What is Concurrent Connections?
Concurrent connections are the number of proxy connections open at the same moment, counted across your workers rather than over a period. It is a capacity limit, separate from requests per second and from bandwidth. Ten connections can carry thousands of requests an hour when each is short, or almost none when each holds a slow page open.
What counts as one connection depends on the protocol. HTTP/1.1 with keep-alive holds a TCP socket open and reuses it for sequential requests, so a pool of fifty sockets is fifty concurrent connections however fast you cycle requests through them. HTTP/2 multiplexes many streams over a single connection, which makes a request-count limit and a connection-count limit diverge sharply. A SOCKS5 tunnel is one connection per destination socket, so a browser loading a page with sixty assets opens dozens at once through a single proxy endpoint.
Limits stack, and the lowest one wins. The provider enforces a cap per account or per endpoint and answers anything above it with a refused connection or a queue. Your own machine runs out first more often than people expect, between the open file descriptor limit, the local ephemeral port range and the connection pool size configured in your HTTP client. On the far side, the target's WAF counts sockets arriving from one address and reads a burst of parallel connections as an automation signal.
On mobile proxies, concurrency is shared radio capacity. A single 4G line has one uplink, and every open connection competes for the same air interface, so pushing sixty parallel requests through one modem raises latency for all of them and drops some outright. Spreading the same workload across more endpoints keeps per-request latency sane. Tuning is a loop rather than a calculation. Raise concurrency until median latency starts climbing or errors appear, then step back a notch and hold there.
Concurrency is a billing lever at many providers, sold as thread limits or per-port pricing, which pushes teams into buying capacity they use in bursts. Where billing runs on traffic instead, another worker costs whatever traffic it moves. PROXIES.SX charges for bandwidth alone, $4/GB dropping to $2.40/GB at volume with endpoints and rotation free, so concurrency becomes a question of what the network and the target tolerate rather than what the plan permits. Either way, find your ceiling by measurement before you size a plan around it.
Where you meet it
You tune concurrency the first time a crawl runs too slowly and the first time it starts collecting blocks, which tend to be the same week. The setting lives in your scraper as a worker or thread count: Scrapy's concurrent request settings, a semaphore around an async client, the pool size in your HTTP library. Watch median latency and error rate together, because throughput that rises while latency doubles is not progress.
Common questions
How many concurrent connections do I need for scraping?
Start low, around five to ten per endpoint, then raise it while watching median latency and error rate. The right number depends on how fast the target responds and how tolerant it is of parallel sockets from one address. Adding workers past the point where latency climbs earns you more errors, not more pages.
What is the difference between concurrent connections and threads?
Threads are a client-side execution model; connections are open sockets. One thread can hold several connections, and an async client on a single thread can hold hundreds. Provider limits count sockets reaching their gateway, so tune the connection pool size rather than assuming your thread count controls how many connections exist.
Does higher concurrency use more bandwidth?
It spends the same bandwidth faster. Total traffic is set by how many pages you fetch and how heavy they are, so concurrency changes the rate rather than the amount. It can raise waste indirectly, since requests that get blocked and retried still transfer bytes you pay for.
Related terms
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.