I second this. I had a $200 router a few years back and it had issues with torrent. I never had these issues until I got my dedicated connection and could actually get decent speed from torrent, but there you go. And it showed up as packetloss, which is strange. I could get 100% of my connection speed with the router and torrent, but as long as I limited the connections. As the connection count got higher, packetloss and jitter started to increase, even with low bandwidth, like 1% of my connection.
It was very binary. Once I reached a certain connection count in the thousands, all of a sudden packetloss and jitter kicked in, which is very unexpected since it didn't actually break existing connections or deny me from creating new ones. Strange way to degrade.
My ISP just promises that congestion will never happen on their network. Is that so much to ask for others to do the same? They explicitly told me congestion management is a hard moving target with too many problems. Much cheaper to just throw more bandwidth at the problem until the congestion is gone. They took the approach to the point of no over-provisioning. My personal sales person, who had been with the company for over 20 years, told me I can use my connection however I want, as long as it's not illegal. If I want to use it 100%, I am free to do so, they promise to never punish me in any way. They also promise I will never experience congestion on their network. If I do, to call them immediately and report the issue. Even their ToS says they will never QoS or traffic shape any traffic. All traffic will be treated exactly the same.
I've load tested my connection during peak hours by downloading a bunch large files from many servers, enough to keep the link saturated for hours on end. On a 100Mb connection, one minute averages were 99.5Mb/s +- 0.25Mb/s, and my ping stayed below 0.2ms and 0 packets lost. All during peak hours. Oh yeah, zero buffer bloat.
All of this tagging crap just to manage congestion sounds like a bunch of busy work. Not to say they aren't using TE's in other ways.
Edge routers, there's your problem? My residential connection plugs pseduo-directly into the core router. 0.14ms pings to the core router, yeah baby. CDNs? Blasphemy. I asked a manager when they were getting a Netflix CDN, their response, "We don't like to prefer one company over another and our trunk is just fine, we have no reasons to use CDNs".
Topology: My Firewall - ONT - Aggregation switch with zero oversubscription - Core Router - 3 links to Level 3 each pair with a failover. My trace route looks like this: Hop 1 ISP Hop 2 Level 3, and like I said, that first hop is 0.14ms.
The network topology is provisioned that 100% of all customers could use 100% of their connections at the same time, ignoring melting the trunk. But the trunk is sized 3x the 95th percentile with another 3x for fail over, so 6x 95th percentile in total. Because they have no CDNs or peers, they're very proactive about keeping their trunk sized with room to breath.
If a small ISP in the sticks can do this, why can't you?
I have a 6 hop, 110ms ping 1m jitter to AWS Frankfurth Germany from Midwest USA because Level 3 rocks, and I get 100% of my bandwidth. Using DSLReports jitter test, I get 1ms of jitter to all of their servers around the globe.
You are right in some completely asinine and idiotic sense, in that a synchronous interface is always either sending a user payload frame or a pad frame
I was actually going for this, but I didn't get the reason why across very well, but I attempted to explain in the last paragraph. At some really small time frames the link is always at 100% or 0%. The reason I went to this extreme is to make a point that it's only once you start to get to macro level time scales, like milliseconds, that all you care about is averages. At some point you can say "I never have more than 1ms of buffer latency over this link", and that is what you're goal is. You want the latency to always be low, meaning the link is not congested.
95th-percentile billing, we have to provision/pay capacity for peak demand
My ISP has roughly 6x peak bandwidth and all traffic is transit via Level 3, no CDNs or peering. 100/100 for $90. No caps, ISP claims you should get 100% of your bandwidth 100% of the time, which is why they have 6x peak. Technically 3x peak as live links and 6x because each of those links has a fail over that can be teamed to double bandwidth on request.
A lot of small ISPs complain about similar things as you. Is it a scale thing or a lack of competition? Maybe my ISP is lucky and has access to several trunk options, even out here in the middle of farm land. our first hop is about 250 miles, but takes a 400 mile route. Do you not have access to a decent hub within 250 miles?
Cost wise, a customer consuming 1TB via torrent is about the same as one consuming 100Gb via Netflix. Most high speed dedicated links are billed based on 95th percentile, which means the peak 1.5hours.
