← Workload Catalog

Ad-Tech

Impression & click aggregation

Per-campaign impression, click and spend counters aggregated across bid/serve nodes in real time, exact for pacing and capping.

SURPASSmeets the rate with 4.0x headroom AND a byte-identical, lock-free correctness guarantee
ThroughputPASS
200 M/s required → 800 M/s measured on silicon (100 M/s/bank)
4.0× headroom — needs 2 of 8 banks
Correctness under concurrencyPASS
byte-identical across 1/2/4/8 banks (0x17e9fe29b5dc0aad)
'sum' is abelian → provably order-independent, lock-free

Measured 2026-06-21 on ALINX/HamGeek AX7020 (XC7Z020-2CLG484); reproduce with abench 65537 0xDEADBEEF0BADF00D. Throughput is the engine's accumulation rate (the aggregation step); end-to-end depends on your ingest path, which ATOMiK's order-independence frees from global ordering / locks.

Run this on our silicon → sign conditional on your bar

This is you if

  • You run a programmatic exchange, DSP, or SSP.
  • You keep per-campaign / per-creative budget, pacing, and frequency counters.
  • Hot campaigns contend, and stale counts mean overspend or over-serving.

The workload

  • Per-campaign / creative / placement counters: impressions, clicks, conversions, spend.
  • Updated concurrently across many bid and serve nodes.
  • Budget pacing and frequency capping read the counts, so they must be exact.
  • Events arrive out of order across the serving fleet.

Today — serialized behind a lock

Exact per-campaign counts under concurrent serving means a per-campaign lock / CAS — or accepting stale, approximate counts and eating the overspend / over-serve. Hot campaigns contend; the lock caps serve throughput at exactly peak traffic.

With ATOMiK — order-independent, lock-free

Serve nodes feed count deltas into a shared accumulator with no per-campaign lock. The count is order-independent and byte-identical, so pacing and capping decisions use an exact count regardless of which node served first.

  • No per-campaign lock contention — hot campaigns don't serialize the serve path.
  • Exact counts for budget pacing & frequency capping — no overspend from stale reads.
  • Linear scaling with banks; serve throughput holds under traffic spikes.
  • Serve nodes never coordinate — order of impressions is irrelevant.
Want to test a different rate or operator? Open the interactive benchmark tool · Representative archetype, not a named customer. Numbers are measured engine facts; the value is lock-elimination + the byte-identical guarantee.