For gaming/low latency: AT&T DSL vs TWC cable vs U-Verse

[Ohrami]

I posted here a while ago asking about routing issues I had with Time Warner Cable in Louisville, Kentucky, and how I got pings of 46 to Chicago (~270 mile distance), pings of 30 to Ashburn (~450 miles distance), and still my ping wasn't as good as I would have expected considering the distance.

Recently, I've been experiencing extremely high latencies at night time (60-70 to Ashburn at around 9 PM and peaks to 120-400 jitter by 10 or 11). I know that DSL doesn't have these problems since they run everyone through a dedicated line. The only ISPs available here in Kentucky are AT&T and Time Warner Cable, as well as a couple of smaller local ISPs. Which would you expect to have lower ping: AT&T's DSL, AT&T's U-Verse, Time Warner's cable, or one of the local DSL companies? Bandwidth is not an issue. I'm willing to give as many specifics as I possibly can. Any other tips about decreasing ping through hardware tweaks or software tweaks are appreciated. Thanks.

 

[PliotronX]

I'd say easily the U-Verse simply because fiber optics do not require electronic interleaving of the signal so the latency is purely based on how the ISP routes traffic. You will have more consistent pings with DSL than cable yes, but unless you can confirm that your line has been configured with Fast Path rather than interleaved, it will average a higher ping than your cable connection during the good parts of the day. Overall I would pursue the U-Verse. Fiber is the way of the future and if you can hop on the bandwagon you might have a chance of getting to the speeds some day that were promised to us by the year 2000 with a $200 billion grant (don't mean to go off on a tangent but the state of American broadband is depressing).

 

[Mushkins]

I remember that thread, and I'll make the same point that I did before. Your pings of 30 and 46 over a 300-500 mile distance are actually pretty good. Your typical ping from east coast to west coast US is about 120-160ish depending on the route and the endpoints.

The bottom line is that unless one service has some sort of glaring issue in your local node or a legitimate routing problem, it's not going to make a lick of difference in your latency which one you use. All three are providing you the "last mile," the second your transmission crosses from their network into someone elses custody (comcast, verizon, whoever else owns pieces of the backbone between you and Point A) your service provider stops mattering entirely and it's just regular old traffic getting routed around the internet. Unless you're renting a dedicated pipe from your house to the game server, which would cost a small fortune, playing musical chairs with your ISPs is not going to increase your performance at all.

You're not going to decrease your latency with hardware or software tweaks in any meaningful way.

 

[PliotronX]

I agree in general but the interleaving depth is a special case as far as DSL goes. Most ISP's will force interleaving by default to decrease instances of page timeouts for marginal lines but they default to it on spectacular lines because of laziness in measuring line noise. The interleaving causes a baseline increase of 30-50ms which means after the craptacular routing of residential ISP's you're looking at 90-110ms average. Once Fast Path is enabled on DSL, this will drop to 25-60ms for the same damn servers. It was a point of contention with Qwest because they refused to even talk with customers about the possibility of enabling Fast Path. You also may need to switch to a POTS splitter rather than using the same sockets and filters on the voice ports as those filters can increase latency further.

FTTH is the best of both worlds, no interleaving and non-shared nodes so any crappy routing is entirely past the last mile where anything on your end truly won't make a difference.

 

[Ohrami]

I remember that thread, and I'll make the same point that I did before. Your pings of 30 and 46 over a 300-500 mile distance are actually pretty good. Your typical ping from east coast to west coast US is about 120-160ish depending on the route and the endpoints.

The bottom line is that unless one service has some sort of glaring issue in your local node or a legitimate routing problem, it's not going to make a lick of difference in your latency which one you use. All three are providing you the "last mile," the second your transmission crosses from their network into someone elses custody (comcast, verizon, whoever else owns pieces of the backbone between you and Point A) your service provider stops mattering entirely and it's just regular old traffic getting routed around the internet. Unless you're renting a dedicated pipe from your house to the game server, which would cost a small fortune, playing musical chairs with your ISPs is not going to increase your performance at all.

You're not going to decrease your latency with hardware or software tweaks in any meaningful way.

