# Proxies for ISP performance research by US access network

> ISP performance research: compare how sites and CDNs treat cable, fiber and fixed wireless homes by ASN and city, from $0.50/GB. No KYC, crypto.

- URL: https://broadbandproxies.com/use-cases/isp-performance-research/
- Last updated: 2026-10-18
- Publisher: Broadband Proxies (https://broadbandproxies.com)

Cable, fiber and fixed wireless homes reach the web through different networks and address schemes. Sampling each one from the same city shows where CDNs and anti-abuse systems treat them differently.

**Quick answer — How can researchers compare US access networks from the outside?**

ISP performance research compares how websites, CDNs and anti-abuse systems respond to traffic from cable, fiber and fixed wireless homes. **Broadband Proxies** supplies US home IPs by ASN and city, so one study can sample Comcast, AT&T and T-Mobile Home Internet in the same place. Traffic starts at $2.63 for 1 GB and falls to $0.50/GB on 1 TB. There is no KYC, and you pay in crypto.

## Network research at a glance

| Attribute | Value |
| --- | --- |
| Questions it can answer | Which CDN edge answers each network, how often challenge pages appear, whether redirects or page weight differ |
| Access types | Cable, fiber, copper DSL and fixed wireless, chosen through the ASN of each provider |
| One city, three networks | Houston: cable 94.9%, fiber 55.2%, fixed wireless 78.5% of 1,199,820 residential units (FCC, December 31, 2025 data) |
| Example usernames | `USERNAME-country-us-state-tx-city-houston-asn-7922` (Comcast), `-asn-7018` (AT&T), `-asn-21928` (T-Mobile Home Internet) |
| Limits | Timings include the gateway and the home line; results are relative, not a subscriber's own speed |
| Ethics | Public pages, low request rates, no load testing of third-party sites |
| Price | $2.63 for 1 GB, $50 for 100 GB, $500 for 1 TB |

## Why compare cable, fiber and fixed wireless?

The three access types reach the internet in different ways. Cable and fiber homes usually get their own public IPv4 address. Fixed wireless homes usually share one through carrier-grade NAT, as phones do. A site that scores risk per IP can therefore treat a shared fixed wireless address very differently from a cable one.

The groups are large. Comcast reported 28,486,000 residential broadband customers on June 30, 2026. AT&T counted 12,144,000 consumer AT&T Fiber customers on the same date. T-Mobile reported 8,450,000 5G broadband customers on December 31, 2025. All figures come from company filings.

Coverage also varies by city. In Kansas City, fiber reaches 93.5% of residential units; in Phoenix, only 27.2%. In Los Angeles, Starry fixed wireless reaches 28.5% (FCC data as of December 31, 2025).

## What can access-network research measure, and what can it not?

A proxy request travels from your machine to our gateway, then to a home device, then to the target. Response times include all of those hops. They do not equal a subscriber's own latency, but differences between networks seen the same way can still be compared.

The method measures well what the target decides from the IP:

- the CDN edge or region named in the response headers;
- the HTTP status, and whether a challenge or block page appears;
- redirects to a regional site, and the weight of the page served.

It measures poorly anything about the line itself, such as speed tests or packet loss. The home device's own load and Wi-Fi add noise you cannot see. Record a baseline per exit, as in the script below, and report medians over many slow samples.

## Which networks make a balanced US sample?

| Access type | Network (ASN) | City with wide coverage | Username part |
| --- | --- | --- | --- |
| Cable | Comcast (AS7922) | Denver: 98.7% | `state-co-city-denver-asn-7922` |
| Cable | Spectrum, Texas (AS11427) | Dallas: 94.3% | `state-tx-city-dallas-asn-11427` |
| Fiber | Google Fiber (AS16591) | Kansas City: 80.5% | `state-mo-city-kansas_city-asn-16591` |
| Fiber | Verizon Fios (AS701) | New York: 92.6% | `state-ny-city-new_york-asn-701` |
| Fiber and copper | AT&T (AS7018) | Charlotte: 91.1% | `state-nc-city-charlotte-asn-7018` |
| Fixed wireless | T-Mobile Home Internet (AS21928) | Charlotte: 88.2% | `state-nc-city-charlotte-asn-21928` |
| Fixed wireless | Verizon 5G Home (AS22394) | Charlotte: 88.3% | `state-nc-city-charlotte-asn-22394` |

Share of residential units each provider reports to the FCC National Broadband Map, data as of December 31, 2025. AS7018 carries AT&T Fiber homes and legacy copper lines, so an AT&T exit is not always fiber.

## How do I design a low-impact network study?

1. **Write the protocol first** — State the question, the target pages, the request rate and the duration. Prefer your own pages or test objects you control.
2. **Check each target's rules** — Read robots.txt and the terms of use. Drop any site that forbids automated access.
3. **Pick networks from FCC data** — Use the city pages to choose a city where each access type has wide coverage.
4. **Baseline every exit** — Run the script below to time a tiny request and confirm the origin ASN of each exit.
5. **Sample slowly and publish limits** — One request per target per exit, tens of seconds apart, spread over days. Publish the method with the results.

