User GuideNIXL Backends

GDS_MT

View as Markdown

Overview

GDS_MT is a compatibility name for the same multi-threaded engine selected by GDS with mode=mt. It uses the same NVIDIA cuFile API and hardware requirements as GDS but distributes I/O across multiple threads for higher throughput on parallel file operations. The default behavior of createBackend("GDS_MT") remains unchanged.

GDS_MT will be removed in a future update. New integrations should use GDS with mode=mt.

PropertyValue
Transfer TypeVRAM ↔ File; DRAM ↔ File
ProtocolGPUDirect Storage (cuFile API, multi-threaded)
Best ForParallel GPU-to-file transfers

Installation

GDS_MT shares the same prerequisites and installation requirements as the GDS backend. See the GDS Installation section for prerequisites, cuFile verification, and build options.

Configuration

GDS_MT accepts only the multi-threaded engine configuration. The batch parameters documented for GDS do not apply.

Backend Parameters

ParameterDefaultDescription
thread_countmax(1, hardware concurrency / 2)Number of persistent TaskFlow workers.

Environment Variables

VariableTypeDefaultDescription
CUFILE_ENV_PATH_JSONString (path)Default cuFile configPath to the cuFile configuration JSON file. This is a cuFile SDK variable, not NIXL-specific. Override to customize GPUDirect Storage behavior.

Build Options

OptionDefaultDescription
gds_path/usr/local/cuda/Path to GDS cuFile installation. Shared with the GDS backend.
disable_gds_backendfalseDisable GDS backend entirely. Disables both GDS and GDS_MT.

When to Use

  • Existing integrations using the compatibility name — GDS_MT continues to select the multi-threaded engine until the compatibility name is removed.
  • New multi-threaded integrations — Use GDS with mode=mt instead of adopting GDS_MT.