Point this at your own endpoint, set how many concurrent users and for how long, and watch response times as the load runs. Requests are fired from this server, not your browser, so the count you enter is the real concurrency — not capped by the browser's per-site connection limit.
What this measures: the container this page is served from opens the requested number of connections to your target and holds them open for the whole test, so "2000 concurrent users" means 2000 requests genuinely in flight at once — not queued in waves behind a browser's ~6-connection-per-host limit. Because the requests come from the server, real HTTP status codes are visible (see the breakdown below the chart), not just success/failure of the network call. Very large counts (thousands+) are still bounded by your target server's own capacity, your network link, and this container's file-descriptor limit — if you see errors climb, that's usually the target (or the link to it), not this tool queuing.