That clearly isn't true when my TWC ping gets ridiculously high at night time, which wouldn't happen with DSL. Also, 30-46 ping is a bad ping for that distance. 270 miles distance shouldn't have +15 ping versus a 450 miles distance, anyway. I would expect 10-20 ping at these distances, which is what most people who I've surveyed usually get.

I don't really care if you think my ping is good or not; I just want to know which among those I've listed would be best for latency.

 

[Ohrami]

I'd say easily the U-Verse simply because fiber optics do not require electronic interleaving of the signal so the latency is purely based on how the ISP routes traffic. You will have more consistent pings with DSL than cable yes, but unless you can confirm that your line has been configured with Fast Path rather than interleaved, it will average a higher ping than your cable connection during the good parts of the day. Overall I would pursue the U-Verse. Fiber is the way of the future and if you can hop on the bandwagon you might have a chance of getting to the speeds some day that were promised to us by the year 2000 with a $200 billion grant (don't mean to go off on a tangent but the state of American broadband is depressing).

I read that U-Verse uses ADSL technology which inherently causes a bad ping, and that it's thus one of the worst ISPs to use for low latency/ping. How much truth is there to that?

 

[serpretetsky]

I read that U-Verse uses ADSL technology which inherently causes a bad ping, and that it's thus one of the worst ISPs to use for low latency/ping. How much truth is there to that?

... Most ISP's will force interleaving by default to decrease instances of page timeouts for marginal lines but they default to it on spectacular lines because of laziness in measuring line noise. The interleaving causes a baseline increase of 30-50ms which means after the craptacular routing of residential ISP's you're looking at 90-110ms average. Once Fast Path is enabled on DSL, this will drop to 25-60ms for the same damn servers. It was a point of contention with Qwest because they refused to even talk with customers about the possibility of enabling Fast Path...

Translation: by default many ISP turn on interleaving. This typically adds lag.

My parents have att uverse, they ping google at roughly 65ms. Compare that to 23ms when they had regular adsl with att before we were forced to switch to uverse.

I have never bothered trying to call ATT to see if this interleaving bologna is true or not and whether or not ATT can enable fastpath instead.

 

[JackMDS]

All of the above does not really matter. There is Nothing that you can do on your System to improve the general Pings over the ISP servers' systems.

If pings of game are so important to you well being Find where the ISP/Server first in line to where that you favorite Game site is connected to the Internet. Relocate your domicile ti the same place, and you will be very happy.

 

[Ohrami]

That's precisely why I was asking which ISP servers' systems would be best for low ping.

 

[Mushkins]

That clearly isn't true when my TWC ping gets ridiculously high at night time, which wouldn't happen with DSL. Also, 30-46 ping is a bad ping for that distance. 270 miles distance shouldn't have +15 ping versus a 450 miles distance, anyway. I would expect 10-20 ping at these distances, which is what most people who I've surveyed usually get.

I don't really care if you think my ping is good or not; I just want to know which among those I've listed would be best for latency.

/sigh, and this is how the last thread ended up going too.

It's absolutely true. What you're experiencing with high pings during peak times *is* a local node problem and you should be calling TWC to complain. You share a block of bandwidth with your entire physical area, who all also have TWC service. Even if you're not topping out your purchased bandwidth from TWC, your neighbors usage directly impacts your own service. This is how cable internet works as a technology, there is nothing you can do about it but complain that your service is sub-par during peak hours to TWC and hope they do something about it. Could be a line is damaged on your node, could be the node is saturated from hitting capacity and they need to allocate more bandwidth or update physical infrastructure. If TWC is anything like Comcast, they'll send a tech out to check your lines and tell you everything is working properly before they do nothing about it.

And no, 30-46 ping over the public internet at that distance is not a bad ping, it's an average ping whether you want to believe it or not. If you want better pings, the only thing you can do is precisely what JackMDS said, physically move to a place with a more optimal route to your target server. Assuming all of the ISPs in question are not having a local issue and all equipment is working to spec, none of them will be better or worse for latency as the latency is not coming from their network.

 

[PliotronX]

I read that U-Verse uses ADSL technology which inherently causes a bad ping, and that it's thus one of the worst ISPs to use for low latency/ping. How much truth is there to that?

Well shoot, here my thinking was AT&T went as Verizon with FiOS but as long as they're using copper there is going to be an interleaving depth issue.

Translation: by default many ISP turn on interleaving. This typically adds lag.

