1. Purpose of the Document

• XRCLOUD is a service that enables multiple users to communicate through synchronized virtual reality spaces with various interactions, including voice. There is a need to limit the number of users per environment for optimal performance.
• The objective is to assess the current status and identify areas for improvement.
• The goal is to adjust the number of voice connections based on distance, similar to large-scale online games.

2. Current Issues

• The biggest problem currently is the real-time transmission of voice:

  • XRCLOUD uses Mediasoup as the WebRTC server and employs the SFU method.

  • Utilizing the SFU approach of Mediasoup’s WebRTC, the relay server ends up using network bandwidth proportional to the square of the number of users (n2).

    (n2)(n^2)

  • In a single XRCLOUD space, users synchronize data using WebRTC technology through the SFU method. This causes the WebRTC server to send identical data to all users, utilizing outbound bandwidth. If every user turns on their microphone and sends data, the number of connections on the WebRTC server grows quadratically, leading to 625 connections for 25 users, consuming the bandwidth of the WebRTC (Dialog) server. The most problematic aspect is the voice signal.

  • There are also some issues on the Android side of WebRTC.

• Avatar Synchronization:
  • The synchronization of all avatar movements is attempted on a frame-by-frame basis.
  • Although the data size is small and does not cause a bottleneck, it sends data too frequently even if not in VR, which needs adjustment.
• Video Synchronization:
  • Videos within a room are synchronized with users.
  • During early 2023 tests with multiple users, poor bandwidth from one user affected the entire room, causing freezes.
  • This issue hasn't been extensively tested and requires further investigation as it is not of the highest priority

3. Current Status of Mass Connections

1) Difficulty Ensuring Stability for Over 120 Users

• On June 14, 2024, when scaling up to 8 cores on XRCLOUD, the 3D screen freezes with up to 175 users connected, and connections start dropping from 252 users. This requires investigation, as it's not a server performance issue but likely a code-related problem. In tests with 1000 users, disconnections start at 125 users, with massive drops at 140 users.

2) Goal: 1,000 Users

• The NIPA Education Platform project aims for 1,000 simultaneous connections and requires objective testing for up to 1,000 users.

3) Successful Testing with 1,000 Users Split into 10 Rooms of 100 Each

• Tested 10 rooms with 100 users each, confirming normal operation.
• With 32 cores, 60% usage, and 3.5GB of memory, it was sufficient. It is believed it would have also worked with 20 cores.

4) Performance Observations

• Resource Consumption: 0.02 cores and 2 MB of memory and network bandwidth per user.

  • Network usage does not increase when rooms are separated.
  • The server’s network bandwidth showed no significant difference from testing with 100 users in a single room, indicating that simultaneous connections are dependent on the number of users per room (data transmission is may be time-division based as room count increases?).

• While there are ways to address the concurrency issue, the following operational method is also available:

  Operation Plan for Events with 1000 Participants Across 11 Rooms