One curl command splits the delay into stages, so you can tell a slow route to the proxy from a slow exit or a slow site.
| Item | What it measures | If it is large, suspect |
|---|---|---|
| time_namelookup | Time until the name was resolved | Your DNS, or the proxy host name lookup |
| time_connect | Time until the TCP connection to the remote host or the proxy was made | The path between you and the proxy |
| time_appconnect | Time until the SSL handshake to the remote host was done | A new connection every time, or a slow tunnel through the exit |
| time_starttransfer | Time until the first byte came back, including server processing | The exit, the site itself, or a slow first request |
| time_total | Total time of the whole request | Large pages or slow download after the first byte |
Definitions follow the curl manual page, read on 11 November 2026; the "suspect" column is our reading of them, not a curl statement.
Ping uses ICMP. RFC 1928 says SOCKS5 does not forward ICMP messages, and an HTTP proxy only tunnels the connection you ask for with CONNECT host:port. So a good ping to the proxy address says little about page loading. Time a real request instead.
Run the first command in the box below without -x. If the site is slow with no proxy, the proxy is not the cause.
Run the second command with HOST, PORT, LOGIN and PASSWORD from My assets.
Large time_connect: the path to the proxy. Small time_connect but a late time_starttransfer: the exit or the site.
Run it at least five times and also time https://ipinfo.io/json. One slow run is not a result; a site that is slow through the proxy but fast on ipinfo.io is a site problem.
Slow route to the proxy: see the table in the next section and the mainland China note. Slow exit: pick an IP country closer to the target site. Slow only in the browser: check extensions and the system proxy or PAC settings.
FMT="dns=%{time_namelookup} connect=%{time_connect} tls=%{time_appconnect} first_byte=%{time_starttransfer} total=%{time_total}\n"
# 1) directly
curl -s -o /dev/null -w "$FMT" https://www.example.com/
# 2) through the proxy
curl -s -o /dev/null -x http://LOGIN:PASSWORD@HOST:PORT -w "$FMT" https://www.example.com/Example only: replace https://www.example.com/ with your target and the placeholders with the values from My assets. On Windows use curl.exe and write NUL instead of /dev/null.
Check Is it large for other sites too, and do you connect from mainland China?
Fix From mainland China, open Proxy302 Fast Connect (free, Windows and Mac), turn on One-click boost (一键加速) and keep it open while you use the proxy. Elsewhere, try another network such as a phone hotspot to see whether your line is the cause.
Check Is the target site far from the exit country, and is it slow without the proxy too?
Fix Choose an IP country closer to the target. Per-IP prices are the same in every country, so a closer country costs no more.
Check Does your tool open a new connection for every request?
Fix A new connection pays for connect and the handshake each time. Reuse one connection (a session) where your job allows it.
Check Extensions, a PAC script, or another proxy client in the chain?
Fix Test in a clean browser profile with extensions off, and check the system proxy settings.
In our October 2026 connectivity report the median time to first response across seven countries was 1.4 s, with country medians from 0.7 s (Singapore) to 2.6 s (Brazil), measured from one probe in Asia. A test of the rotating datacenter line on 10 November 2026 gave about 0.8 to 1 s per HTTPS request from Asia. Expect a first response of around a second through a proxy, not a few milliseconds, and measure your own route.
Ping uses ICMP, which SOCKS5 does not forward and HTTP proxies do not carry. Time a real request with curl instead.
We do not promise a figure. Our published reference points are in the section above; a first response in the range of a second or two from a nearby region is what our own checks saw.
Run the timing in the steps above, directly and through the proxy, five times. Then compare time_connect with time_starttransfer.
No. It is for people who connect to the proxy from mainland China; elsewhere use the details in My assets directly.
A static ISP IP costs US$0.50 for 1 day and US$7.99 for 30 days, the same price in every country. There is no free trial; the smallest purchase is one IP for one day.
When a proxy won't connect or is slow, check six things first: address, port and type, login and password, one proxy app only, the antidetect browser port, the exit IP.
Citable open data from our hourly checks: 38 platforms through residential IPs in 8 countries, with response times, exit networks, methods and limits.
Check the exit IP, country and ISP your proxy really gives you in 30 seconds — and why IP databases disagree on the city. With our own test of three lookup services.
The ERR_PROXY_CONNECTION_FAILED error appears in Chrome and Edge when the browser cannot establish a connection with a proxy server.
Fixed static ISP proxies: a residential-grade IP that stays the same for the whole term. For social accounts, stores and AI tools. HTTP and SOCKS5.
View options and pricingOfficial documentation · Confirmed on:
Read on 11 November 2026: -x/--proxy accepts socks4://, socks4a://, socks5:// and socks5h://, and a user and password in the proxy string are URL-decoded (%40 for @); --write-out defines time_namelookup, time_connect (TCP connect to the remote host or proxy), time_appconnect, time_starttransfer and time_total.
Official documentation · Confirmed on:
Read on 11 November 2026: lists the commands CONNECT, BIND and UDP ASSOCIATE and the authentication methods no authentication, GSSAPI and username/password; the introduction says the protocol does not forward ICMP messages.
Official documentation · Confirmed on:
Read on 11 November 2026: CONNECT asks a proxy to open a tunnel to a destination; the request target is the destination host and port (CONNECT host:port).
Product observation · Confirmed on:
Our published residential connectivity report (October 2026, one probe in Asia, 6 of 8 hourly rounds, one exit per country per round): the median time to first response across seven countries was 1.4 s, with country medians from 0.7 s (Singapore) to 2.6 s (Brazil). The product facts file records a test of the rotating datacenter line on 10 November 2026 at about 0.8 to 1 s per HTTPS request from Asia.
Product observation · Confirmed on:
Product facts file of the site (values checked against the console, the backend and the price table on 18-19 October 2026, protocol tests of 4 November 2026 and the rotating datacenter line on 10 November 2026): products, prices in US$ (1 PTC = US$1), lease terms, protocols and ports, countries, no free trial.
Create an account, top up once, then pick rotating residential, static ISP or datacenter IPs and have one running in your browser within minutes.
