselves.
You have to show proof that you need them and ICANN can reclaim IP blocks if they "feel" you don't "need" them anymore.
In the USA, new rules are going into affect now that we on the last block. One of those is transferring a block has a fixed cost of $10/ip. Can get pricey for a large blocks. You soon must also show they you actually need all the IPs you have, otherwise they can forcefully take them back. And the proof of "needing" will get progressively harder.
You can't change "change" IPv4, it would require an entirely new protocol, even if everything else was treated the same. Large address spaces allow for wasteful use to reduce fragmentation. Fragmentation is the bane of core routers.
Just have more IPs that you could ever want in 100 years, and 100 years from now, you'll need a new protocol. That's IPv6.
And you electricity, for free! They also increase the amount of interference going on and they are getting a service for free that was not agreed upon. They are getting access to not only your electricity, but also your location.
Imagine if your cell phone company enable wifi hotspot for free on your cell phone for whomever was near by and had an account with your cell company. What if Microsoft started implementing CDN services into Windows to reduce load for Windows Update? Hey, Microsoft owns the OS, so no problem, right?
What if your car dealership allowed other people to use your leased car while you were at work? Hey, you're not using it and you don't own it.
All DOCSIS networks are like this, they just hope that most users don't actually make use of their connection. Adding the hot spot increases the number of customers making use of a given node, causing more spikes of traffic.
But you share that 1gb with up to 500 other customers and your 1gb COAX connection shares a 10gb uplink with up to 5,000 customers. With COAX, there is a lot of sharing going on. This is another way to increase the over-subscription even more.
The same way the bank owns your house of you have a mortgage, so the bank can stop over and let people into your house without warning, right?
While it's possible that your local link to the node is fine, a node can have upwards of 5,000 customers all sharing a single or teamed 10gb uplink. With 100mb internet becoming common, that's 500,000mb of customer bandwidth being forced into a 10gb-20gb uplink. The last thing you need is a bunch of freeloaders putting more stress on those poor nodes.
If Comcast was doing the Google Fiber setup, it would be more like 4tb of customer bandwidth sharing 4tb of "node" bandwidth.
"ignorance" it is! Really, it is.. sorry. Thanks for the correction.
I r blind. Thank you my good sir.
I don't pay to access my ISP's network, I pay to access the Internet. An ISP sells "Internet access".
The second parts really isn't caching examples. If it's not 100% transparent to the end user, it's not caching, then it's just mirroring.
Buffer bloat exasperates congestion by causing more extreme swings between over-saturated and under-saturated, but there shouldn't be congestion in the first place. Netflix is not attempting to send data at your full connection speed, except for the fraction of a second for initial buffering. Most of the time, they're only attempting to send 2mb/s
Some people have already asked Netflix what they mean and Netflix has stated that their own network and servers are not over-loaded, so the issue must be on the other side.
Here's an analogy for you. I'm wondering why my mail is getting lost or delayed. So I watch each step of the process between my business and the post office. I see that the mail is always getting to the post office on the same day, never delayed or lost. Then I find that my customers are getting mail late or lost.
I would blame the post office.
So you could have a 10gig agreement with AT&T and a 5gig agreement with Level3 and be doing fine. 30% of all peak traffic comes from Netflix. But Netflix has their peering agreement with AT&T so you're all good. Then, suddenly, Netflix switches peering hosts and goes to Level3.
Level 3 has stated that this is common issue across the entire Internet, which is why Level 3 has an average peak port utilization of 37%. Level 3 has designed their network to handle large shifts. You can pay Level 3 to handle peering for you or you can do it yourself, but don't come crying when someone changes routes.
Netflix stated that they had to hire Cogent because Verizon refused to accept Netflix traffic from any other CDN. Netflix stated that they were willing to pay the higher price of Level 3, but Verizon wouldn't accept it.
Maybe Verizon knew that Cogent was bad and wanted to try to cause Netflix into a "guilt by association" situation. Or maybe Verizon finds it easier to flex against Cogent than Level 3, who is many times larger than Verizon when it comes to transit.
Im not sure where people are getting these "facts" about the dismantled middle class but theyre terribly wrong
$18k was middle class black in 1950, but once you adjust for inflation, that means in 2004, they should be making about $54k assuming a low 2% inflation, but they're only making $31k. So the median white man when from middle class in 1950 to "lower class" in 2004. The other demographics have gone up, and that's great, but we have fewer middle class.
At least Toyota took is seriously and eventually paid NASA to look over their system. GM just tried to ignore the issue and cover it up.
OpenSSL design is fundamentally flawed. Bug fixes will probably introduce more bugs in many cases. I mean, OpenSSL will use your actual private key as a source of entropy. How messed up is that?
Netflix tried to acquire access to Comcast from several other CDN providers and Comcast said they will not allow Netflix data from any other CDN, only Cogent. Then Comcast complained about Cogent.
I'm glad people are coming with ideas of how to detect it other than gravitational lensing far far away from us. I can't wait to see what data we get in a few years.
This has no effect on gravitationally bound systems, which confuses me considering the whole universe, i would think, is a gravitationally bound system
Once an object reaches escape velocity, it is no longer gravitationally bound, no matter how close. You can think of the expansion of space time as a form of acceleration, and as long as two objects are gravitationally bounded, the gravity will over-whelm the expansion.
I wonder how that could affect the perceived orbits, if at all. I'm sure someone could contrive an extreme situation where one object is right at the break even point between gravity and space expansion, and instead of the two objects falling towards each other, they just stay motionless.