4% on revenues included TV services, which make up a huge portion of the overall revenue and are incredibly small margins. ISPs make decent margins, TV providers do not. The only reason Verizon is selling off is because they want to liquidate a reliable 4% income so they can dump that money into wifi and make much higher revenues. Remember when text messaging costed more per byte than renting time from the Hubble Telescope? There is a lot of money to be made in price gouging mobile data.
Almost no ISPs use any kind of AQMs. You could do a high sample trace route and watch for jitter and avg ping increases. In general, a health link's avg ping should not be more than 1-5ms over the minimum. If you do enough samples in a trace route, you should be able to see which hops are causing issues. If your ISP is doing high quality shaping along with an AQM, it would be much harder to watch for congestion because latency should be quite stable. Then you need to somehow measure a route's bandwidth.
The good news is most decent traffic shaping algorithms are very CPU intensive relative to the amount of bandwidth an ISP's core network must handle and the algorithms do not scale well with the number isolated groups. In other words, you should be able to detect jitter and avg ping increases.
Your argument sounds like if you pay for X bandwidth, but less bandwidth than someone else, you should get priority over them. That makes no sense. I think what you want is a consumption based bill. Pay for what you use. But expect your bill to go up. The increased cost of managing everyone's bills will increase by more than the cost of just adding more bandwidth.
You have it all wrong, Torrent only makes up 6% of peak bandwidth. It's all of those lousy customers paying for 100Mb connections and trying to stream 3Mb/s of Netflix. Those are the greedy bastards. They should pay for 100Mb and only use 0Mb/s. Even better, lets just get rid of ISPs and make a tax that gets funneled directly into the ex-ISP's pockets.
Because the ISP model (at least in the majority of the US) is to oversubscribe their bandwidth based on the assumption that not everyone will be using it at once
The entire internet is that way, even dedicated connections. The difference is the amount of oversubscription.
First come first serve. If your ISP can't keep up, blame them, not other users for trying to make use of what they paid for. Bandwidth is the cheapest part of an ISP, cheaper than customer support even. If you want to blame people for consuming more than you, blame grandma for calling support and running up the ISP's cost of operations faster than the 24/7 torrent seeder.
TCP attempts to consume all available bandwidth. If you have a 1Gb connection, you can't consume more than 1Gb/s. The trunk should never be the bottleneck. If you're streaming Netflix, even if you transfer 1Gb/s, you're only averaging 3Mb/s. Very short lived bursts. The bigger offenders are bulk data transfers and not streaming. Even then, how much data can you download before you fill up your HD? For every customer using 1Gb/s, there's 1,000 others using 0.
In this day and age, 100Gb trunks are becoming common. Of course links are always at 100% or 0%, there is no in between. The question is how long can they be at 100% before they're considered "congested". If your trunk is never at 100% for more than a few fractions of a milisecond, who cares.
which is what causes single user fsck requirement.
An FS that needs fsck is broken by design. Use a modern FS. If you need a reliable boot volume, just use ZFS and do a 4 way mirror or something outrageous, two disks on one controller and 2 on another. If that gets corrupted, there is no hope, like a damaged CPU or bad motherboard.
Ms didn't lay off 18k MS programmers, they laid off 13k Nokia designers, marketing, sales, etc from the company purchase. There was a lot of redundant positions and MS trimmed the "fat".
About 80% of the known OpenSSL bugs that have been fixed, were inadvertently fixed in LibreSSL during the refactoring. Many of OpenSSL's bugs are entirely do to horrible coding practices. Of the remaining 20%, a sizable portion were actually found by LibreSSL during the clean up.
I bet you don't like some things the government does. You are invited to run for Senate or President. Because obviously if you don't, you should just shut up and gtfo.
Complaining about open source software is like voting, you're letting your voice be heard but letting the other run the show. Submitting patches is like being a politician, you're the only actually doing the work.
They want it paid off in 6 months. Technically it's not the hospital, it was a debt that went to collections and was paying $100 on it for a year, then the collections got taken over by some other d-bags and said they have a new minimum.
All mater in the Universe is made out of quantised elements, but we don't see steps because the elements are small enough. Increase the resolution and get rid of it.
Slashdot Mirror
I second this. I had a $200 router a few years back and it had issues with torrent. I never had these issues until I got my dedicated connection and could actually get decent speed from torrent, but there you go. And it showed up as packetloss, which is strange. I could get 100% of my connection speed with the router and torrent, but as long as I limited the connections. As the connection count got higher, packetloss and jitter started to increase, even with low bandwidth, like 1% of my connection.
