DPDK/dpdk
Dpdk
DPDK este un framework de procesare a pachetelor și o bibliotecă de networking cu kernel bypass, concepută pentru a accelera performanța planului de date al rețelei. Oferă un stack de rețea în user space care mută datele de rețea direct de la placa de interfață de rețea (NIC) către user space, eliminând overhead-ul kernel-ului sistemului de operare pentru a obține throughput ridicat și latență scăzută.
Framework-ul utilizează o arhitectură de driver poll mode care înlocuiește networking-ul bazat pe întreruperi cu un mecanism de polling constant. Această abordare permite implementarea funcțiilor și protocoalelor de rețea în afara kernel-ului.
Proiectul acoperă capabilități pentru procesarea pachetelor de înaltă performanță, virtualizarea funcțiilor de rețea (NFV) și dezvoltarea de appliance-uri de rețea personalizate, cum ar fi firewall-uri și load balancer-e.
Features
- Packet Processing Frameworks - Provides a high-performance framework for accelerating packet processing by moving network data directly from the NIC to user space.
- Poll-Mode Drivers - Employs a poll-mode driver architecture that continuously checks hardware for packets to eliminate interrupt latency.
- Network Stack Bypasses - Implements a networking library that bypasses the standard kernel network stack to achieve high throughput.
- User-Space Network Interface Access - Moves network driver logic from the kernel to user space to achieve faster data transfer.
- High-Performance Packet Processing - Handles massive volumes of network traffic by bypassing the OS kernel to reduce latency and CPU overhead.
- User-Space Packet Processing - Provides high-speed network packet manipulation by bypassing standard kernel overhead.
- Direct I/O Access - Provides direct I/O access to move network traffic from hardware to user-space memory, bypassing the kernel.
- User-Space Implementations - Provides tools for implementing network protocols in user space to maximize throughput and avoid kernel context switches.
- Zero-Copy Buffers - Implements zero-copy buffer management by passing pointers to packet data across processing stages.
- Single-Producer Single-Consumer Queues - Implements single-producer single-consumer lockless ring buffers for high-speed data exchange between CPU cores.
- Network Appliance Frameworks - Provides a framework for building specialized network appliances such as firewalls and load balancers.
- Hugepage Allocations - Implements hugepage-based memory allocation to improve memory access speed and reduce TLB misses for packet buffers.
- Virtualization Networking - Enables the creation of virtualized network services that process packets at near-wire speed.
- Run-To-Completion Models - Uses a run-to-completion model by assigning a packet to a single CPU core for all processing steps.
- Biblioteci de rețea - Libraries and drivers for fast packet processing.
