How LVS Powers Million-Connection Load Balancing: Modes, Algorithms, and HA

LVS (Linux Virtual Server) achieves million‑level concurrency because it is a high‑performance Layer‑4 load balancer that runs entirely in the Linux kernel, avoiding the costly user‑space processing required by solutions like Nginx.

Operating at the IP and transport layers, LVS eliminates frequent user‑kernel context switches, dramatically improving throughput.

Since LVS only handles network and transport packets without parsing application‑layer protocols such as HTTP, its processing speed is extremely fast and its performance overhead is minimal, which is the key to supporting high concurrency.

LVS Core Modes

LVS offers three core forwarding modes to suit different scenarios.

NAT mode : suitable for low traffic but imposes higher load on LVS.

Direct Routing (DR) mode : packets bypass LVS after routing, the most commonly used mode that supports million‑level concurrency.

<ol><li>Client → LVS (request)</li><li>LVS → Real Server (direct response to client)</li></ol>

TUN mode : high‑performance cross‑network scenario, ideal for geographically distributed data centers.

Powerful Load‑Balancing Algorithms

LVS supports multiple algorithms, including Round‑Robin (RR), Weighted Round‑Robin (WRR), Least Connections (LC), and Weighted Least Connections (WLC).

RR distributes requests sequentially across servers, suitable when server capacities are similar.

WRR assigns a weight to each server; higher‑weight servers receive more requests.

LC sends new requests to the server with the fewest active connections, providing dynamic load distribution.

These algorithms intelligently route traffic based on real‑time server load, ensuring balanced and efficient operation.

High Availability Assurance

Even with strong performance, a single LVS failure would disrupt services. To prevent this, Keepalived is commonly used to provide high availability.

Keepalived implements the VRRP (Virtual Router Redundancy Protocol) and manages a cluster with one master and one or more backup servers.

It continuously monitors the master’s health; if the master fails, Keepalived automatically migrates the virtual IP (VIP) to a backup, maintaining uninterrupted service.