My parents have att uverse, they ping google at roughly 65ms. Compare that to 23ms when they had regular adsl with att before we were forced to switch to uverse.

I have never bothered trying to call ATT to see if this interleaving bologna is true or not and whether or not ATT can enable fastpath instead.

If you are able to log into the interface of the DSL modem, some of them will show you the ATM mode and fast path is no joke. My gateway pings went from 30ms to 8 when I lived in California with Verizon. This made the absolute lowest pings I had ever seen go from 50ms to 23ms and was a whole new ballgame. It's worth a shot as long as you're not dealing with Qworst/CenturyLink.

 

[imagoon]

That clearly isn't true when my TWC ping gets ridiculously high at night time, which wouldn't happen with DSL. Also, 30-46 ping is a bad ping for that distance. 270 miles distance shouldn't have +15 ping versus a 450 miles distance, anyway. I would expect 10-20 ping at these distances, which is what most people who I've surveyed usually get.

I don't really care if you think my ping is good or not; I just want to know which among those I've listed would be best for latency.

High pings at night is a node local issue. The exact same thing happens on DSL. It is not "dedicated" like you claim. DSL is a shared service from the CO to the internet, dedicated "last mile." Effectively the same thing.

DSL and cable can both have single digit latency. It depends on the implementation and the plant. With both, you still cannot control the internet routing.

 

[serpretetsky]

All of the above does not really matter. There is Nothing that you can do on your System to improve the general Pings over the ISP servers' systems.

I don't think anyone above you recommended anything about doing anything to the OP's system

/sigh, and this is how the last thread ended up going too.

It's absolutely true. What you're experiencing with high pings during peak times *is* a local node problem and you should be calling TWC to complain. You share a block of bandwidth with your entire physical area, who all also have TWC service. Even if you're not topping out your purchased bandwidth from TWC, your neighbors usage directly impacts your own service. This is how cable internet works as a technology, there is nothing you can do about it but complain that your service is sub-par during peak hours to TWC and hope they do something about it.

Isn't that why the OP started this thread, to ask which ISP technologies (and which companies in which areas) do not have this problem?

High pings at night is a node local issue. The exact same thing happens on DSL. It is not "dedicated" like you claim. DSL is a shared service from the CO to the internet, dedicated "last mile." Effectively the same thing.

DSL and cable can both have single digit latency. It depends on the implementation and the plant. With both, you still cannot control the internet routing.

Don't most cable providers service smaller residential neighborhoods than DSL typically does. Don't they use slower equipment for each neighborhood?

Doesn't this "high ping at peek usage" issue appear more often in cable companies? I really don't know, i've never seen any statistics on it.

 

[imagoon]

I don't think anyone above you recommended anything about doing anything to the OP's system

Isn't that why the OP started this thread, to ask which ISP technologies (and which companies in which areas) do not have this problem?

Don't most cable providers service smaller residential neighborhoods than DSL typically does. Don't they use slower equipment for each neighborhood?

Doesn't this "high ping at peek usage" issue appear more often in cable companies? I really don't know, i've never seen any statistics on it.

He has a previous thread where we explained this issue in detail. His specific issue is the routing of the tier 3 internet providers and not a local last mile issue in most cases. As for the high pings at night, that is first example of a local last mile issue that he had so he should complain to his provider.

As for your "general" questions I can't answer them specifically down to the location. All I can draw on is the fact that Cable and DSL are both shared bandwidth techs. The latency issues come down to upstream management. DSL can be choked at the street node (ala uverse style) or the central office. It also has some extreme distance issues in some areas and also may have interleaving techs involved etc. As for your cable question about "slower connections" the question is far to broad to answer and I am not a cable internet plant operator. It would also depend on the end point to central node ratio among a lot of other things. Your ping question is also to general to answer. It depends on the plant designs of each tech. The DSL company can drop 45mbps 5mbps to 8 users in a pedestal. Cable can have 45mbps or 5mbps to 8 people also. It all depends on the plant and there is no way to solidly say "most" one way or the other.

 

[serpretetsky]

He has a previous thread where we explained this issue in detail. His specific issue is the routing of the tier 3 internet providers and not a local last mile issue in most cases. As for the high pings at night, that is first example of a local last mile issue that he had so he should complain to his provider.

