HKHFTKernel

Release notes

HFTKernel evolves with every stable Linux release.

Latency-relevant improvements in networking, timekeeping, memory behavior, virtualization, tracing, build reproducibility, cpujitter, and memjitter.

Read HFTKernel 7.1.8 Open kernel docs

Public changelog

Release notes

Each entry focuses on upstream changes that support HFTKernel’s primary goals: minimal jitter, controlled tail latency, a stable data path, accurate timekeeping, and measurable behavior under pressure.

HFTKernel 7.1.8

More Reliable Zero-Copy Networking, More Accurate Timing, and More Predictable Memory Behavior

Current changelog entry

HFTKernel has been updated to Linux Kernel 7.1.8.

This release incorporates upstream improvements that are particularly relevant to HFT, market data, execution gateways, and other latency-sensitive infrastructure.

Key improvements

  • More reliable AF_XDP zero-copy and multi-buffer TX
  • Improved network queue, IRQ, and TX completion handling
  • Greater PTP, PHC, and DPLL stability during reset recovery
  • Improved MGLRU accounting and HugeTLB behavior for memory-intensive workloads
  • Stronger KVM, VMX, and TLB paths
  • More accurate RTLA timerlat detection and tracing of rare latency events
  • Improved build reproducibility and packaging workflows

Native measurement tools included

cpujitter

for measuring CPU execution jitter, IRQ interference, and tail latency.

memjitter

for analyzing allocation latency, first-touch behavior, locked working sets, and memory performance under pressure.

HFTKernel continues to evolve as a predictable and measurable Linux platform for workloads where average performance is not enough, and the rarest latency outliers matter most.

View OS packages Review boot parameters Request production boot profile Permanent link

HFTKernel 7.1.7

More Stable Networking, More Accurate Timekeeping, and More Predictable Latency

HFTKernel 7.1.7 incorporates the cumulative improvements from Linux Kernel 7.1.6 and 7.1.7 that directly support its core goals: minimal jitter, a stable data path, accurate timekeeping, and controlled tail latency.

Key improvements

  • More balanced network IRQ distribution for Azure MANA
  • Automatic recovery of GVE RX queues
  • Improved GRO/GSO and AF_PACKET handling for stable market data processing
  • Improvements to PHC/PTP and hardware timestamping
  • More reliable KVM timers and AMD IOMMU behavior
  • Improvements across memory allocation paths, tracing, and observability
  • Additional AMD Safe-RET hardening

Native measurement tools included

The release also includes cpujitter and memjitter, our tools for measuring CPU execution jitter, memory latency, and tail-latency behavior under controlled workloads.

cpujitter

measures CPU execution gaps, IRQ and softIRQ interference, and high-percentile latency on controlled CPU workloads.

memjitter

measures allocation, first-touch, locked working-set, and memory-pressure tail behavior.

Together, HFTKernel, cpujitter, and memjitter provide both the optimized runtime and the measurement tools needed to validate latency-sensitive infrastructure.

HFTKernel 7.1.7 continues to turn Linux into a more predictable and measurable platform for HFT, market data, and other latency-sensitive workloads.

View OS packages Review boot parameters Request production boot profile Permanent link

Release philosophy

Upstream stability, workload-specific policy, and measurable outcomes

HFTKernel starts from an exact stable Linux release, records the toolchain and build identity, applies the selected Cloud or Hardware profile, packages cpujitter and memjitter with the runtime, and validates the result as an immutable release.

Exact baselineVerified stable Linux source and pinned kernel version.
Controlled profileCPU, memory, network, timekeeping, and isolation policy matched to the platform.
Native packagesDEB, RPM, or TGZ bundles for the supported production matrix.
Measured deliverycpujitter and memjitter provide repeatable evidence at the tail.

Deployment planning

Evaluate the release against your CPU, NIC, memory, and timing topology.

Package selection is only the first step. The production result depends on boot parameters, runtime affinity, queue placement, PTP/PHC design, memory policy, and a controlled measurement plan.

Request a production boot profile Read the boot-parameter reference