It was very binary. Once I reached a certain connection count in the thousands, all of a sudden packetloss and jitter kicked in, which is very unexpected since it didn't actually break existing connections or deny me from creating new ones. Strange way to degrade.
My ISP just promises that congestion will never happen on their network. Is that so much to ask for others to do the same? They explicitly told me congestion management is a hard moving target with too many problems. Much cheaper to just throw more bandwidth at the problem until the congestion is gone. They took the approach to the point of no over-provisioning. My personal sales person, who had been with the company for over 20 years, told me I can use my connection however I want, as long as it's not illegal. If I want to use it 100%, I am free to do so, they promise to never punish me in any way. They also promise I will never experience congestion on their network. If I do, to call them immediately and report the issue. Even their ToS says they will never QoS or traffic shape any traffic. All traffic will be treated exactly the same.
I've load tested my connection during peak hours by downloading a bunch large files from many servers, enough to keep the link saturated for hours on end. On a 100Mb connection, one minute averages were 99.5Mb/s +- 0.25Mb/s, and my ping stayed below 0.2ms and 0 packets lost. All during peak hours. Oh yeah, zero buffer bloat.
All of this tagging crap just to manage congestion sounds like a bunch of busy work. Not to say they aren't using TE's in other ways.
Edge routers, there's your problem? My residential connection plugs pseduo-directly into the core router. 0.14ms pings to the core router, yeah baby. CDNs? Blasphemy. I asked a manager when they were getting a Netflix CDN, their response, "We don't like to prefer one company over another and our trunk is just fine, we have no reasons to use CDNs".
Topology: My Firewall - ONT - Aggregation switch with zero oversubscription - Core Router - 3 links to Level 3 each pair with a failover. My trace route looks like this: Hop 1 ISP Hop 2 Level 3, and like I said, that first hop is 0.14ms.
The network topology is provisioned that 100% of all customers could use 100% of their connections at the same time, ignoring melting the trunk. But the trunk is sized 3x the 95th percentile with another 3x for fail over, so 6x 95th percentile in total. Because they have no CDNs or peers, they're very proactive about keeping their trunk sized with room to breath.
If a small ISP in the sticks can do this, why can't you?
I have a 6 hop, 110ms ping 1m jitter to AWS Frankfurth Germany from Midwest USA because Level 3 rocks, and I get 100% of my bandwidth. Using DSLReports jitter test, I get 1ms of jitter to all of their servers around the globe.
You are right in some completely asinine and idiotic sense, in that a synchronous interface is always either sending a user payload frame or a pad frame
I was actually going for this, but I didn't get the reason why across very well, but I attempted to explain in the last paragraph. At some really small time frames the link is always at 100% or 0%. The reason I went to this extreme is to make a point that it's only once you start to get to macro level time scales, like milliseconds, that all you care about is averages. At some point you can say "I never have more than 1ms of buffer latency over this link", and that is what you're goal is. You want the latency to always be low, meaning the link is not congested.
95th-percentile billing, we have to provision/pay capacity for peak demand
My ISP has roughly 6x peak bandwidth and all traffic is transit via Level 3, no CDNs or peering. 100/100 for $90. No caps, ISP claims you should get 100% of your bandwidth 100% of the time, which is why they have 6x peak. Technically 3x peak as live links and 6x because each of those links has a fail over that can be teamed to double bandwidth on request.
A lot of small ISPs complain about similar things as you. Is it a scale thing or a lack of competition? Maybe my ISP is lucky and has access to several trunk options, even out here in the middle of farm land. our first hop is about 250 miles, but takes a 400 mile route. Do you not have access to a decent hub within 250 miles?
Cost wise, a customer consuming 1TB via torrent is about the same as one consuming 100Gb via Netflix. Most high speed dedicated links are billed based on 95th percentile, which means the peak 1.5hours.
4% on revenues included TV services, which make up a huge portion of the overall revenue and are incredibly small margins. ISPs make decent margins, TV providers do not. The only reason Verizon is selling off is because they want to liquidate a reliable 4% income so they can dump that money into wifi and make much higher revenues. Remember when text messaging costed more per byte than renting time from the Hubble Telescope? There is a lot of money to be made in price gouging mobile data.
