Low Latency System

A curated collection of resources for building low-latency systems, with a focus on financial trading. Every reference below carries a summary badge linking to a plain-language summary of that reference, and a level tag — beginner intermediate advanced — so you can decide what to read before you click.

New here? Read The Complete Guide — a single page that organizes this whole topic from beginner to expert: what latency is, where the nanoseconds go, and which techniques buy what.

Content fully refreshed in July 2026. The research behind the update — every link checked, dead links replaced, new state-of-the-art references added — is documented in the research notes.

Start Here

Tick-to-Trade Latency summary beginner — what the metric means and 2024-era reference numbers for software, FPGA and ASIC tiers.
Kernel Bypass summary beginner — the clearest introduction to why and how trading systems leave the kernel.
What Every Programmer Should Know about How CPUs Work — Matt Godbolt summary beginner — the best current 43-minute “mechanical sympathy” primer.
What is Low Latency C++? — Timur Doumler (2 parts) summary intermediate — the best structured pedagogical treatment of the whole topic.

Linux OS Tuning

Monitoring and Managing System Status and Performance (RHEL 9) summary beginner — the modern successor to the retired RHEL 7 Performance Tuning Guide.
Optimizing RHEL for Real Time for Low Latency Operation summary intermediate — Red Hat’s canonical low-latency guide; since RHEL 9 the real-time kernel is included with standard RHEL.
Erik Rigtorp — Low Latency Tuning Guide summary intermediate — the most-cited practitioner checklist for HFT-style x86_64 tuning.
TuneD — the Linux tuning daemon summary beginner — home of the latency-performance, network-latency and cpu-partitioning profiles.
Getting Started with PREEMPT_RT summary beginner — real-time preemption is in mainline Linux since 6.12 (2024); here’s how to use it.
The Realtime Preemption End Game (LWN) summary intermediate — how and why PREEMPT_RT finally merged after 20 years.
SUSE Labs — CPU Isolation series summary advanced — the nohz_full maintainer explains what CPU isolation really does inside the kernel.
CPU Isolation for HFT: what isolcpus doesn’t do summary intermediate — a trading case study of the full isolcpus + nohz_full + rcu_nocbs recipe.
RHEL Performance Guide (Myllynen) summary intermediate — the scattered official docs condensed into one actionable page.
Brendan Gregg — Linux Performance summary beginner — the observability hub: perf, ftrace, eBPF, flame graphs, and the standard textbooks.
TLB Shootdowns (JabPerf) summary advanced — the latency source most tuning checklists miss.
Historical: Low Latency Tuning for RHEL 7 (PDF) summary · Red Hat Summit 2018 slides (PDF) summary — RHEL 7 is EOL; kept for background.

Server Hardware & CPUs

Since this site’s last update, the landscape changed: HPE retired its overclocked “Trade and Match” servers, and boutique vendors (Blackcore, Hypertec CIARA) now own that niche. On the mainstream side, AMD EPYC 9005 F-series (e.g. 9575F), EPYC 4004/4005 X3D parts with 3D V-Cache, and Intel Xeon 6 with Latency Optimized Mode are the current choices. Single-socket is the settled architecture for the latency path, and core-to-core latency is now a first-class CPU selection metric.

AMD EPYC 9005 BIOS & Workload Tuning Guide summary advanced — the current canonical AMD server tuning document (rev 2.2, June 2026).
Dell PowerEdge 16G BIOS Tuning — Low Latency summary intermediate — replaces Dell’s dead 2010 low-latency whitepaper.
Configuring HPE ProLiant Servers for Low-Latency Applications summary intermediate — the long-running HPE whitepaper at its new home, updated through Gen11.
Intel Xeon 6 Performance & Power Profiles (PDF) summary advanced — Latency Optimized Mode and Efficiency Latency Control, from Intel.
Phoronix — Xeon 6 Latency Optimized Mode benchmarks summary intermediate — independent measurements: ~17% geo-mean gain.
Xeon 6 High-Priority Cores (SST-BF) Explained summary beginner — guaranteed-fast cores for your hot threads.
Chips and Cheese — Xeon 6 memory subsystem summary advanced — measured L3/DRAM/core-to-core latencies, Intel mesh vs AMD chiplets.
core-to-core-latency summary intermediate — measure your CPU’s topology; community results database.
Blackcore — why overclocked trading servers summary beginner — the boutique-server perspective (vendor content).
hft-server-settings summary intermediate — a community checklist of the settings practitioners actually change.
Historical: Intel “Optimizing Applications for Latency” parts 1 & 2 summary · IBM/Lenovo Linux tuning Redpaper (2008) summary

