What is a proxy pool?
Definition
The set of exit addresses a provider can route your requests through. Its depth where you need it matters more than the advertised total.
Every rotating or backconnect product draws its exits from one. Rotation can only pick from what is there, so the shape of the pool decides how a job fails: how many addresses are online in your country right now, and how many separate networks they sit on. The advertised total decides very little.
Why the headline number is weak evidence
A pool size is a vendor claim, and there is no agreed way to count one. A total can mean every address seen over a year, the addresses online this minute, or the share your plan can reach. Those three can be far apart, and a product page rarely says which it is.
Counting from the outside is possible, just expensive. The IEEE Security and Privacy study of residential proxy services by Mi and colleagues, whose measurements are now several years old, sent 62 million requests through five services over four months and observed 6,183,876 distinct exit IPs. For one provider that claimed 30 million addresses, the authors found about 4 million with 16 million probes, and wrote that it was unclear where the gap came from. The same study saw exits thin out during local night hours, because the hosts are home devices that get switched off. A pool is a population that breathes.
Four questions that make the number usable
| Ask | Why it changes the answer |
|---|---|
| Counted over what window? | A year of sightings includes addresses that left months ago |
| How many are online right now? | Concurrent availability is what rotation actually draws from |
| How many in the country I need? | A global total says nothing about one market |
| How many distinct networks? | Many addresses on a few ASNs behave like a few addresses |
The last row is the sharp one. Addresses announced by the same network share a fate, because a defence can classify a whole network in one decision. Ten thousand exits on three networks are three bets, not ten thousand.
Measure the pool your account actually gets
You do not need the provider’s figure. Draw exits through your own gateway, in your own country, and count what comes back. The loop below uses placeholder credentials; swap in yours and the country parameter your provider documents.
# 200 draws through a rotating gateway, one exit address per line
for i in $(seq 200); do
curl -s --max-time 15 -x http://user:pass@gw.example.net:8000 https://api.ipify.org; echo
done > draws.txt
Then group the addresses by network. This script sends them to Team Cymru’s free IP-to-ASN service in one bulk query and prints distinct addresses, distinct /24 prefixes and distinct origin networks. It needs only the Python standard library; we ran it on Python 3.15.0.
import ipaddress, socket, sys
ips = sorted({line.strip() for line in open(sys.argv[1]) if line.strip()})
query = "begin\nverbose\n" + "\n".join(ips) + "\nend\n"
with socket.create_connection(("whois.cymru.com", 43), timeout=30) as s:
s.sendall(query.encode())
reply = b"".join(iter(lambda: s.recv(4096), b"")).decode()
asns = {}
for row in reply.splitlines()[1:]: # first line is a banner
cols = [c.strip() for c in row.split("|")]
if len(cols) >= 7:
asns[cols[1]] = (cols[0], cols[6])
prefixes = {ipaddress.ip_network(ip + "/24", strict=False) for ip in ips if ":" not in ip}
print(f"addresses {len(ips)} /24 prefixes {len(prefixes)} networks {len({a for a, _ in asns.values()})}")
for asn, name in sorted(set(asns.values())):
share = sum(1 for a, _ in asns.values() if a == asn)
print(f" AS{asn:<7} {share:>3} {name}")
To check the script without publishing anyone’s exits, we fed it six public DNS resolver addresses (8.8.8.8, 8.8.4.4, 1.1.1.1, 1.0.0.1, 9.9.9.9 and 149.112.112.112). Its output:
addresses 6 /24 prefixes 6 networks 3
AS13335 2 CLOUDFLARENET - Cloudflare, Inc., US
AS15169 2 GOOGLE - Google LLC, US
AS19281 2 QUAD9-AS-1 - Quad9, US
Six addresses in six different /24 blocks look like six independent exits. They are three networks. That is the gap the headline number hides.
The script opens the socket itself for a reason. Team Cymru’s documentation says to use GNU netcat, not nc, and on macOS the bundled nc returned an empty reply with exit status 0 for the same bulk query.
Reading your own count
Run the draw more than once. If new draws keep returning addresses you have already seen, the distinct count levels off, and that plateau is the pool available to you, in that country, at that hour. Run it again at a different time of day before you trust it.
The network count matters more than the address count when a target starts refusing you. A pool spread across many networks degrades a little at a time. One concentrated on a few loses most of its capacity in a single decision, which is also why rotation settings cannot rescue it. For the buying side of this, the proxy type guide lists what to verify before committing, and the curl guide covers checking an exit address and its geography from the command line.