# HFTKernel

> HFTKernel provides predictable Linux execution for latency-critical trading systems across cloud and bare-metal infrastructure.

## Four production problems HFTKernel solves

1. Keep trading cores free from operating-system noise by moving IRQ, softIRQ, RCU callbacks, timers, and background work away from latency-critical CPUs.
2. Reduce p99.9, p99.99, and p99.999 tail latency across market data, strategy, risk, execution, and NIC paths.
3. Make memory behavior predictable by controlling reclaim, compaction, transparent huge pages, NUMA balancing, and allocator behavior.
4. Stabilize network processing and time synchronization through NIC queue placement, interrupt affinity, PTP/PHC, and timestamping.

## Product

- Verified upstream stable Linux baseline with recorded source and build identity.
- Cloud profile for AWS EC2, ENA, and PTP/PHC workflows.
- Hardware profile for bare-metal and colocation systems with physical NIC, NUMA, and storage requirements.
- Native OS packages with HFTKernel, cpujitter, memjitter, matching kernel tools, and headers.
- Immutable SHA-256-verified releases delivered through signed private links.
- Sequential public release names for each kernel version: <kernel-version>-r1, <kernel-version>-r2, <kernel-version>-r3. A new kernel version starts at r1.
- Free technical consultation and custom kernel engineering for CPU, NIC, memory, isolation, and measurement requirements.

## Supported operating-system targets

- Ubuntu 20.04 LTS, 22.04 LTS, 24.04 LTS, and 26.04 LTS.
- Debian 11, 12, and 13.
- RHEL-compatible 8, 9, and 10.
- AlmaLinux 8, 9, and 10.
- Rocky Linux 8, 9, and 10.
- CentOS Stream 9 and 10.
- Fedora Linux 44.
- Arch Linux base-devel.
- NixOS-compatible base built in the official minimal Nix container.

## Package formats

- Ubuntu and Debian: DEB.
- Enterprise Linux and Fedora: RPM.
- Arch Linux and the NixOS-compatible base: TGZ.

## Measurement tools

- cpujitter measures scheduling and execution gaps on a pinned CPU, high-percentile and maximum latency, tail events, IRQ activity, and softIRQ activity.
- memjitter measures page-allocation and first-touch latency under controlled working-set and memory-pressure phases, including allocator-tail percentiles, maximum latency, and throughput.

## Controlled measurement results

- cpujitter p99.999 CPU gap: 7.413 microseconds to 0.443 microseconds, 94.02% lower.
- cpujitter gaps above 5 microseconds: 122,322 to 1,021, 99.17% fewer.
- cpujitter IRQ / softIRQ activity on the isolated CPU: 120,133 / 2,136 to 0 / 0.
- memjitter p99.999 4K first-touch latency: 256.000 microseconds to 11.840 microseconds, 95.4% lower.
- memjitter median maximum latency: 273.572 microseconds to 17.187 microseconds, 93.7% lower.

## Primary audience

Proprietary trading firms, hedge funds, market makers, exchange-connectivity teams, low-latency infrastructure teams, cloud trading teams, and bare-metal colocation teams.

## Canonical pages

- https://hftkernel.com/
- https://hftkernel.com/#trading-outcomes
- https://hftkernel.com/#results
- https://hftkernel.com/#packages
- https://hftkernel.com/#measurement
- https://hftkernel.com/#contact
- https://hftkernel.com/privacy.php
- https://hftkernel.com/terms.php