Networking, NICs & Kernel Bypass

The vendor map was redrawn: Solarflare → Xilinx → AMD (X4-series NICs, Oct 2025), Mellanox → NVIDIA (VMA replaced by XLIO), Exablaze → Cisco (which then exited the ULL switch business in 2023, leaving Arista 7130 as the de facto layer-1 platform).

Understand the kernel path first

Understanding TCP/IP Network Stack (CUBRID) summary beginner — the mental model: how a packet moves through Linux.
Monitoring and Tuning the Linux Networking Stack (packagecloud) summary advanced — the definitive source-level RX-path walkthrough.
Linux Network Performance Ultimate Guide (ntk148v) summary intermediate — one maintained page consolidating all the classic tuning material.
How to Receive a Million Packets Per Second (Cloudflare) summary intermediate — RSS, SO_REUSEPORT and NUMA pinning, hands-on.
Why We Use the Linux Kernel’s TCP Stack (Cloudflare) summary beginner — the case against bypassing everything.
Extreme HTTP Performance Tuning (talawah.io) summary advanced — 5× throughput without bypass, every knob measured with flame graphs.
Zero Copy I: User-Mode Perspective summary intermediate — the 2003 classic behind today’s zero-copy techniques.

Kernel bypass stacks

Onload User Guide (AMD UG1586) summary advanced — Onload, ef_vi and TCPDirect: the whole Solarflare stack, actively maintained.
OpenOnload on GitHub summary intermediate — open source since 2021; now runs on non-Solarflare NICs via AF_XDP.
NVIDIA XLIO documentation summary advanced — the successor to Mellanox VMA for socket acceleration on ConnectX.
Rivermax + NEIO FastSockets for financial services summary intermediate — NVIDIA’s 2025 answer for market-data workloads.

NICs

AMD Solarflare X4 launch summary beginner — the current flagship trading NIC family (Oct 2025).
Performance Tuning for Mellanox Adapters (NVIDIA) summary advanced — successor to the classic Mellanox tuning PDF.
Cisco Nexus SmartNIC (ex-ExaNIC) summary intermediate — 568 ns trigger-to-response; now in maintenance mode.

Switches

Arista 7130 series summary intermediate — layer-1 (~4 ns), MetaMux (~39 ns), MetaWatch; the de facto ULL switching platform.
Measuring the Latency of a 4 ns Switch (PDF) summary advanced — how to measure below your test gear’s resolution.

RDMA & InfiniBand

RDMA Aware Networks Programming User Manual summary advanced — verbs, queue pairs and rdma_cm, at its new NVIDIA home. Note: in trading, RDMA now means RoCEv2 on internal fabrics; exchange-facing paths are Ethernet UDP multicast.

FPGA & Hardware Acceleration

AMD Alveo UL3524 product brief (PDF) summary intermediate — the first FPGA card purpose-built for trading: <3 ns transceivers.
Exegy + AMD: the 13.9 ns STAC-T0 record summary intermediate — the 2024 tick-to-trade world record.
Where next for ultra-low latency trading? (A-Team) summary beginner — software vs FPGA vs ASIC, strategically framed.
Orthogone — HFT FPGA engineering summary intermediate — a modern FPGA trading stack with real latency numbers.
FPGA trading with Nasdaq ITCH & OUCH (Design Gateway) summary intermediate — an all-hardware tick-to-trade pipeline, protocol by protocol.
Algo-Logic summary beginner — long-running FPGA-for-finance vendor, still active; classic Lockwood slides on the Wayback Machine.

