Master Linux Clusters: High Availability, Load Balancing & Performance

Preface

Linux

is a free, open‑source Unix‑like operating system based on POSIX and UNIX. Although the term originally referred only to the kernel, it is commonly used to describe the whole system built on the Linux kernel together with GNU tools and libraries.

1. Cluster Concepts

Distributed computing aims to shorten the execution time of a single task, while a cluster improves overall efficiency by increasing the number of tasks completed per unit time.

Two Key Characteristics

Scalability – new service nodes can be added dynamically to boost performance.

High availability – redundant service nodes ensure that if one fails, another takes over, preventing service interruption.

Two Core Capabilities

Load balancing – distributes tasks evenly across compute and network resources.

Error recovery – if a resource fails on one node, another node continues the task transparently.

Two Core Technologies

Cluster address – a single virtual address that clients use to reach the internal service nodes; managed by a load balancer.

Internal communication – heartbeat and context information exchanged between nodes to support load balancing and error recovery.

2. Cluster Types

Main Categories

(1) High‑availability cluster – e.g., two‑node active‑passive (dual‑machine hot standby) and multi‑node HA solutions using heartbeat, keepalived, RAC, shared, or share‑nothing architectures.

(2) Load‑balancing cluster – implementations such as LVS, Nginx Proxy, HAProxy, keepalived, F5, Netscaler.

(3) High‑performance cluster – e.g., Beowulf clusters.

3. High‑Availability Clusters

(1) Overview

A high‑availability cluster reduces service downtime by grouping computers that present a single set of network resources to users. In a two‑node configuration (dual‑machine hot standby), one server automatically takes over when the other fails, without manual intervention.

(2) Sub‑categories

High‑throughput computing – tasks can be split into independent parallel subtasks.

Distributed computing – parallel subtasks are tightly coupled and require extensive data exchange.

(3) Standards

(4) Solutions

Linux‑HA’s heartbeat system and related components; Red Hat’s open‑source RHCS suite.

4. Load‑Balancing Clusters

(1) Overview

All nodes are active and share the workload. Typical examples include web‑server, database, and application‑server clusters.

(2) Network Layers

Layer‑4 Load Balancing

Maps a public IP address to multiple internal server IPs, selecting a different internal IP for each TCP connection.

Layer‑7 Load Balancing

Controls traffic at the application layer, inspecting HTTP headers to direct requests, detect errors, and route different content types appropriately. Drawbacks include protocol limitation (mostly HTTP) and higher resource consumption.

(3) Implementation Methods

(1) Hardware – dedicated appliances such as F5, Array, Redware (expensive, network‑layer only, suitable for high traffic).

(2) Software – solutions like LVS, Nginx, HAProxy (cost‑effective, flexible, meet most load‑balancing needs).

(4) Common Applications

DNS load balancing – multiple IPs share a single domain name.

Proxy load balancing – forwards requests to multiple internal servers to improve static content delivery.

Reverse‑proxy load balancing – dynamically forwards external connections to internal servers.

Address‑translation gateway load balancing – maps an external IP to several internal IPs.

Protocol‑level load balancing – e.g., HTTP redirects.

NAT load balancing – translates external to internal IPs, useful for hiding internal topology.

Hybrid load balancing – combines different methods across multiple server groups for optimal performance.

5. High‑Performance Clusters

(1) Overview

High‑performance computing (HPC) clusters solve extremely complex problems in fields such as weather modeling, crash simulation, genomics, and nuclear testing, and are used in government, education, manufacturing, finance, energy, life sciences, and digital media.

(2) Architecture

Compute nodes – execute user programs.

Storage nodes – provide storage services.

Monitoring nodes – monitor system health and manage power/events.

Storage‑area network – connects storage nodes and disk arrays.

Video switching system – allows remote control of nodes.

Firmware layer – low‑level software for devices.

Node OS layer – operating system and drivers.

Cluster management layer – job scheduling, monitoring, and management software.

Application support layer – parallel environments, math libraries, debugging tools.

Application layer – scientific and engineering workloads.