Daniel Marks... Developer?

AT&T 5000 Fiber Review

·3 mins

Speedtest

Ordering & Install #

So AT&T released their “hyper-gig” service in January, but in the Detroit market where I live, it wasn’t marked as availible.

Intro Promo

However, I still checked anyways, and lucky for me, it allowed me to order as a new customer:

Purchase Page

Unfourtunately, since I was a current customer, I had to call in to upgrade. This was actually not terrible, except for the first rep not knowing they even offered service above 1000, and had to ask their manager how to even place the order.

Install day #

My old setup looked like this (note the lack of bypass):

Old Topology

After the installation, they removed the ONT and plugged the fiber directly into the BGW-320:

New Topology

The tech that came seemed to be pretty knowlegable and even offered to move the ONT to my server cabinet. However, I live in an apartment so I didn’t want to make it rough for the next person.

Performance #

So I have a few gripes with the way AT&T runs their network:

  • AT&T does not interconnect with anyone except at their major “hubs”, and in my case being in Michigan, that nearest hub is in Chicago.
  • AT&T generally only peers with Cogent and does not allow residential users to use it’s own (excellent) IP backbone.
  • (For me specifically) Cloudflare does not have direct interconnections with AT&T, which means it has to traverse congested GTT links in Chicago.
  • As of late, AT&T is choosing to route all traffic going to Detroit through Toledo instead of going directly to Chicago. In the past, they have used a lower latency link meaning you could reach Chicago in 8ms, but now it’s up to 13ms because of the detour. Sometimes, AT&T reverts back to the old route, and my monitoring software detects significatly lower latency to the rest of the internet.

Keep in mind the traffic engineering portion when we’re checking out speed tests and such.

Is it really 5000mbps? #

Eh.

Because AT&T has such poor interconnections with the rest of the internet (from a residential perspective), you aren’t ever going to see 5 Gbps outside of connecting to someone else on AT&T with a 5Gbps connection.

With heavy threading, you can get close.

iperf3 from AS398351 with 50 download threads #

I used the command iperf3 -c REDACTED -P50 -R -t20

[ ID] Interval           Transfer     Bitrate         Retr
[SUM]   0.00-20.02  sec  9.98 GBytes  4.44 Gbits/sec  4162             sender
[SUM]   0.00-20.00  sec  9.94 GBytes  4.42 Gbits/sec                  receiver

This is a decent result, but using 50 threads to get this result is pretty high. Normally, you’d really only need 3 or 4 to reach 5gbps with 20ms of latency, but Cogent does their own traffic engineering to limit the amount of theads.

Real world usage #

Steam #

Downloading games is probably the most bandwidth intensive thing a lot of people do, so I natually tried it first

Steam Download

Reaching ~2.4gbps is meh, but expected.

AWS #

So this one is tricky, because different AWS serices perform differently. Cloudfront for instance, has a similar limitation as Steam, and really doesn’t go higher than 200MB/s over HTTP downloads. However, using S3 and downloading/uploading files can fill up my circuit if you’re using an optimized script and multithreading the process.

Summary #

For 180/mo, the Fiber 5000 plan is not worth it with the current way AT&T handles their network unless you’re in a huge household with many power users. The Fiber 2000 plan would be something you can actually fill up if you’re the only power user.

I’m going to be keeping my Fiber 5000 plan, because I deal with AWS extensively and able to max out my connection.