System Configuration:
-
CPU: AMD Ryzen 9 9950X3D (32 Threads) @ 5.75 GHz
-
GPU: NVIDIA RTX 6000 Ada Generation (96GB VRAM)
-
RAM: 128GB (126,300 MiB)
-
OS: Ubuntu 22.04.5 LTS (x86_64)
-
Software Stack: Isaac Sim 5.1.0 (Host) + Isaac ROS / nvblox (Docker)
-
RMW Implementation:
rmw_fastrtps_cpp
Description: Hello, I am encountering a critical memory leak/accumulation issue leading to a system crash while running isaac_ros_nvblox.
I am running Isaac Sim on the native host (Ubuntu 22.04) and the Isaac ROS nvblox node inside a Docker container. Despite using high-end hardware (Ryzen 9 9950X3D + RTX 6000 Ada + 128GB RAM), the system RAM (not VRAM) rapidly fills up until the Linux OOM Killer terminates the Isaac Sim process.
Key Observations & Problem Analysis:
-
System RAM Overflow: VRAM usage is stable, but Host System Memory usage spikes uncontrollably. It appears that the ROS 2 Bridge in Isaac Sim keeps buffering messages because the subscriber (nvblox in Docker) cannot process them fast enough.
-
CPU Saturation: Even with a Ryzen 9 9950X3D, I observe extremely high CPU usage. This suggests a significant bottleneck in data serialization/deserialization between the Host and the Docker container.
-
Hardware Context: On my previous setup (RTX 3090, WSL2), the network throttling likely prevented this overflow. However, on this high-performance Native Ubuntu setup, the unthrottled data generation from Isaac Sim overwhelms the ROS 2 pipeline.
Any guidance on architecture configuration for this high-throughput setup would be appreciated.