Sub-microsecond tick-to-store. PTP hardware-clock nanosecond timestamps. Millions of messages per second. 14 C++ microservices with zero-copy shared-memory IPC, io_uring kernel-bypassed I/O, Raft 2-of-3 fault-tolerant clustering, and three-tier WORM immutable storage, built for real-time trading, deep analytics, and MiFID II RTS-25 compliance.
A horizontally scalable tick-data platform built on shared-memory SPSC ring buffers, io_uring kernel-bypassed I/O, and three-tier immutable storage.
6 native exchange protocol parsers (NYSE Pillar, Nasdaq ITCH 5.0, Cboe PITCH, IEX TOPS, IEX DEEP, MEMX Memoir), 33 direct exchange venues, 22 consolidated vendor normalizers, plus Finnhub, OpenBB, WorldMonitor & LSEG RKD adapters via io_uring UDP multicast and WebSocket/REST. Optional qtick-shipper sidecar ships remote feed batches to a central sequencer.
Global sequence numbering with Raft 2-of-3 consensus replication via shared-memory SPSC ring buffers
Three-tier WORM storage: WAL on NVMe, columnar .acol with LZ4, Parquet archival with SHA-256 hash chains
ANSI SQL 2016 with ASOF JOIN, SAMPLE BY, VWAP, OHLC via PGwire, Arrow Flight SQL, and REST/WebSocket. K-methodology vectorized executor delivers 10.6M rows/s GROUP BY and 11.5M rows/s SampleBy throughput.
Complex event processing with 45 alert models, LSTM inference, embedded Python VM, and io_uring TCP fan-out
Everything you need to capture, store, query, and analyze tick data at scale in a single platform.
WAL on NVMe for write durability, columnar .acol files with LZ4 compression and sparse indexes for hot queries, and Parquet for long-term cloud archival. WORM storage with SHA-256 hash chains for tamper-evident audit trails.
SPSC ring buffers in shared memory enable zero-allocation, zero-copy data transfer between all 14 microservices. No serialization overhead on the hot path. Packets flow from kernel to storage without a single heap allocation.
Full ANSI SQL 2016 with window functions (ROW_NUMBER, RANK, LEAD, LAG, PARTITION BY), ASOF JOIN, SAMPLE BY, LATEST ON, time_bucket(), OHLC(), VWAP(), TWAP(), and Python UDFs. Query via PGwire, Arrow Flight SQL, or REST/WebSocket.
6 native exchange protocol parsers (NYSE Pillar, Nasdaq ITCH 5.0, Cboe PITCH, IEX TOPS, IEX DEEP, MEMX Memoir) drive 33 direct exchange venues (NYSE, NASDAQ, CME, LSE, Euronext, HKEX, and 27 more). Plus 22 consolidated vendor normalizers (Bloomberg B-PIPE + SAPI, LSEG TREP, ICE, dxFeed, Databento, Exegy, FactSet, and 15 more) and 9 crypto / alt-data connectors (Binance, Bybit, Coinbase, OKX, Upbit, Finnhub, OpenBB, RKD, WorldMonitor). All 49 sources emit the canonical QTickQuote schema — one query covers every venue and vendor.
Complex event processing engine with 45 built-in alert models. LSTM model training and inference, model registry, SQL UDFs for ML predictions, and an embedded Python VM for custom analytics.
Full Q IPC binary protocol server (qtick-qserver) lets existing KDB+ clients connect and query QuantumTick data using native Q syntax. Transparent query translation. No client code changes are required.
Native SDKs for C++ (coroutine async, zero-copy Arrow), Java (JDBC, virtual threads), .NET (ADO.NET, source generators), and Python (pip install quantumtick, zero-copy polars DataFrames, qt.replay() historical replay, qt.stream() live feed, Jupyter %%qt magic). Plus a universal REST/WebSocket API.
