QPS (queries per second)

QPS, queries per second, is the number of bid requests a bidder or exchange receives or sends each second, and is the primary unit for sizing RTB infrastructure, negotiating exchange connections and measuring the waste in programmatic traffic.

A DSP connected to several large exchanges can see hundreds of thousands of requests per second, most of which it will not bid on. Each one costs network, parsing and model time, so DSPs throttle (ask exchanges for less traffic), filter (drop requests before full parsing) and sample. On 15 April 2026 Amazon Ads donated its Dynamic Traffic Engine to IAB Tech Lab: a file-based framework in which a DSP publishes the bid-request types it wants to prioritise and SSPs poll that file, with the stated goal of cutting "QPS waste" and infrastructure cost (IAB Tech Lab, 15 April 2026). IAB Tech Lab's ARTF makes a stronger claim, that running enrichment agents next to the auction can cut request and response time "up to 80%" (IAB Tech Lab, 13 November 2025).

Capacity planning starts from QPS and the latency budget. Instances needed is roughly peak QPS multiplied by per-request service time, divided by the concurrency each instance sustains, with headroom for the 99th percentile; the formulas are in QPS, latency and bid performance and the calculator at Bidder capacity planner.

Example: a bidder receiving 200,000 QPS with a mean service time of 8 ms needs 1,600 concurrent request slots on average; at 50 slots per instance that is 32 instances before headroom.

Related: timeout, bid request, DSP, SPO.

Source: IAB Tech Lab, OpenRTB 2.6 (tmax); IAB Tech Lab, Dynamic Traffic Engine donation (15 April 2026).

1. IAB Tech Lab, 15 April 2026
2. IAB Tech Lab, 13 November 2025