Network

What is Latency?

Latency is the delay between sending a request and receiving the first byte back, measured as round-trip time in milliseconds. Through a proxy it spans the client's hop out, the proxy's hop to the target and the return leg. A mobile exit adds more delay than a datacenter one, since radio scheduling and carrier NAT sit in the path.

Be precise about what you measure. An ICMP ping reports round-trip time to a host that may deprioritise or rate-limit those packets, so the number can be flattering or meaningless. Time to first byte over the real protocol includes the proxy's own handling, name resolution and the target's processing. Those are different quantities. Measure with the same client, protocol and endpoint you use for production work, take a distribution rather than a single sample, and record the median alongside the slow tail.

The milliseconds come from several places. Distance sets a hard floor, since light in fibre covers roughly two hundred kilometres per millisecond and real paths are longer than the great-circle line, so a transatlantic round trip cannot drop below the tens of milliseconds. Cellular adds radio scheduling, and a modem that has dropped to an idle radio state pays a wake-up penalty on the first packet after a pause. Handshakes multiply everything: a TCP connection costs one round trip, a TLS 1.3 handshake one more, older TLS two. Reusing connections removes that cost entirely.

Latency and throughput are separate problems. A high round-trip time does not limit how much data you can move; it limits how many sequential exchanges you can complete per second. A crawler doing four round trips per page at 200 ms will finish a little over one page per second per worker, no matter how much capacity the link has. The fix is concurrency and keep-alive connections rather than hunting for a lower ping, and HTTP/2 multiplexing helps on targets that support it.

Variance breaks automation more often than the average does. Tower congestion, a rotation event or a carrier middlebox produces occasional multi-second responses inside an otherwise steady stream, and a timeout set near the median converts those into failures and retry storms. Set timeouts from the 95th or 99th percentile you observe on your own traffic, back off on retry, and treat a rising tail as a signal to check the exit rather than to raise concurrency.

Where you meet it

Latency turns into a real decision when you size timeouts for a crawler or compare two providers whose sales pages both quote a ping figure. Run your own measurement through the client that does the work, watch the slow tail over an hour rather than a minute, and remember that a lower median buys nothing if a sequential pipeline is the thing holding you back.

Common questions

What is a good latency for a mobile proxy?

Judge it against the same route without a proxy rather than against a figure on a sales page. An exit in your own country will beat one on another continent, since distance sets a floor you cannot argue with. Consistency is worth more than a low median, because a steady response time beats a lower average punctuated by multi-second spikes.

Why is a mobile proxy slower than a datacenter proxy?

Cellular traffic waits for radio scheduling, shares cell capacity with real subscribers, and passes through carrier NAT before reaching the internet. That overhead buys the thing datacenter addresses cannot provide, which is an exit that looks like an ordinary phone on a carrier network.

Does lower latency mean faster scraping?

Only for work that runs one request after another. Once you add concurrent workers and reuse connections, total throughput depends on how many requests are in flight and how the target responds, not on the round-trip time of any single one.

Related terms

Bandwidth
Bandwidth carries two meanings in proxy work. One is link capacity, measured in bits per second. The other is transferred volume, measured in gigabytes, and that is what proxy providers bill. Metering counts every byte crossing the proxy in both directions, including protocol overhead, headers, redirects, images and requests that end in a block.
IP Address
An IP address is the numeric identifier a device presents on an IP network so packets can be routed to and from it. IPv4 writes 32 bits as four decimal octets, such as 203.0.113.7. IPv6 writes 128 bits as hexadecimal groups, such as 2001:db8::1. Every request a server logs is tied to one.
IPv4
IPv4 is version 4 of the Internet Protocol, addressing hosts with 32 bits written as four dot-separated octets, which gives roughly 4.3 billion addresses. The top-level pool ran dry in 2011, pushing providers toward private addressing, carrier-grade NAT and a paid transfer market. Nearly all proxy traffic still runs over IPv4 because that is what target sites accept.
IPv6
IPv6 is version 6 of the Internet Protocol, addressing hosts with 128 bits written as eight hexadecimal groups, such as 2001:db8::1. The space is large enough that carriers assign each subscriber a whole subnet, usually a /64 or wider, so reputation systems score prefixes instead of single addresses. Adoption is broad but far from universal.
CGNAT
Carrier-grade NAT is address translation performed by an ISP rather than in the customer's router, putting many subscribers behind one public IPv4 address. Mobile networks depend on it. Each customer holds a private or shared-range address, usually from 100.64.0.0/10, and the carrier rewrites the source address and port on every outbound flow.
NAT
Network Address Translation rewrites the address and port fields of packets as they cross a router, letting many private hosts share one public IP. The router keeps a translation table, mapping each outbound flow to a free public port and reversing the rewrite on replies. Home routers do this for a household; carriers do it for whole regions.
ASN
An Autonomous System Number identifies a network that announces its own IP address ranges to the rest of the internet through BGP. Every routable address sits inside a prefix announced by some AS, so the ASN behind an IP tells you who operates it - a mobile carrier, a hosting provider, a home ISP or a corporate network.

Real 4G/5G mobile and residential IPs

PROXIES.SX runs carrier IPs in 100+ countries with HTTP and SOCKS5 on every endpoint. $4/GB down to $2.40/GB at volume, free endpoints and rotation, and your GB never expire.