Scalable Distributed Image Storage and Load Balancing for Web Sites

Today we record the principles of distributed storage and load balancing for images.

For web servers, serving image files consumes significant resources because each image creates a separate connection, increasing concurrency and server load.

Small sites typically store pages and images in a single directory, which is simple but not scalable.

Larger sites store many images; image traffic dominates page size, and browsers cannot download all images from one server simultaneously, making I/O a performance bottleneck as concurrent users grow, prompting the need for distributed storage.

For a medium‑scale business site, a dynamic distributed image storage and load‑balancing solution can be built with minimal hardware cost, allowing the number of image servers and their directories to be adjusted on demand, ensuring scalability and elasticity.

After adding image servers, the system must address four problems:

(1) How to deploy images across servers and dynamically decide where to store an uploaded image.

(2) How to achieve load balancing so each image server has an equal chance to store images.

(3) How to distribute images across multiple sub‑directories on a server to avoid filesystem limits.

(4) How to expand the image server pool dynamically as performance needs grow.

Below is the architecture diagram:

The web server serves pages, retrieves image URLs from the database, and returns the page to the client, which then downloads images directly from the appropriate image servers.

When a user uploads an image, the web server queries the database for the current status of all image servers, selects a server and directory using an algorithm, calls the server’s web service to store the image, and records the image ID and URL in the database. The database keeps records of image locations and server status; the image server cluster can be expanded as needed.

Image Server Information Table

The database stores each server’s status: ServerId (primary key), ServerName, ServerUrl (root URL), PicRootPath (physical directory), MaxPicAmount (maximum images), CurPicAmount (current count), and FlgUsable (availability).

Saving Images to an Image Server

Image servers can expose services via Web Service, WCF, WebClient, shared files, etc.

Random Selection Algorithm for Image Servers

From the status table, select the set C of usable servers and its count N. Generate a random number R1, compute I = R1 % N, and choose C[I] as the target server.

Detecting Image Server Health

A heartbeat mechanism can be used to monitor server availability.

When a client requests a page, the web server obtains image URLs, checks each server’s FlgUsable flag; if false, it replaces the image URL with a default image hosted on the web server, otherwise it returns the original URL. Clients then download images directly from multiple servers, enabling concurrent downloads, reducing load on the web server’s I/O, and improving site performance.

Note: This is a conceptual overview; implementation details require further consideration.