.NET MAUI real-time monitoring GUI with 8 dashboard panels, Prometheus metrics export, Grafana dashboards, gRPC telemetry streaming, thread-level CPU heatmaps, and Raft cluster health visualization.
Runtime-gated HotTier facade with Kove SDM / local mmap / disabled backends. A full L3 order book for the US equities universe is ~12 TB — Kove SDM puts that hot DRAM on the network, accessible over RDMA in ~1 µs. Huge pages + pre-fault + probe-latency measurement are exposed through [store.hot_tier] in TOML and verified by the qtick-kove-bench tool before production rollout.
Built-in Model Context Protocol server exposes 29 tools for AI assistants to query tick data in natural language, manage CEP alert rules, monitor cluster health, run ML experiments, and optimize storage. Connect Claude, GPT, or any MCP-compatible agent directly to live market data with zero custom integration. The natural-language surface lets non-technical staff run audit queries without q, OTQ, or SQL expertise.
Full ANSI window function support: ROW_NUMBER, RANK, DENSE_RANK, LEAD, LAG, FIRST_VALUE, LAST_VALUE, NTH_VALUE, plus aggregate-over-window (SUM, AVG, MIN, MAX, COUNT OVER). Essential for signal research and momentum analysis.
Dedicated replay service streams historical ticks at configurable speed (1x, 100x, max throughput) with the same API as live streaming. Strategy code works unchanged in both backtest and production. Deterministic, repeatable replay.
Reconstruct full limit order book state at any historical nanosecond from ITCH, Pillar, PITCH, and TOPS packet streams. Query midpoint, spread, imbalance, VWAP, and queue position as first-class columns.
Extend SQL with Python functions: CREATE FUNCTION momentum(...) LANGUAGE PYTHON AS $$...$$. Sandboxed subprocess execution with numpy, 30s timeout, and 256MB memory limit. No C++ required.
CRSP-style price adjustment for splits and dividends. Point-in-time index membership tracking (S&P 500 constituents over time). Symbol change resolution chains. Eliminates survivorship bias in backtests.
Dual-timestamp engine stores both exchange time and ingestion time on every tick. Queries default to AS OF ingestion time, making look-ahead bias structurally impossible. Zero ABI impact on the 64-byte tick struct.
Active-active Raft 2-of-3 consensus with <50 ms failover. Per-service Raft groups (sequencer, store, CEP) limit blast radius to a single service. Cross-region async WAL shipping with CRC32C validation, epoch-based split-brain prevention, and cuckoo-filter dedup across sources. Chaos-tested under 1 119 regression tests.
qtick-import reads kdb+ HDB splayed tables, CSV, and Parquet files directly into QuantumTick storage. Shadow-run reconciliation tool proves tick-for-tick equivalence before cutover. Migration timeline: weeks, not years.
Capabilities that set QuantumTick Database apart from traditional tick databases, time-series stores, and cloud data warehouses.
Capture and fan-out paths use Linux io_uring for kernel-bypassed packet processing, achieving 6M+ packets/second with registered buffers and zero system calls on the hot path. No other tick database offers this at the system level.
SHA-256 hash chains on every write create tamper-evident, append-only WORM storage that satisfy the strictest regulatory retention requirements. Combined with ACID-compliant ingestion in strict, relaxed, or best-effort modes.
Every thread pins to a specific CPU, NUMA node, and cache-line alignment. The qtick-numa-plan launcher auto-generates systemd drop-ins that pin feed handlers to ingest nodes and isolate compaction on the mirror node. Huge-page reservation, cross-CCX isolation on AMD Zen, AVX-512 vectorized aggregation (4 B int64/s target). Result: p99 tick-to-store latency within 1.2× of p50 — not 10× like NUMA-naïve systems.