Well, since we agree that his late night latency problems is probably NOT level 3 problems, we can also agree that there is some probability that another ISP in his area will not have the same late night problems (although they might not improve latency during other times). Sure he can complain to his ISP about it, but somehow I doubt the ISP would really change anything. But i suppose it's worth a try.

If there is, however, an ISP without this late night issue, then what the OP really needs is for someone in his specific part of town with a different ISP to chime in and relate his experiences using the same testing methodology that he is using.

yay /nay?

 

[Mushkins]

Well, since we agree that his late night latency problems is probably NOT level 3 problems, we can also agree that there is some probability that another ISP in his area will not have the same late night problems (although they might not improve latency during other times). Sure he can complain to his ISP about it, but somehow I doubt the ISP would really change anything. But i suppose it's worth a try.

If there is, however, an ISP without this late night issue, then what the OP really needs is for someone in his specific part of town with a different ISP to chime in and relate his experiences using the same testing methodology that he is using.

yay /nay?

That's the practical answer, but not what the OP was asking. If you read the other thread he made on essentially the same topic a while back, it's clear he does not understand networking at all and just wants a magic button to make his internet connection as fast as he thinks it should be. When people try to explain the technology, he gets angry and defensive and tells us we're all wrong and don't know what we're talking about.

His original question was which one of these technologies is inherently going to provide lower latency to his gameservers than the others, and the right answer is "none of them" because the server is not on his local node. Assuming none of them have local issues, once it leaves the local node it's ending up on the same backbone, with the same routes, to the same server, and back. Changing the last mile provider does not address the actual cause of the latency and there's nothing that can be done about it.

 

[imagoon]

I would also add that in their purest forms, most copper services are comparable in latency. DSL vs Cable is nearly identical. The performance difference comes from plant design and over subscription ratios. In our network we run a lot of VPNs to small 2-10 person offices. We have cable and DSL lines based on availability. The latency at the sites varies a lot based on location more than service. I have a few cable vpns showing sub 12ms round trips and another one 3 blocks away that is showing 45ms round trips. Same for DSL.

 

[PliotronX]

That's the practical answer, but not what the OP was asking. If you read the other thread he made on essentially the same topic a while back, it's clear he does not understand networking at all and just wants a magic button to make his internet connection as fast as he thinks it should be. When people try to explain the technology, he gets angry and defensive and tells us we're all wrong and don't know what we're talking about.

His original question was which one of these technologies is inherently going to provide lower latency to his gameservers than the others, and the right answer is "none of them" because the server is not on his local node. Assuming none of them have local issues, once it leaves the local node it's ending up on the same backbone, with the same routes, to the same server, and back. Changing the last mile provider does not address the actual cause of the latency and there's nothing that can be done about it.

I didn't see the other thread so I'll write him off as hopeless but the bolded part caught my attention because, perhaps this city of 80k is unique, but up until a backbone that Bresnan/Optimum piped out to the internet, a 10Gb pipe was the only way out to Denver and the internet but the routing is completely different betwixt the cable and DSL providers out here.

 

[PliotronX]

I would also add that in their purest forms, most copper services are comparable in latency. DSL vs Cable is nearly identical. The performance difference comes from plant design and over subscription ratios. In our network we run a lot of VPNs to small 2-10 person offices. We have cable and DSL lines based on availability. The latency at the sites varies a lot based on location more than service. I have a few cable vpns showing sub 12ms round trips and another one 3 blocks away that is showing 45ms round trips. Same for DSL.

Not always true, while cable always has interleaving, the depth is measured in nanoseconds rather than milliseconds while DSL standard depth is north of 20ms so the only way to compare apples to apples is with Fast Path on DSL. If you're lucky enough to have a DSL provider enable FP by default, you've never seen the woes of interleaving

 

[imagoon]

Not always true, while cable always has interleaving, the depth is measured in nanoseconds rather than milliseconds while DSL standard depth is north of 20ms so the only way to compare apples to apples is with Fast Path on DSL. If you're lucky enough to have a DSL provider enable FP by default, you've never seen the woes of interleaving

Yes, hence the "The performance difference comes from plant design" comment. That is a plant design decision.

 

[PliotronX]

Yes, hence the "The performance difference comes from plant design" comment. That is a plant design decision.

