UCCL is an efficient communication library for GPUs, covering collectives, P2P (e.g., KV cache transfer, RL weight transfer), and EP (e.g., GPU-driven)

Infiniband Verbs Performance Tests

The simplest, fastest repository for training/finetuning medium-sized GPTs.

个人构建MoE大模型：从预训练到DPO的完整实践

RDMA core userspace libraries and daemons

Optimized primitives for collective multi-GPU communication

NVIDIA Inference Xfer Library (NIXL)

A tiny yet powerful LLM inference system tailored for researching purpose. vLLM-equivalent performance with only 2k lines of code (2% of vLLM).

This is a repo for a mini distributed warehouse management system based on gRPC

Network Benchmarking Utility

LITE Kernel RDMA Support for Datacenter Applications. SOSP 2017.

Fastsocket is a highly scalable socket and its underlying networking implementation of Linux kernel. With the straight linear scalability, Fastsocket can provide extremely good performance in multi…

mimalloc is a compact general purpose allocator with excellent performance.

Pond: CXL-Based Memory Pooling Systems for Cloud Platforms (ASPLOS'23)

The Artifact Evaluation Version of SOSP Paper #19

Checkpoint/Restore tool

Linux kernel source tree

a library version of FreeBSD's TCP/IP stack plus extras

Remote Persistent Memory Access Library

AIFM: High-Performance, Application-Integrated Far Memory

Tools for profiling the Linux network stack.

neper is a Linux networking performance tool.

libunwind official github repo (in need of new / additional maintainer, mail/open issue if interested)

Modern HTTP benchmarking tool

Main gperftools repository

Fork from official iperf-3.1.3, and run on the dpdk user space TCP/IP stack(ANS).

iperf3: A TCP, UDP, and SCTP network bandwidth measurement tool

nettrace is a eBPF-based tool to trace network packet and diagnose network problem.

BCC - Tools for BPF-based Linux IO analysis, networking, monitoring, and more