Shop Now, Avoid Delays - Get Our Best Holiday Deals!
Get superfast AT&T Fiber internet
rich719's profile

Contributor

 • 

3 Messages

Mon, Nov 2, 2015 9:37 PM

Gigapower seems to have a speed limit when leaving AT&T's network

I have gigapower (1 Gbps) in the Orlando, FL area. I can test to AT&T's speedtest.net in Orlando and get 950 Mbps or so which is what I expect on both up/down. When I switch to another test server in the area that is not hosted by AT&T, at best I can get 350 Mbps down and 200 Mbps up.

 

Okay, I understand all the networking, hops, latency, etc, etc. Been doing networking since it was still in its infancy during that mid 1980s. I have run several tests and used my network connections I have on remote servers outside of AT&T's network to test to the same speedtest.net servers. For example, I can test from Dallas to the same Orlando server that gives me bad results on Gigapower and I can get virtually my full bandwidth from Dallas over more hops.

 

Also when I test with end-to-end test equipment where I control both sides, the bandwidth seems to be clipped everytime it approaches over 200 Mbps and drops down to near 0 for a period before it ramps up again. Certain protocols seemed to be clipped worse than others for example, using an IPSec or PPTP VPN has really bad performance (16 Mbps) whereas the same computer using an iPhone and LTE data network can get over 40 Mbps.

 

I tried talking with AT&T technical support until I am blue in the face. They refuse to do anything since "Their" server tests okay and in fact, they can't/won't test the same servers from their side. I tried to get them to escellate it and was told to basically eat dirt and they will not because their server works to spec'd speeds.

 

Excuse me AT&T, I was willing to consider you had an issue that was unknown, but when a Tech support person refuses to look into it, it tells me this is on purpose and AT&T is ripping me off by charging me for 1Gbps and giving me at best 300Mbps. This is fraud or at best a horrible network.

buckeyegoose

Scholar

 • 

186 Messages

6 y ago

Not every server has a gig network connection, that would be the reason why you see slower on some servers vs others.

JefferMC

ACE - Expert

 • 

25.6K Messages

6 y ago

I think there are a whole lot of things to consider that you have chosen not to.

 

Did you check the networks being traversed? Maybe there are some peering issues with some of the networks?

 

Did you realize that diferent protocols may be handled differently by routers along the way (e.g. some routers now handle IPv4 via ASIC, not all of them will handle IPSEC in ASIC and have to let it be processed by the General Purpose CPU, which slows things down).

 

Did you realize that every ISP gets hundreds of calls a day claiming they have a problem, and that 99% of them are dead wrong, 0.9% of them are partially wrong, and 0.1% of them have a point?  How are they to know you're the 0.1%?

 

 

 

Contributor

 • 

3 Messages

6 y ago

 

Not every server has a gig network connection, that would be the reason why you see slower on some servers vs others.

 

 

I spent the time to explain all the testing I have done over the last couple of weeks and you either didn't read it or really didn't understand what I was saying. This has nothing to do with different servers being fast/slow. I eliminated that possibility by testing the same servers from different ISPs and locations.

 

I understand overhead/differences/load/saturation/etc. That is not what is in play here.

Contributor

 • 

3 Messages

6 y ago


@JefferMC wrote:

I think there are a whole lot of things to consider that you have chosen not to.

 

Did you check the networks being traversed? Maybe there are some peering issues with some of the networks?

 

Did you realize that diferent protocols may be handled differently by routers along the way (e.g. some routers now handle IPv4 via ASIC, not all of them will handle IPSEC in ASIC and have to let it be processed by the General Purpose CPU, which slows things down).

 

Did you realize that every ISP gets hundreds of calls a day claiming they have a problem, and that 99% of them are dead wrong, 0.9% of them are partially wrong, and 0.1% of them have a point?  How are they to know you're the 0.1%?

 

 

 


Actually, I have spent nearly 2 weeks testing and considering all the possible issues ruling them out through various tests. I understand different hops and the network traversed. In fact, in various tests I have mapped out the networks to see where the common networks were that were shared with each test.

 

I agree that there may be a peering issue, however, I cannot get anyone at AT&T to consider that or even look into it. The answer is, if it works on OUR network, it must be okay. You can see as soon as it leaves their network there is a potential issue which is why I want them to look into it.

 

I do know different protocols are handled different and some require CPU vs built-in hardware. I am well versed in this, however, the performance degredation down to 16 Mbps for IPSEC is not something I would consider to be a CPU vs hardware or IP packet overhead. The encryption happens at the endpoints so that overhead is not part of the ISPs, the only thing they are handling is the transfer of TCP packets. Protocols are built into those packets since we are really dealing with TCP here, not UDP or anything else like that.

 

All I am asking is for AT&T to look into the possibility of a problem within their network or more specifically with their peering connections. A couple of tests that I ran below show an example of what I am talking about.

 

Eliminating all the IPSEC, speedtest.net servers, etc. from the equation. I should not see the following when transfering TCP packets from AT&T Uverse network to a server when the same server is hit from a different ISP and gets stable results.

 

Orlando - Farm - Internet.PNG

 

When running a test through another ISP to the same destination I see a graph like the one below. This is much more consistent and stable.

 

Verizon - Farm

Need help?
New to the AT&T Community? Start by visiting the Community How-To.
New to the AT&T Community?
Visit the Community How-To.