Time Synchronization & Timestamping

How PTP is deployed at Meta summary intermediate — tens-of-ns sync at hyperscale, fully open-sourced stack.
OCP Time Appliances Project — Time Card summary advanced — open hardware that turns any server into a stratum-1 time appliance.
White Rabbit (CERN) summary intermediate — sub-nanosecond timing, now the IEEE 1588-2019 High Accuracy profile.
Deutsche Börse / Eurex White Rabbit service summary beginner — how an exchange distributes its own clock to co-location customers.
linuxptp documentation summary intermediate — ptp4l, phc2sys, ts2phc: the standard Linux PTP stack.
MiFID II RTS 25 and time sync on Linux (Red Hat) summary beginner — the regulatory requirement mapped to Linux capabilities.
Real-world PTP accuracy across NICs (scottstuff.net) summary advanced — measured 2025 data: NIC choice changes PTP accuracy by 45×.
Intel rdtsc benchmarking whitepaper (mirror) summary advanced — the classic Paoloni paper on cycle-accurate timing (Intel removed the original).

Measurement, Benchmarks & Jitter Detection

STAC benchmarks summary intermediate — STAC-T0/N1/TS: the audited industry benchmarks, with full configuration disclosures.
cns-sysjitter summary advanced — Solarflare’s classic OS-jitter meter at its new GitHub home.
cyclictest & rt-tests summary intermediate — the standard RT latency suite; pair with the in-kernel rtla tracer on 5.17+.
hiccups summary advanced — a simple OS-hiccup meter (unmaintained but readable in one sitting).
sockperf summary beginner — socket-level latency benchmarking with percentile histograms.
Corvil Analytics (Pico) summary beginner — what production-grade nanosecond latency monitoring looks like.
c2clat & core-to-core-latency summary intermediate — characterize inter-core latency before choosing a pinning layout.
Google Benchmark summary beginner — the standard C++ microbenchmark harness.

Programming Languages & Code

C++

When Nanoseconds Matter — David Gross (CppCon 2024 keynote) summary advanced — the current definitive HFT C++ talk.
Trading at Light Speed — David Gross (Meeting C++ 2022) summary advanced — the companion full-system design talk.
When a Microsecond Is an Eternity — Carl Cook (CppCon 2017) summary intermediate — the classic introduction to HFT C++.
What is Low Latency C++? — Timur Doumler (C++Now 2023) summary intermediate — two parts; the best structured modern treatment.
SPSC Lock-free FIFO From the Ground Up — Charles Frasch (CppCon 2023) summary intermediate — build the workhorse queue, benchmark every step.
C++ Design Patterns for Low-latency Applications (arXiv) summary advanced — the HFT folklore, measured with statistical rigor.
Google Benchmark summary beginner — prove your optimization actually helped.

Concurrency & lock-free

C++ Atomics, From Basic to Advanced — Fedor Pikus (CppCon 2017) summary advanced — what atomics really do; still canonical.
Lock-Free Programming — Herb Sutter (CppCon 2014, parts I & II) summary advanced — the conceptual foundation (Part II link fixed — the old one was truncated).
Erik Rigtorp — essays & libraries summary advanced — SPSCQueue, MPMCQueue, Seqlock, and the awesome-lockfree list.
1024cores — Dmitry Vyukov summary advanced — the old 1024cores.net domain is hijacked; use this Google Sites URL.

Java

LMAX Disruptor & Chronicle Queue summary intermediate — the GC-free Java toolkit.
Aeron summary advanced — low-latency messaging and clustering; note the new aeron-io org.
Evolution of Financial Exchange Architectures — Martin Thompson summary intermediate — 25 years of exchange architecture in one talk.
Azul Platform Prime (Zing) & the C4 collector summary intermediate — pauseless Java; compare with OpenJDK’s free generational ZGC (JDK 21+).
Historical: SubMicroTrading blog + open-sourced code summary · epickrram (LMAX) jitter posts summary

Rust

Rust vs C++ for trading systems (Databento) summary intermediate — the grounded mid-2026 industry consensus.
hftbacktest summary advanced — the flagship open-source Rust HFT project.

Communities & protocols

Mechanical Sympathy Google Group summary intermediate — quieter than its heyday, but the archive is gold.
Databento Microstructure Guides summary beginner — ITCH, OUCH, MoldUDP64 and tick-to-trade, explained properly.

Contribute

Found a dead link, or a resource that belongs here? Please open a GitHub issue.