OSVR
Disable Client Side Prediction for HDK because of poor performances
Hi,
According to this doc the server side prediction is already enabled. The client side prediction is enabled too, so it can be a good idea to disable it by default for clients OR disable the server side prediction and keep the client side prediction enabled.
Today both are enabled resulting poor performances and a lag when you move your head.
I vote for disabling server-side prediction by default, because the client side actually knows how far ahead it needs to predict.
Except when in extended mode, where at the moment we don't have accurate timings, and so disabling server-side overall would be a regression. We already have separate configs for direct and extended mode, just need to make the direct mode ones not use server-side prediction - I'm not sure why this wasn't done before the client-side prediction config enabled by default was merged in to begin with, but we can do it now in any case.
Ryan
On Thu, Jun 23, 2016 at 7:44 AM Russell Taylor [email protected] wrote:
I vote for disabling server-side prediction by default, because the client side actually knows how far ahead it needs to predict.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/OSVR/OSVR-Core/issues/441#issuecomment-228038425, or mute the thread https://github.com/notifications/unsubscribe/AADuyQL0fcfchWk5a1tUKgLwt1J_vxo1ks5qOn--gaJpZM4I7yPV .
Ryan A. Pavlik, Ph.D. CTO - OSVR Platform Sensics, Inc. www.sensics.com
Latest news and blog posts (subscribe here http://sensics.com/subscribe-to-our-mailing-list/ to get weekly updates):
June 22: The three (and a half) configurations of eye trackers http://sensics.com/the-three-and-a-half-configurations-of-eye-trackers/
June 17: Notes from the Zero Latency free-roam VR gameplay http://sensics.com/notes-from-the-zero-latency-free-roam-vr-gameplay/
June 15: VRguy podcast: Jens Christensen, former CEO of Jaunt VR, on the video value chain in VR http://sensics.com/vrguy-podcast-episode-15-jens-christensen-on-360-and-3d-video-in-vr/
Client-side prediction can be done in extended mode, and the RenderManager config files are set up to do it. It just predicts for a fixed interval, which is the same thing that server side does. The config files for extended mode add a frame of prediction to the ones in direct mode. Note that we need separate prediction for each eye, so the client-side prediction will always be active. May as well have it do all the work.
Hi,
Any news about this issue? The client side prediction is still enabled resulting to bad performances for new users. I can prepare a pull request that fix that if all of you are agree with that.
@demonixis it sounds like there is consensus to make this change, so you could prepare a pull request.
To disable server-side prediction, add:
"aliases": {"/me/head":"/com_osvr_Multiserver/OSVRHackerDevKit0/semantic/hmd"}
to any HDK server config that uses a renderManagerConfig with (client-side) prediction enabled.
You'd want to edit all the HDK configs in the sample-configs directory which have a renderManagerConfig in them or reference an external renderManagerConfig with prediction enabled, like this one: https://github.com/OSVR/OSVR-Core/blob/master/apps/sample-configs/osvr_server_config.HDK13DirectMode.sample.json#L4
You wouldn't add it to this file because it doesn't use a renderManagerConfig: https://github.com/OSVR/OSVR-Core/blob/master/apps/sample-configs/osvr_server_config.autodetectonly.sample.json
Don't need to change this either because it already disables server-side prediction: https://github.com/OSVR/OSVR-Core/blob/master/apps/sample-configs/osvr_server_config.renderManager.HDKv2.0.direct.json#L96
It looks like the default osvr_server_config.json file is a copy of "sample-configs/osvr_server_config.HDK13DirectMode.sample.json" https://github.com/OSVR/OSVR-Core/blob/master/apps/CMakeLists.txt#L220 so that's probably the most important one in terms of changing the default user experience.
I'm not sure which config files are packaged as the defaults in Razer's installer.
@DuFF14 OK thanks. Are you ok if I send a pull request with these changes?
So the caveat is that we don't have the timing data in extended mode to do client-side prediction well, isn't that right? so we should actually be disabling client side prediction in extended mode rm configs...
If the timing data ever shows up, it will be on the client side, and there may be some APIs available to get it. We still know the relative eye timing. We also know the expected value (I think I've added a half frame of delay to the extended-mode files compared to the DirectMode ones). When timing data is not available, timing returns "0" as time to vsync, so it sort of auto-disables itself.
@rpavlik ok, do you want to keep server-side prediction for extended mode?