Mobile Web Performance on Real US Carriers

Lab performance tests slow a fast connection down to imitate a phone. That is useful, but it is an imitation. Real carrier paths add variable latency, carrier DNS, NAT behaviour and occasional slow starts that a synthetic profile does not capture, and field data tells you something is slow without telling you why. GlobalProxies lines let performance engineers run repeatable tests over real AT&T, T-Mobile and Verizon connections in eight US metros. This page explains what a proxied carrier test measures, how to set it up without distorting the numbers, and how to use it alongside lab and field data.

What a carrier path adds to the picture

Three things mostly. First, latency: round trips over a mobile network are longer and less predictable than on office fiber, and every extra connection a page opens pays that cost. Pages that load many third-party origins suffer most. Second, the carrier's own infrastructure: carrier DNS resolvers, carrier-grade NAT and the route from the carrier's gateway to your CDN's nearest edge. A CDN that maps a Houston carrier gateway to a distant edge will show up here and nowhere else. Third, variability. A carrier connection on a busy evening behaves differently from the same connection at dawn, which is exactly what your users experience.

What a proxied test does not reproduce is the phone's own radio and CPU. The test machine still renders the page. Combine the carrier path with CPU slowdown or real mid-range devices if you want both halves of the mobile experience.

Setting up a fair test

Point your testing browser at the line with its HTTP or SOCKS5 port. Chrome and its automation tools accept a proxy server argument, and most performance tooling built on them can pass one through. Turn off the tool's simulated network profile, because the carrier is now providing the real conditions and a simulated profile on top would double-count. Keep CPU slowdown if you want mid-range device timing. Run each URL several times and report the median, discarding the first run if you are testing a warm cache scenario.

Keep the line on sticky during a test series, so the numbers reflect one path. Rotate between series if you want to sample different addresses from the carrier's regional pool; each rotation re-attaches the device to the network, which occasionally lands on a different gateway and shows you another route to your edge.

• Simulated network profile: off, because the carrier supplies real conditions.
• CPU slowdown: on if you want mid-range device timing, since rendering still happens on the test machine.
• Runs per URL: several, reporting the median, because carrier timing varies from run to run.
• Session mode: sticky for each series, so one route stays under test.

Choosing metros, carriers and 4G or 5G

Test from where your users are, and from at least two carriers. A site that is quick on T-Mobile in Los Angeles and slow on AT&T in Miami has a routing or CDN mapping issue worth raising with the provider. Phoenix is strong on Verizon, Miami and North Carolina lean AT&T, and Houston has the widest choice of hardware. Boston runs 4G LTE only.

Choose 4G lines to represent a typical or conservative user, since 4G connections usually run between 20 and 45 Mbps, and 5G lines for users on newer phones and networks, often at 50 Mbps or more depending on signal. Many teams keep one 4G and one 5G line in their main metro and report both, because a page that only feels fast on 5G is not fast for everyone.

Reading the results

Compare carrier results with your lab numbers and your field data from real users. If the carrier test matches field data closely, you have a reliable reproduction and can test fixes before shipping them. If the carrier test is much faster than field data, the gap is probably device CPU or older networks. If it is much slower than lab tests, look at connection setup: DNS lookups, TLS handshakes and the number of origins. Waterfall views over a carrier path make third-party cost painfully clear.

Each line includes 15 GB of data per day. Performance runs repeat full page loads with cache disabled, so heavy pages add up; plan series sizes accordingly. A 5G line is $15 per day or $75 per week, a sensible budget for a focused optimization sprint.

Setting up a Mobile web performance testing proxy on GlobalProxies

1. Order lines in your main user metros, mixing carriers and at least one 4G and one 5G line.
2. Copy the host, HTTP or SOCKS5 port, username and password for each line.
3. Launch your test browser or tool with the proxy server setting pointed at the line.
4. Disable the simulated network profile and choose your CPU slowdown setting.
5. Run each URL several times on sticky and record the median.
6. Rotate between series and compare results with lab and field data.

Mobile web performance testing proxy questions