*Baseline one exit per access type (Python, requests)*

```python
# Baseline for a network study: one slow sample per access type in one city.
# It times a tiny request through each exit and records the origin ASN.
import csv
import sys
import time
from datetime import datetime, timezone

import requests

GATEWAY = "gw.broadbandproxies.com"
PORT = 7777
PASSWORD = "PASSWORD"
CITY = "country-us-state-nc-city-charlotte"
NETWORKS = {"cable": 11426, "fiber_or_copper": 7018, "fixed_wireless": 21928}
PAUSE_SECONDS = 30

writer = csv.writer(sys.stdout)
writer.writerow(["utc_time", "access_type", "asked_asn", "exit_ip", "origin_asns", "ipify_ms"])

for access_type, asn in NETWORKS.items():
    user = "USERNAME-%s-asn-%d" % (CITY, asn)
    proxy = "http://%s:%s@%s:%d" % (user, PASSWORD, GATEWAY, PORT)
    started = time.perf_counter()
    reply = requests.get("https://api.ipify.org?format=json",
                         proxies={"http": proxy, "https": proxy}, timeout=30)
    elapsed_ms = round((time.perf_counter() - started) * 1000)
    ip = reply.json()["ip"]
    ripe = requests.get("https://stat.ripe.net/data/prefix-overview/data.json",
                        params={"resource": ip, "sourceapp": "isp-research"},
                        timeout=30).json()
    origins = " ".join("AS%d" % a["asn"] for a in ripe["data"].get("asns", []))
    writer.writerow([datetime.now(timezone.utc).isoformat(timespec="seconds"),
                     access_type, asn, ip, origins, elapsed_ms])
    time.sleep(PAUSE_SECONDS)
```

## What are the ethical limits of this research?

Measure public pages and your own test objects, at rates no site would notice. Never load test a third-party site or CDN: our [Acceptable Use Policy](https://broadbandproxies.com/acceptable-use/) forbids floods and scans of systems without the owner's permission.

When a challenge page appears, record it as a result and move on. Do not solve it, by hand or with a tool; automatic CAPTCHA solving is forbidden. Do not sign in anywhere, and keep personal data out of what you store. Each exit is a home device on a real line, which is one more reason to keep each sample small.

## How much traffic does a network study use?

A slow panel of small requests uses little traffic; full page loads use more. Try 1 GB ($2.63) to validate the protocol on three networks. Long panels across many cities fit 100 GB ($50) or 500 GB ($250). Traffic never expires between waves.

| Plan | Quantity | Price | Price per unit |
| --- | --- | --- | --- |
| Test | 1 GB | $2.63 | $2.63/GB |
| Business | 100 GB | $50 | $0.50/GB |
| Scale | 500 GB | $250 | $0.50/GB |

In every plan: Protocols: HTTP, HTTPS, SOCKS5 · Targeting: ASN (ISP), state, city and ZIP in the US; country elsewhere; city availability varies · Sessions: Rotating or sticky (up to 30 minutes) · Authentication: Username and password, or an IP allowlist (up to 10 IPs) · Traffic expiry: Never expires

**Before you pay**

- Minimum top-up $25. Top up once in crypto, then buy any plan from your balance.
- Unused balance stays on your account and never expires. When a purchase costs less than the minimum top-up, the part added only to reach the minimum is refundable on request to support within 30 days of the payment, in USDT or USDC, as long as it has not been spent. The rest of the balance is not paid back, except where the law requires it.
- Credited once your payment is checked on the blockchain. Typical network confirmation time: BTC ≈ 20 min · USDT TRC-20 ≈ 1 min · USDT ERC-20 ≈ 5 min · USDC ERC-20 ≈ 5 min · ETH ≈ 5 min · SOL ≈ 1 min · LTC ≈ 15 min

## ISP performance research: questions

### Is latency through a residential proxy the same as a subscriber's latency?

No. The path adds our gateway and the home device's own line, so absolute numbers run higher than a subscriber's. What you can compare is the difference between networks measured the same way, at the same time, from the same city. Record a baseline per exit and report medians, not single values.

### Can I load test a CDN or a third-party site?

No. Load testing a system you do not own, or that has not agreed to it, is forbidden by our Acceptable Use Policy. Load tests also say nothing about access networks. Keep to a few slow samples per target, or measure test objects that you host yourself on the CDN.

### Can I tell fiber homes from copper lines on AT&T?

Not from the ASN alone. AS7018 carries AT&T Fiber homes, legacy copper lines and many business circuits. Pair it with a city where AT&T has built fiber widely, and treat the result as a mix. Google Fiber (AS16591) or Verizon Fios (AS701) give a cleaner fiber sample.

### Can I cite the FCC figures shown on this site?

Yes. Cite the FCC National Broadband Map, data as of December 31, 2025, which is the source of every coverage figure here. Each [city page](https://broadbandproxies.com/cities/) lists the providers with their share of residential units. Subscriber counts come from company filings, each dated on its [ISP page](https://broadbandproxies.com/isps/).

### How many samples does a study need?

Enough to see a pattern over noise, spread across days and hours. One sample per network says little, because home devices come and go. ASN availability varies by network, city and time of day; the checkout shows what is online, and no IP count per ISP is published. Plan waves on several days, and note when a network had no exit online.

### Can a university or lab buy without documents?

Yes. Broadband Proxies works without KYC, so a lab uploads no ID card, selfie or papers. Crypto is the only payment method, with a minimum top-up of $25, which may not fit every procurement process. Account-number login; email is optional. Prepaid traffic never expires, so a study can run in several waves.

---

Broadband Proxies: Broadband Proxies is a US residential proxy provider with ISP (ASN) and city targeting, paid in crypto with no ID upload: pick Comcast, Spectrum, AT&T, Verizon Fios or 30 other US networks, plus state, city and ZIP, billed per GB.