GroupeZ Docs

Optimization

Discover how zAuctionHouse is optimized for large-scale servers

zAuctionHouse has been designed and developed from the ground up to handle very large servers. Performance has been a core priority throughout the entire development process, not an afterthought.

Built for Scale

Whether you're running a small survival server with a few dozen players or a massive network with thousands of concurrent players, zAuctionHouse is engineered to maintain optimal performance under any load.

Design Philosophy

  • Async-first architecture - All database operations run asynchronously to never block the main thread
  • Efficient data structures - Carefully chosen algorithms and data structures for minimal memory footprint
  • Smart caching - Intelligent caching strategies reduce database queries
  • Lazy loading - Data is loaded only when needed, reducing startup time and memory usage
  • Connection pooling - Optimized database connection management for high throughput

Internal Performance Tools

zAuctionHouse includes built-in tools for developers and server administrators to measure and monitor performance.

Debug Mode

Enable debug mode in the configuration to see detailed performance metrics:

debug: true

This activates internal timing measurements that help identify any performance bottlenecks.

Test Data Generator

The plugin includes a powerful test data generator that allows you to create large amounts of test items to stress-test your server configuration.

Command:

/zah admin generate <amount>

This command generates fake auction items, allowing you to test how your server handles large volumes of data before going live.

Only use this command in a test environment! Generated items should be cleared before production use.

Performance Benchmarks

We have conducted extensive benchmarks to ensure zAuctionHouse performs well under heavy load.

Sorting 100,000 Items

One of the most demanding operations in an auction house is sorting large numbers of items for display. We tested sorting 100,000 items simultaneously.

Description
MetricResult
Items sorted100,000
Time taken45ms
OperationFull sort with filters

Test Environment

The benchmark was conducted on the following hardware:

ComponentSpecification
Operating SystemWindows 11
ProcessorAMD Ryzen 7 5700X 8-Core
Server RAM4 GB allocated
Java VersionJava 21
Server SoftwarePaper 1.21.11

These benchmarks were performed with default plugin settings. Performance may vary based on your specific configuration and hardware.

Real-World Performance

zAuctionHouse is actively used on some of the largest Minecraft networks, handling:

  • Thousands of concurrent players
  • Hundreds of thousands of active listings
  • Millions of transactions per month

Optimization Techniques Used

  1. Indexed database queries - All frequently accessed columns are properly indexed
  2. Batch operations - Multiple database operations are batched together when possible
  3. Memory-efficient serialization - Items are serialized using optimized byte formats
  4. Thread-safe operations - Concurrent access is handled efficiently without locks where possible
  5. Pagination - Large result sets are paginated to reduce memory pressure
  6. Selective placeholder resolution - Item lore placeholders are pre-detected at configuration load time. At render time, only the placeholders actually referenced in the lore template are computed. For example, the default expired-item lore only uses %time-remaining%, so the plugin skips computing %price%, %seller%, %status%, and all other placeholders — saving CPU and memory per item rendered

Monitoring

Server TPS

Monitor your server TPS when the auction house is heavily used. zAuctionHouse should have minimal impact on TPS due to its async design.

Database Performance

For MySQL/MariaDB, monitor:

  • Connection pool usage
  • Query execution time
  • Table sizes

Memory Usage

The plugin is designed to be memory-efficient, but monitoring Java heap usage during peak times is recommended for very large installations.

Conclusion

zAuctionHouse is not just another auction house plugin - it's a performance-engineered solution built for the demands of modern, large-scale Minecraft servers. The 45ms sorting time for 100,000 items demonstrates our commitment to delivering a plugin that won't slow down your server, no matter how popular your auction house becomes.

On this page

GroupeZ MCP

Documentation in your AI

MCP lets your AI search GroupeZ documentation and read current examples to help you configure your plugins.

Read-only · No documentation API key needed

1. Copy the server URL

Transport: HTTP (Streamable HTTP). Authentication: none. Your tool uses its own model and subscription limits.

  1. Open Plugins in ChatGPT on the web, then + → Add custom MCP server.
  2. Name it ‘GroupeZ Documentation’, paste the URL above and choose ‘No authentication’. Create and install the plugin.
  3. In a conversation, type @ and select the plugin before asking. If needed, enable search, get_page and list_pages in its settings.
Official setup guide

3. Ask a question

Use GroupeZ MCP to help me create a zMenu inventory. Search relevant pages, read the examples and cite your sources.
Other clients and troubleshooting

For clients using mcpServers, adapt this example to their format:

{
  "mcpServers": {
    "groupez-docs": {
      "url": ""
    }
  }
}

Available tools are search, get_page and list_pages. Refresh the connector’s tools after updates. If your AI cannot access them, check that the connector is enabled in the conversation and your account allows custom MCP servers. Opening the URL in a browser is not an MCP test.