3 个仓库
Measures packet processing speed by sending hundreds of millions of packets per second to evaluate raw throughput.
Distinct from Packet Processing Frameworks: Distinct from Packet Processing Frameworks: focuses on benchmarking speed rather than providing a framework for processing.
Explore 3 awesome GitHub repositories matching networking & communication · Packet Processing Speed Benchmarks. Refine with filters or upvote what's useful.
dperf is a high-performance network testing tool that generates massive multi-protocol traffic for benchmarking network devices and servers. It uses DPDK to bypass the kernel network stack, enabling line-rate packet generation and capture with zero-copy processing, and implements a full TCP/IP stack in user space to achieve millions of connections per second from a single server. The tool is designed for stress-testing Layer 4 load balancers and gateways, capable of simulating billions of concurrent TCP or UDP sessions and generating tens of millions of new connections per second. It supports
Measures packet processing speed by sending hundreds of millions of packets per second to evaluate raw throughput.
DPDK 是一个数据包处理框架和内核旁路网络库,旨在加速网络数据平面性能。它提供了一个用户空间网络栈,将网络数据直接从网卡移动到用户空间,消除了操作系统内核开销,从而实现高吞吐量和低延迟。 该框架利用轮询模式驱动架构,用持续轮询机制取代了中断驱动的网络处理。这种方法允许在内核之外实现网络功能和协议。 该项目涵盖了高性能数据包处理、网络功能虚拟化以及防火墙和负载均衡器等自定义网络设备开发的功能。
Provides high-speed network packet manipulation by bypassing standard kernel overhead.
f-stack 是一个内核旁路(kernel-bypass)框架和用户态 TCP/IP 协议栈,旨在提高数据包处理速度和整体吞吐量。它利用 DPDK 将网络流量直接路由到用户态,并移植了 FreeBSD 网络栈,从而在宿主操作系统之外处理传输层和网络层协议。 该项目提供了一个 Socket API 兼容层,允许标准网络应用程序在高性能后端上运行,而无需修改代码。它还具有基于协程的网络 API 和微线程接口,用于管理异步网络 IO 和有状态任务,无需复杂的异步回调逻辑。 该框架涵盖了高吞吐量 TCP/IP 实现、并发连接扩展和事件驱动的 IO 多路复用。它还包含路由配置功能,可确定传入流量是由用户态协议栈还是标准内核网络栈处理。
Directs network traffic to user space via DPDK to eliminate kernel overhead and minimize latency.