multiqueue

Multiqueue is a design concept in computer networking and input/output systems that splits workloads into multiple queues. Each queue can be processed independently by hardware or software, enabling parallelism and reducing contention on shared data structures. In network interfaces, multiqueue refers to the ability to create several transmit and receive queues on a single device. The operating system distributes packets across queues using a hash of packet headers or other metadata, which improves cache locality and enables per-queue interrupt processing, often with IRQ affinity so that queues map to different CPU cores. In Linux, core features include Receive-Side Scaling (RSS) and Transmit Packet Steering (XPS). The kernel may also expose per-queue NAPI polling to further reduce interrupt overhead. Virtualization and cloud environments leverage multiqueue to support higher virtual network throughput with virtio multiqueue devices and SR-IOV virtual functions.

In storage and I/O subsystems, multiqueue I/O queues allow multiple outstanding I/O requests to be processed in parallel by storage controllers, reducing queuing delays and improving throughput on fast SSDs.

Benefits include higher throughput, lower latency under load, and better scalability on multi-core systems. Challenges include achieving effective load balance across queues, preserving in-order delivery when required, and managing increased complexity in drivers and the kernel’s scheduler. The optimal number of queues often matches the number of CPU cores or hardware channels, and tuning may depend on workload characteristics.