How to enable and test IPC buffer sharing between VST and DeepStream?

Accelerated Computing Intelligent Video Analytics Metropolis Microservices for Jetson
, ,
1

We are testing JPS on:

  • HW: Nvidia Jetson AGX Orin
  • Jetpack: 7.0

VST and deepstream-ipc-test works, but there is no guideline to check feature of sharing IPC buffer as mentioned in JPS intro (VST — Jetson Platform Services documentation).

Your helps are really appreciated.

3

Here is the config for IPC between VST and DeepStream: VST Config — Jetson Platform Services documentation. Are you meet any issue with the feature?

1 Like
4

Thank you for pointing out two settings for ipc.

It seems that I still cannot enable that feature:

  • HW: Nvidia Jetson AGX Orin
  • Jetpack: 7.0
  • vst image: nvcr.io/nvidia/jps/vst:1.3.0-24.08.1-aarch64
  • vst_config.json: defaulf file in vst image file, with following changes:
    “enable_ipc_path”: true,
    “ipc_socket_path”: “/tmp/”,
    …
    “gpu_indices” : [0],

I have a few questions:

  1. I expect a socket path is created in /tmp, but it seems there is no new file is created in /tmp when vst is running, and what socket name should it be?
  2. If DS microservice wants to use IPC, it should also map the local /tmp to docker container when running?
  3. (Not related to IPC - I might create new topic if necessary) can VST use the iGPU and build-in encoder/decoder of AGX Orin? they are all offline when VST microservice is running. I tried with “gpu_indices” : [0].

Thank you

-------- docker-compose.yml ---------------
services:
vst:
image: nvcr.io/nvidia/jps/vst:1.3.0-24.08.1-aarch64
user: “0:0”
runtime: nvidia
container_name: vst
entrypoint: sh -c ‘/opt/vst_release/launch_vst --debug-level 3 2>&1 | tee -a /opt/vst_release/webroot/log/vst.log’
environment:
- ADAPTOR=vst_rtsp
volumes:
- ./default_vst_config.json:/opt/vst_release/configs/vst_config.json
- /usr/lib/aarch64-linux-gnu/nvidia/:/usr/lib/aarch64-linux-gnu/tegra/
- /tmp/:/tmp/

----------- some logs info --------------------------------
[2025-3-17T8:28:30.694Z:281473481041344:fs_utils.cpp:265: createDir] Creating dir_path: /home/vst/vst_release/streamer_videos/
GPU INDEX: 0
Host HTTP port: 30000
Recorded Video Root: /opt/store/video
…
enable_grpc: 0
grpc_server_port: 50051
webrtc_in_audio_sender_max_bitrate: 128000
webrtc_in_video_degradation_preference: resolution
enable_websocket_pingpong: 0
websocket_keep_alive_ms: 5000
webrtc_in_video_sender_max_framerate: 30
gpu_indices: 0
Remote VST address:
Enable latency logging: 1
WebRTC Out Encode Fallback Mechanism (software/pass_through): software
Enable IPC Path: 1
IPC Socket Path: /tmp/
IPC Src Buffer Timestamp Copy: 1
IPC Src Connection Attempts: 5
IPC Src Connection Interval: 1000000
IPC Sink Buffer Timestamp Copy: 1
IPC Sink Buffer Copy: 1
Use Webrtc HW decoder: 1
Enable frame drop in Stream Recorder: 1
Max frame queue size in Stream Recorder: 16000000
analytic_server_address: http://:30080/emdx
webrtc_out_enc_quality_tuning (ultra_low_latency / low_latency / high_quality): ultra_low_latency
webrtc_out_enc_preset(ultra_fast / fast / slow): ultra_fast
enable drc: 1
Enable loopback multicast: 0

[2025-3-17T8:28:30.695Z:281473481041344:utils.cpp:1147: setRecvMaxSocketBufferSize] Setting OS’s socket receive max buffer size to = 1000000
sysctl: cannot stat /proc/sys/net/core/rmem_max: No such file or directory
[2025-3-17T8:28:30.698Z:281473481041344:utils.cpp:1158: setSendMaxSocketBufferSize] Setting OS’s socket send max buffer size to = 1000000
sysctl: cannot stat /proc/sys/net/core/wmem_max: No such file or directory

(gst-plugin-scanner:28): GStreamer-WARNING **: 08:28:31.200: adding type GstEvent multiple times
[2025-3-17T8:28:31.946Z:281473481041344:Main.cpp:184: main] g_main_loop is running now
[2025-3-17T8:28:31.949Z:281473481041344:redis_subscriber.cpp:180: redisInit] Redis Subscriber Connect failed. Exiting
[2025-3-17T8:28:31.949Z:281473481041344:server.cpp:290: VmsServer] #### VST version: 1.3.0-24.08.1 ####
[2025-3-17T8:28:31.949Z:281473481041344:utils.cpp:2213: detectGPU] ############## GPU ID DETECTED = 0 ##############
[2025-3-17T8:28:31.949Z:281473481041344:utils.cpp:2214: detectGPU] ############## GPU Device = ##############
[2025-3-17T8:28:31.949Z:281473481041344:server.cpp:83: initialize] Host Ip = 172.21.0.3
[2025-3-17T8:28:31.950Z:281473481041344:adaptor_loader.cpp:145: loadAdaptor] device_manager->name:vst_rtsp
[2025-3-17T8:28:31.950Z:281473481041344:adaptor_loader.cpp:161: loadAdaptor] Loading control adaptor: prebuilts/arch/rtsp_streams.so

5

I think the IPC feature on JPS isn’t full tested. Why you must use the feature in your project?

6

we want to minimize DS loader processing time: RTSP – VST – IPC – DS seems to be a shortcut compared to RTSP – VST — RTSP — DS.

7

Are you check the performance bottleneck of your use case? Can you share more on your use case?

8

we are exploring new features of JPS with provided samples to understand what it can and can’t do.