1

COMSLIVE Project: Wonderland Stress Test 2: 26/2/10

Aims

Main aim: Find any problems that may arise from multiple users (>10) connecting to a Wonderland system, that are not present when few users are connected (<5).

Test how many low-res avatars (texture files manually scaled from 2k to 0.5k) can be present before ‘OutOfMemoryException’ thrown. Files are now 15 times smaller as in the last stress test only 8 avatars could be loaded, so technically there should be enough memory for over 100 of the smaller avatars.

Test X11 Apps; The major bottleneck here should be the network, as each client will need to be fed a video feed from the Shared App server (SAS).

Formally test the in-world audio; Over LAN the audio has proved very successful, however it has not been tested with more than 4 people at any given time. In the past audioquality has been varied over the internet, and has not worked perfectly.

Method

Node 1will be hosting the virtual ward (approx 75 items), using OWL version Preview 3.

Node 2will be hosting the application server.

Make sure to use a headset, or turn off audio, to avoid echoes/background noise. The ‘[‘ key stops the microphone from sending audio.

We will communicate during the test using OWL’s chat box.

  1. We will set up 11 laptops in the library, and log them into Node 1.
  2. We will thenswitch to the reduced Evolver avatars checking for any problems that were encountered in the previous test.
  3. One user will manipulate one of the web browsers in the ward, whilst other users will watch for changes.
  4. The VCC (Virtual Case Creator) will be loaded into a web browser, and the handover video played.
  5. Attempt to access a video site (i.e. BBC iPlayer or YouTube) and assess playback quality.
  6. In order to check VoIP lag, one laptop will be set up with a microphone whilst the others will be set as speakers. One of the team will speak into the microphone and observe how the time delays to the speakers.

Client Specifications

Toshiba Laptop:

Processor / Intel Core Duo CPU T9400@ 2.53GHz 2.53GHz
Installed RAM / 3.00GB
Sys Type / 32-bit Operating sys
Operating sys / Win 7
Video Card / ATI Mobility Radeon HD 3400 Series

Results

Reducing the size of the avatars resolved the problem of running out of memory. In this test13 clients with avatars were logged in, and the platform still ran smoothly on each client.

In the ward space with 13 avatars the Toshiba laptops were able to draw 30-50 fps.

Static websites were no problem for Open Wonderland, with the updates appearing for all clients showing little lag.

In one browser an Adobe Flash ECG graph played jerkily but it did play.

The VCC handover video appeared jerky, and of course lacked sound. However it did technically play.

The SAS was unable to keep up with a YouTube video, a couple of clients managed to keep up for a moment (albeit at <1 fps), but most simply showed a static screen. After a minute or so the all the clients were unusable, and the application and world servers had to be restarted.

The VoIP was of good and consistent quality across all the laptops. The lag was comparable to the lag of a mobile phone, so only noticeable when you can hear both the original and transmitted sound.

The SAS; Node 2’s maximum utilisation was about 50Mbs-1, and as we ran the server till it crashed, it could be assumed that this is as much as it can provide. It is unknown at this time where that limit comes from.

The amount of data sent out by the SAS during the test was 21GB, which would appear to be quite high considering that the test was conducted for less than two hours. However it must be taken into consideration the amount of video that was sent to several clients, making this amount reasonable.

Figure 1: The bandwidth graph for SAS.

The world node’s network stats are similar to the last test but with a lower network utilisation max, which will be as a result of all users logging on at the same time, which did not occur in this test.

Figure 2: The bandwidth graph for Node 1.

Analysis

For now the problem with the Evolver avatars has been solved. Changing the texture size from 2048 to 512 reduced the amount of memory each one used and stopped the client from crashing when more than 8 were loaded.

Anything that is played over a timeline, such as the ECG graph or a video, will appear jerky in an X app when viewed by multiple people. This is seems to be due to increased pressure on network.

Over the LAN the VoIP audio system works fine, with no noticeable difference between one user and >10 users. The problems will arise when users connect from outside of the LAN, due to (partially)blocked connections and variable ping and bandwidth.

Client Bandwidth:

Inbound (upload bandwidth) / Outbound (download bandwidth used)
WhiteRabbit / 0.3484615 Mbs-1
167.26 MB/min / 0.6630769 Mbs-1
318.28 MB/min
Incl. SAS / 0.4653846 Mbs-1
223.38 MB/min / 4.4392307 Mbs-1
2130.83 MB/min