// research lab · time series infrastructure

Ingestbillions.Query fast.

Epoch ingests 1.24 billion timestamped data points per second and returns query results in under 1ms — before the next tick arrives.

Run a Benchmark →Read the Docs

1.24B

points/sec ingest

sustained write throughput

0.31ms

query p50

on 847M row dataset

11.2×

compression

vs raw float64 storage

2.1B

unique series

zero index memory bloat

LIVE

Ingestion Counter

2,84,73,92,104

points ingested total

+12,47,382 pts/sec

Data Stream — Real-time

// infrastructure failure modes

What breaks at scale.

Three real failure patterns. Each one resolved — watch the metrics heal as you scroll.

Single Node — FAILURE⚡

Cardinality Explosion

Prometheus runs out of heap at 10M series. Scrape intervals slip. Alerts stop firing. The dashboard goes dark.

14.2M

unique series → OOM

Single Node — RESOLVED

Cardinality Explosion

Epoch's bitmap inverted index scales horizontally. 2 billion unique series, same sub-ms tag lookup. No heap pressure.

2.1B

series, zero index bloat

Distributed Cluster — FAILURE💾

Disk I/O Bottleneck

Five years of tick data means 40GB/s read bursts during backtests. Postgres chokes. The quant team waits.

94%

disk utilization → stall

Distributed Cluster — RESOLVED

Disk I/O Bottleneck

Gorilla XOR encoding collapses tick data to 9% of raw size. Your 40GB/s burst becomes 3.6GB/s. Disks breathe again.

0.4 GB/s

effective I/O via compression

Global Replication — FAILURE🌐

Query Timeout

IoT platform. 80,000 sensors. Each dashboard query fans out to 80K series. TimescaleDB hits 47s. Users leave.

47s

p99 latency → timeout

Global Replication — RESOLVED

Query Timeout

Vectorized SIMD execution across columnar segments. 80,000 sensor aggregation in under 1ms. Dashboards feel instant.

0.89ms

p99 across 80K series

Single Node

→

Distributed Cluster

→

Global Replication

// infrastructure crisis averted

Your data is melting. The fix takes 5 minutes.

Run a Benchmark →

// who breaks first

Built for the engineers on call at 2 AM.

The pain

Prometheus cardinality wall

14.2M series → OOM

The fix

Epoch ingests 2.1B unique series with flat memory.

150×

more series, same RAM

epoch.sql

SELECT count(*) FROM metrics
WHERE __name__ = 'http_requests_total'
  AND time > now() - 1h
GROUP BY instance, job;
-- returned 2.1B rows in 0.31ms

▶execute_

// independent benchmark results — 2026-02-24

The numbers don't negotiate.

Same hardware. Same dataset — 5 years of S&P 500 tick data, 847M rows. Cold cache. No query result caching.

Database	Ingest Rate	Query p50	Query p99	Compression	Cardinality
Epoch← fastest	1.24B pts/s	0.31ms	0.89ms	11.2×	2.1B series
InfluxDB	38M pts/s	4.2ms	31ms	3.1×	~10M series
TimescaleDB	22M pts/s	8.7ms	47s	2.8×	~5M series
ClickHouse	180M pts/s	1.8ms	12ms	6.4×	~100M series

* Benchmark methodology: full report → Hardware: 32-core AMD EPYC, 256GB DDR5, NVMe RAID-0.

// no account required

Paste a query.
See the result.

The playground is preloaded with 5 years of S&P 500 tick data — 847 million rows. Write any query. The answer arrives before you expect it.

Run a Benchmark →Read the Docs

No sign-up·Preloaded dataset·Results in <1ms

Ingestbillions.Query fast.

What breaks at scale.

Cardinality Explosion

Cardinality Explosion

Disk I/O Bottleneck

Disk I/O Bottleneck

Query Timeout

Query Timeout

Built for the engineers on call at 2 AM.

The numbers don't negotiate.

Paste a query.See the result.

Paste a query.
See the result.