CUDA kernels for rolling VWAP and OHLCV aggregation run at 40+ GB/s per GPU. LibTorch-powered LSTMs for volatility prediction trained and served in-database. Zero-copy Arrow RecordBatches stream straight into pinned host memory, then cudaMemcpyAsync to device. Multi-GPU via NCCL for backfill and training. On boxes without a GPU, identical operators fall back to AVX-512 through qtick-simd.
Sequencer, store, and CEP each run as independent 2-of-3 Raft clusters for zero-data-loss replication. Automatic leader election, <50 ms failover, deterministic global sequence numbering across all feeds, and cuckoo-filter dedup across sources. No external coordination service required.
Run on bare metal, Docker Compose, Kubernetes (Helm charts with Vault + NetworkPolicies for AWS, Azure, GCP, OCI), Linux RPM (RHEL 8/9), Synology NAS under $2 K of hardware (DSM 7.2+ SPK), Windows (MSI), or Android. Multi-site WAL replication with automated failover for disaster recovery. Same 14-service architecture scales from a home lab to a global trading floor. Optional qtick-shipper sidecar unlocks distributed topologies — feed handlers on separate nodes near their venues stream tick batches over UDP/RoCEv2 to a central sequencer.
Vectorized query executor inspired by kdb+/q inner-loop patterns: per-aggregate Struct-of-Arrays state, int-key fast paths, hoisted type dispatch, and grade-up permutation sorts. Delivers 10.6 M rows/s on GROUP BY, 11.5 M rows/s on SampleBy, and 2.7 M rows/s on AsofJoin, verified by a standalone bench binary wired into CI for regression tracking.
Fail-safe TLS 1.3 at the REST boundary (refuses to start without a valid cert or explicit opt-out), HTTP Basic auth with constant-time comparison on the Prometheus metrics endpoint, whitelist input validation with SQL injection regression tests, non-root containers with dropped capabilities, and a CycloneDX SBOM plus Trivy CVE scan on every CI push.
From micro-structure signals to portfolio-level risk. Analytics run inside the database, not outside it.
ASOF JOIN with TOLERANCE, SAMPLE BY with gap-fill, LATEST ON, window functions (ROW_NUMBER, RANK, LEAD, LAG, PARTITION BY), time_bucket(), OHLC(), VWAP(), TWAP(), and Python UDFs. The only tick database where quants can write SELECT symbol, LAG(price,1) OVER (PARTITION BY symbol ORDER BY ts) natively.
Built-in correlation matrices, PCA, volatility surface fitting, Greeks calculations, and yield curve analytics via the qtick-analytics library. VaR/ES computed directly on tick-level data.
All TCA metrics built in: slippage analysis, venue analytics, arrival price vs. execution price, implementation shortfall, and participation rate that has been validated against known ground truth in the test suite.
Full event-driven backtesting engine with configurable fill models, P&L tracking, and parameter grid search. Run multi-year backtests directly against the columnar store without data export.
LSTM model training and inference, model registry with versioned deployments, SQL UDFs for predictions, and HNSW vector index supporting 1M+ vectors with sub-millisecond top-K nearest-neighbor search.
Multi-standard resolver mapping ISIN, CUSIP, SEDOL, FIGI, RIC, OSI, and LEI at ingest. Bi-temporal mapping tables with full corporate-action chain handling (splits, mergers, spin-offs, ticker changes).
ACID operations, audit trails, fine-grained access control, and long-term retention suitable for global regulatory frameworks.
/dev/ptp0) with CLOCK_TAI fallback — immune to NTP slewSO_TIMESTAMPING on every UDP ingress — the nanosecond is stamped before user-space sees the packetfdatasync completes — exposes a wal_durable_sequence watermark so clients can verify durabilityFS_IMMUTABLE_FL cannot be applied the seal rolls back; no silent compliance degradationSee how QuantumTick Database can deliver sub-microsecond tick-to-store latency, MiFID II RTS-25 compliance with nanosecond PTP timestamps, and Raft-replicated fault tolerance for your firm.
Contact Us