What is the PriceONN price engine?

PriceONN's price engine is the read-path that serves last-tick and 1-minute-candle queries to our charts and public price API. It uses an in-memory hot-path cache fronting a sharded persistent store, so every public chart read is a sub-millisecond cache hit.

What does p97 < 0.12 ms mean?

97 of every 100 queries finish faster than 0.12 milliseconds (120 microseconds). The remaining 3% land in the long tail (rare slow events such as cache cold-starts), but never affect the typical user experience.

How is the latency measured?

Every minute, the engine-benchmark job picks random active symbols and times exactly 1,000 indexed read-path queries with a monotonic, hardware-backed nanosecond clock. The first 50 samples are discarded as warm-up; aggregated p50/p97/p99/mean/min/max are persisted for 35 days.

Yes. Three open JSON endpoints under /api/feed/ return live and historical benchmark data. CORS is open, no authentication required, and the dataset is published under CC BY 4.0.

Why is this faster than other market-data APIs?

Public quote APIs (Bloomberg, Reuters, TradingView, Yahoo) include network round-trip plus their own server-side cost - typically 50–250 ms. Our number is the engine-internal read-path only; network adds the same 20–200 ms on top for any remote call.

● LIVE PROOF

Engine Speed Test

Independent live proof - sub-millisecond query latency on the PriceONN price engine, refreshed every 5 seconds.

✓ TARGET MET

Last-tick query · p97

- ms

p50 - · p99 -

1m candle · p97

- ms

24h average · p97

- ms

rolling 24-hour mean

Symbols indexed

in last sample

Verify it yourself open · no auth

Three public JSON endpoints back this page. CORS open, CC BY 4.0 dataset.

Web availability · independently monitored via UptimeRobot ↗

How we measure

1,000 queries per minute
Nanosecond clock (monotonic, hardware-backed)
First 50 samples discarded as warm-up
35 days public history retained

Full methodology →

Who is this for?

Fintech apps · real-time price widgets
Algo trading · low-latency price feed
Content sites · minute-fresh quotes

Engine v4.11 · 90+ symbols (forex, CFD, crypto, indices, commodities) · multi-source aggregated.

A human eye

blinks once every

100^ms

Roughly the time it takes you to blink once - far longer than a single network round-trip.

In that same blink

PriceONN serves

833^tick

100 ms ÷ 0.18 ms p97 = 555+ indexed price queries served before you finish blinking. Median (p50): 250,000 queries.

priceonn-engine ~ /tick-stream

ticks/s 0 served 0

Live latency distribution

On-demand 200-sample benchmark, refreshed every 5s. p50 / p97 / p99 marked.

Cumulative distribution

% of queries returning under threshold X - closer to step = lower variance.

p97 timeline

Per-minute p97 over time. Yellow band 2-4 ms · red >4 ms.

🏆 Performance Report - Live 24h

Honest, instrumented, public-verifiable. No marketing claims - just measurements.

p50 (median)

- µs

100% under 100µs

p97 (typical)

- ms

- vs 0.18ms

p99 (worst minute)

- ms

spike rare

Sample size

minutes

SLA Transparency

p50 < 100µs primary

p97 < 0.18ms committed

p97 < 0.12ms aspirational

Industry Position - vs Public Financial Data APIs

Honest disclaimer: Server-side query latency only. End-to-end network adds 50-200 ms RTT depending on consumer location. Tier-1 HFT firms operate sub-microsecond on FPGA/kernel-bypass - that is a different market.

Where this matters - and where it does not

✓

Algorithmic trading bots

Backtest 50-100× faster than competitors.

✓

AI financial assistants (LLM tool calls)

Engine consumes <1% of typical LLM tool budget.

✓

Real-time dashboards & terminals

Sub-millisecond engine = "instant" UX feel.

Tier-1 HFT (FPGA territory)

Different market. Citadel-tier needs custom hardware.

Auto-refreshes every 30 seconds. Backed by /api/feed/performance-report.php · 7d · 30d

Industry comparison

Server-side query latency only. Network adds ~20-200ms on top for any remote API.

Source	Typical latency	Scale
PriceONN engine	- ms
Typical in-house quote DB ¹	1–5 ms
TradingView quote feed ²	~120 ms
Bloomberg Terminal API ³	~50 ms (network)
Refinitiv Eikon / LSEG ⁴	~80 ms
Yahoo Finance public API ⁵	~250 ms

Sources: vendor docs and public benchmarks. Numbers represent median round-trip latency for last-tick quote retrieval and may vary by symbol, region and subscription tier. PriceONN figure is server-side engine latency only - network round-trip adds ~20-200 ms on top.

What do p50, p97, p99 mean?

Percentiles describe how a query performs across many samples - better than averages, which hide the slow ones.

p50 · Median

Half of all queries finish faster than this.

Primary performance guarantee. PriceONN engine consistently returns sub-100µs medians - independently verifiable on this page right now.

p97

97 of every 100 queries finish faster than this.

Tail-latency budget. Typical < 0.18 ms. Brief cache-miss spikes (Mongo checkpoint, plan re-eval) are real and visible in the live histogram. We do not hide them.

p99

99 of every 100 queries finish faster than this.

The "tail" - only 1% of queries land beyond this. Used to bound rare slow events.

Methodology & transparency

Every minute, an automated benchmark fires 1,000 indexed last-tick and 1-minute-candle queries against the live PriceONN price engine - the exact same code path our public price API uses. Latency is measured with a monotonic nanosecond clock; the first 50 samples are discarded as warm-up. Each minute's aggregated statistics (p50, p97, p99, mean, min, max) are persisted to a public benchmark dataset and retained for 35 days. The price-feed pipeline is read-only here - measurement never interferes with live tick ingestion.

Anyone can verify these numbers using the public JSON endpoints below. CORS is open, no authentication required.

/api/feed/benchmark.php history?range=24h range=7d range=30d

Frequently asked

Why can't I test the underlying price endpoint directly?

The /api/feed/benchmark.* endpoints are public for verification. Production price endpoints (last-tick, 1m-candle) are gated behind API authentication - anti-abuse, rate-limit-per-license, and customer SLA tracking. The benchmark endpoint runs the same code path on the same infrastructure - its measurement is representative of authenticated production behavior. Standard B2B practice (Polygon, Coinbase, Bloomberg all do this).

Why do I see occasional 0.5-2 ms spikes in the histogram?

Sub-millisecond infrastructure has irreducible tail noise: MongoDB WiredTiger checkpoints, query plan re-evaluation, OS scheduling, brief cache invalidation. We do not hide them. p50 < 100µs is the consistent floor; p97 < 0.18 ms typical; p99 may briefly hit ~0.5 ms during a checkpoint window. Filtering these outliers would be statistically dishonest. Looking at history?range=30d, average steady-state SLA is >95%.

What is the real-world end-to-end latency for a remote API consumer?

Engine latency is what's measured here: 65µs median. End-to-end depends on network distance - typically 50-200 ms RTT depending on the consumer's region. The engine moat shows when our latency consumes <1% of the total round-trip, while competitors (TradingView ~120 ms server, Yahoo ~250 ms server) consume the entire latency budget on the server side alone - before network even starts. A trader 50 ms away from us still gets a fresh tick faster than a co-located TradingView consumer.