Well it's not entirely fair to carte blanche say they are identical because they are not, there are too many nuances end-to-end. For example, the end user's decision between filtering voice with the data signal also introduces throughput and latency impedance versus installing a POTS splitter. DSL is a paired conductor while coaxial is a single conductor utilizing frequency division multiplexing where the varying frequencies travel along the same conductor for reception and transmission. Here is a good read.

 

[imagoon]

Well it's not entirely fair to carte blanche say they are identical because they are not, there are too many nuances end-to-end. For example, the end user's decision between filtering voice with the data signal also introduces throughput and latency impedance versus installing a POTS splitter. DSL is a paired conductor while coaxial is a single conductor utilizing frequency division multiplexing where the varying frequencies travel along the same conductor for reception and transmission. Here is a good read.

You keep bring this up but the picoseconds you are talking about I am not that concerned about. The latency for a small packet (1500) bytes is going to not vary a lot between the techs. 8mbps / dsl vs 80mbps cable changes the local latency by 1.3ms. (not adjusted for distance.)
1500byte packet
DSL: 1.43
cable: 0.14

Typical ICMP ping: 56 bytes
DSL 0.0000534 seconds
cable 0.00000534 seconds.

Anything else like adding voice or shared service at the CO vs the cable being a shared medium is a plant decision that can affect performance.

DSL is FDM same as cable also.

 

[PliotronX]

You keep bring this up but the picoseconds you are talking about I am not that concerned about. The latency for a small packet (1500) bytes is going to not vary a lot between the techs. 8mbps / dsl vs 80mbps cable changes the local latency by 1.3ms. (not adjusted for distance.)
1500byte packet
DSL: 1.43
cable: 0.14

Typical ICMP ping: 56 bytes
DSL 0.0000534 seconds
cable 0.00000534 seconds.

Anything else like adding voice or shared service at the CO vs the cable being a shared medium is a plant decision that can affect performance.

DSL is FDM same as cable also.

I guess I just don't get your point, the "plant decision" is the most significant factor in the differences beyond the location of node sharing and ratio of overselling which is cable's only weakness. The point that copper is copper would mean that 56k has the same latency as DSL.

edit- Also, I'm not sure where you are getting those figures. What you seem to be talking about is the physical layer which is transmission of bits. The MTU framing takes place in the data link layer which is inconsequential of the physical limitations of DSL (by default).

 

[imagoon]

I guess I just don't get your point, the "plant decision" is the most significant factor in the differences beyond the location of node sharing and ratio of overselling which is cable's only weakness. The point that copper is copper would mean that 56k has the same latency as DSL.

edit- Also, I'm not sure where you are getting those figures. What you seem to be talking about is the physical layer which is transmission of bits. The MTU framing takes place in the data link layer which is inconsequential of the physical limitations of DSL (by default).

I did the math. 56k @ single bit has near the same latency as DSL and cable. Larger packets suffer due to the slower bit rate increasing the round trip on the circuit.

The issue you are seeming to miss is that in their base vanilla forms both techs are not that far off in their raw specs. User to node counts and all that is a plan decision. 50 people on a cable node vs 50 users on a DSLAM rack is a plant decision. The end results are pretty similar also. The only variance is the bandwidth which when using FDM, the deltas are minimal based on the channel counts. Cable would be able to attain high speeds but the small packet latency is going to still be very close.

 

[PliotronX]

I did the math. 56k @ single bit has near the same latency as DSL and cable. Larger packets suffer due to the slower bit rate increasing the round trip on the circuit.

The issue you are seeming to miss is that in their base vanilla forms both techs are not that far off in their raw specs. User to node counts and all that is a plan decision. 50 people on a cable node vs 50 users on a DSLAM rack is a plant decision. The end results are pretty similar also. The only variance is the bandwidth which when using FDM, the deltas are minimal based on the channel counts. Cable would be able to attain high speeds but the small packet latency is going to still be very close.

Dude we're in strange territory. The electron latency is irrelevant, the DAC/ADC process means 56k has a minimum of 60ms one direction, that is not a plant decision that is a physical limitation. What I'm saying is that because of DSL's typical implementation it is often inferior to cable in latency to the gateway. You have to consult with the provider to get it to even compete with cable, routing differences granted.