Almost no ISPs use any kind of AQMs. You could do a high sample trace route and watch for jitter and avg ping increases. In general, a health link's avg ping should not be more than 1-5ms over the minimum. If you do enough samples in a trace route, you should be able to see which hops are causing issues. If your ISP is doing high quality shaping along with an AQM, it would be much harder to watch for congestion because latency should be quite stable. Then you need to somehow measure a route's bandwidth.
The good news is most decent traffic shaping algorithms are very CPU intensive relative to the amount of bandwidth an ISP's core network must handle and the algorithms do not scale well with the number isolated groups. In other words, you should be able to detect jitter and avg ping increases.
Your argument sounds like if you pay for X bandwidth, but less bandwidth than someone else, you should get priority over them. That makes no sense. I think what you want is a consumption based bill. Pay for what you use. But expect your bill to go up. The increased cost of managing everyone's bills will increase by more than the cost of just adding more bandwidth.
You have it all wrong, Torrent only makes up 6% of peak bandwidth. It's all of those lousy customers paying for 100Mb connections and trying to stream 3Mb/s of Netflix. Those are the greedy bastards. They should pay for 100Mb and only use 0Mb/s. Even better, lets just get rid of ISPs and make a tax that gets funneled directly into the ex-ISP's pockets.
Because the ISP model (at least in the majority of the US) is to oversubscribe their bandwidth based on the assumption that not everyone will be using it at once
The entire internet is that way, even dedicated connections. The difference is the amount of oversubscription.
First come first serve. If your ISP can't keep up, blame them, not other users for trying to make use of what they paid for. Bandwidth is the cheapest part of an ISP, cheaper than customer support even. If you want to blame people for consuming more than you, blame grandma for calling support and running up the ISP's cost of operations faster than the 24/7 torrent seeder.
TCP attempts to consume all available bandwidth. If you have a 1Gb connection, you can't consume more than 1Gb/s. The trunk should never be the bottleneck. If you're streaming Netflix, even if you transfer 1Gb/s, you're only averaging 3Mb/s. Very short lived bursts. The bigger offenders are bulk data transfers and not streaming. Even then, how much data can you download before you fill up your HD? For every customer using 1Gb/s, there's 1,000 others using 0.
In this day and age, 100Gb trunks are becoming common. Of course links are always at 100% or 0%, there is no in between. The question is how long can they be at 100% before they're considered "congested". If your trunk is never at 100% for more than a few fractions of a milisecond, who cares.
which is what causes single user fsck requirement.
An FS that needs fsck is broken by design. Use a modern FS. If you need a reliable boot volume, just use ZFS and do a 4 way mirror or something outrageous, two disks on one controller and 2 on another. If that gets corrupted, there is no hope, like a damaged CPU or bad motherboard.
"Grasping" a 1000x difference is easy. It's the difference between a millimeter and a meter.
Even if you had the plants, you still need the power. Cali has as much of an electrical crunch as it does water.
Ms didn't lay off 18k MS programmers, they laid off 13k Nokia designers, marketing, sales, etc from the company purchase. There was a lot of redundant positions and MS trimmed the "fat".
About 80% of the known OpenSSL bugs that have been fixed, were inadvertently fixed in LibreSSL during the refactoring. Many of OpenSSL's bugs are entirely do to horrible coding practices. Of the remaining 20%, a sizable portion were actually found by LibreSSL during the clean up.
I bet you don't like some things the government does. You are invited to run for Senate or President. Because obviously if you don't, you should just shut up and gtfo.
Complaining about open source software is like voting, you're letting your voice be heard but letting the other run the show. Submitting patches is like being a politician, you're the only actually doing the work.
You can donate plasma 2x a week, but not blood. Plasma is mostly water and electrolytes.
Intel Haswell 3.3ghz i5 quad
24GB DDR3
Samsung 850 EVO 500GB x2
AMD6950
Intel i210-T1 NIC - because I refuse to use the intergrated RealTek
They want it paid off in 6 months. Technically it's not the hospital, it was a debt that went to collections and was paying $100 on it for a year, then the collections got taken over by some other d-bags and said they have a new minimum.
All mater in the Universe is made out of quantised elements, but we don't see steps because the elements are small enough. Increase the resolution and get rid of it.
Generally when I run speedtest.net, it comes in between 100 and 125 down.
25% variation? What kind of crap is that? Above what you're supposed to get or not, they obviously can't control their network.
Monster Cable SATA cables, power cables, video cables, nothing but the best!