What's with AT&T's routing?

[ssavoy]

I'm in Eastern Pennsylvania and every traceroute I do, AT&T always seems to pipe data to Chicago before going anywhere else. This seems highly inefficient as NYC, Philly, and DC are far closer - but every bit of data goes to Chicago first. Even when I use Speedtest servers - the best results come from the Towerstream Chicago server instead of ones in NYC or DC.

Anyone else think this is terrible routing? It's always been the norm, though.

[formercanuck]

Well, SoCal routes through San Jose - go figure.
This is why my 'speedtest' gives me better latency through Palo Alto (Fiber Internet Center) than it does on any local SoCal speedtest site.
Sad thing is - I still get better thoughput to places like San Luis Obispo - Charter than most other spots.

[Jordan T]

That's mythical nonsense, transit across the US (assuming a non-existant perfectly straight path) adds only 16ms. Within the US, latency doesn't have nearly as much to do with geographical proximity as it does number of hops and the capacity of the network routers in the way - and how "twisted" the path is (crossing the country multiple times, etc). In that way, having everything go in and out of a central data center can be the MOST efficient path.

[formercanuck]

That's mythical nonsense, transit across the US (assuming a non-existant perfectly straight path) adds only 16ms. Within the US, latency doesn't have nearly as much to do with geographical proximity as it does number of hops and the capacity of the network routers in the way - and how "twisted" the path is (crossing the country multiple times, etc). In that way, having everything go in and out of a central data center can be the MOST efficient path.

While SoCal to NorCal and back probably adds ~15-20ms, I will typically get ~90ms to Palo Alto site (Fiber Internet Center), and ~100-110ms to LA (Dreamhost).
The sad part is that I'll typically get ~150-200ms to Charter (Monterey Park or San Luis Obispo) most likely due to peering. The sad part is that I'll get better overall throughput on Charter in SoCal/SLOCal than I will out of anything else (application/load).

[silentjudge]

I'm in Eastern Pennsylvania and every traceroute I do, AT&T always seems to pipe data to Chicago before going anywhere else. This seems highly inefficient as NYC, Philly, and DC are far closer - but every bit of data goes to Chicago first. Even when I use Speedtest servers - the best results come from the Towerstream Chicago server instead of ones in NYC or DC.

Anyone else think this is terrible routing? It's always been the norm, though.

Is your APN either "phone" or "pta"? I'm pretty sure "pta" will get the enhanced routing but you must have a 4G data plan.

[Jordan T]

While SoCal to NorCal and back probably adds ~15-20ms, I will typically get ~90ms to Palo Alto site (Fiber Internet Center), and ~100-110ms to LA (Dreamhost).
The sad part is that I'll typically get ~150-200ms to Charter (Monterey Park or San Luis Obispo) most likely due to peering. The sad part is that I'll get better overall throughput on Charter in SoCal/SLOCal than I will out of anything else (application/load).

Correct my math was wrongish, crossing the US and back adds 32ms (since it's a round trip) but still, physical distance to the data center (within the US) is far from the biggest source of latency as you note with your 150-200ms to Charter's servers...

[formercanuck]

Correct. It takes one bad hop/peer to spike that latency.
A good example of this is my work VPN. Direct from SoCal to Texas (many hops) yields me 50ms - 10 of tjat being the 1st.
Using VPN, i hit 130ms... Mostly due to internal routing of traffic from the vpn from Texas to Michigan and back (1st VPN hop is 50ms as well). Direct traffic may be low in latency, but each hop can/does add latency. I will agree that the first hop is the largest on cellular - showing 70ms here. In general, these carriers will set up large gateways on tier 1 backbone sites... Typically Chicago, San Jose,NYC, Dallas, Miami, Washington DC, etc.
i still can't say how many regional gateways there are for at&t's wireless, but i think that they are highly regional. Just as At&t claimed that everyone between NorCal/San Diego and Las'Vegas were in the West market... I suspect that they all go through the same set of gateways in NorCal.

[Jordan T]

A few large data centers makes sense, from a routing and design efficiency standpoint (fewer points of failure!). And I would love to see anything under 100ms consistently here in Montana. And I'm not talking mobile... my AT&T HSPA+ service has similar latency to my cable, and to what DSL around these parts has (100-150ms to major servers)

[fraydog]

Where I live, both AT&T and Verizon route everything through St. Louis. Good thing that's only an hour away.

[i0wnj00]

I'm in Eastern Pennsylvania and every traceroute I do, AT&T always seems to pipe data to Chicago before going anywhere else. This seems highly inefficient as NYC, Philly, and DC are far closer - but every bit of data goes to Chicago first. Even when I use Speedtest servers - the best results come from the Towerstream Chicago server instead of ones in NYC or DC.

Anyone else think this is terrible routing? It's always been the norm, though.

Not really.
It depends on where the peering points are and where AT&T has their gatewav servers for data for a given area. Most likely these peering points have been around since the days of dial up so it's not likely things will have changed much in terms in routing, regardless of what medium used. But let's bet that Chicago is where AT&T is the ILEC so logically that's where they would first pipe the data to and hand it off there to a peering partner who would in turn make the run to servers/hosting providers on the Eastern Seaboard.

[i0wnj00]

Well, SoCal routes through San Jose - go figure.

It depends on the final destination in question.
None of my So Cal and Vermont data goes through San Jose. LAX and IAH/DFW are the peering points to and from. At least for me, LAX is the first and last peering point based in CA before it splits off, but that depends on where the request is heading to.

Going through San Jose was something from 15 years ago when Road Runner of San Diego couldn't afford to route their data through MCI Worldcom (remember that company?) in San Diego and didn't have a well developed internal backbone for linking RR markets to each other.

[GSMinCT]

It flips all over the place. It just has to do with where they have a fast pipe. Also, all Straight Talk traffic goes through Chicago, so it seems that's where their main datacenters are, and they probably route a good chunk of East-of-Mississippi postpaid traffic through there as well.

[formercanuck]

Take a look at NAP sites -
http://en.wikipedia.org/wiki/List_of...s_by_continent
United States
AADS - AT&T NAP Chicago, Illinois

In general, I suspect that for wireless, there are a handful ...My assumptions are: West (San Jose), East (NY, Washigton DC, Miami), North (Chicago) and South (Dallas).