The majority of the notes in this update are taken from the TPV Developer meeting held on Friday, May 19th, 2017. The video of that meeting is embedded at the end of this update, my thanks as always to North for recording and providing it. Timestamps in the text below will open the video in a separate window at the relevant point for those wishing to listen to the discussions.
Server Deployments Re-cap
- There was no Main (SLS) channel deployment or restart on Tuesday, May 16th.
- On Wednesday, May 17th, the three RC channels were updated as follows:
[1:00] The Voice RC viewer has an elevated crash rate, and the Lab currently haven’t determined why.
The Maintenance RC viewer updated to version 188.8.131.526444 on Thursday, May 18th. This viewer currently has a lower crash rate that the other RC viewers (although it has not been out that long), so might be a candidate for promotion. I have an overview of this viewer for those interested.
[2:23] The last major functional addition for the 64-bit Alex Ivy viewer is currently with the Lab’s QA. If all goes well, a further project viewer update should arrive in week #21 (commencing Monday, 22nd May).
This introduces a new executable to the viewer – SL Launcher – which runs an update check at start-up. If there is a new version of the viewer available, the Launcher manages the download and installation – including ensuring Windows users get the right version for their operating system (32-, or 64-bit). If there is no new version to install, or once the viewer installation has completed, the Launcher will launch the viewer as a child process, and will shut down when the viewer exits at the end of a session.
The plan is to move the crash data capture package to the Launcher in the future, which will give full end-to-end monitoring of the viewer in the event of a crash.
360 Snapshot Viewer
[6:07] The work on the 360 snapshot viewer is once again progressing. A new library has been added, which provides the appropriate meta data so that websites supporting 360-degree viewing can correctly such images taken by the viewer on upload, eliminating the need to process them separately via the web service currently supplied by the Lab.
Region Crossing Hand-off / Caps Router Issues
[7:43] Fantasy Faire experienced very high levels of region crossing hand off problems with avatars trying to move between the various regions. A similar issue has surfaced at the just-opened Home and Garden Expo.
While it issue isn’t new, the Lab found a cause is the Caps Router running out of connections due to the number of avatars it is attempting to serve. New monitoring has been put in place which will determine how many connections the Caps Router is using, and when it is approaching its limits. The data gathered will be used to help better determine how many connections are needed, allowing the Lab to adjust the number supported.
This work is going to be carried out incrementally, starting with an initial RC deployment in week #21 containing conservative adjustments in the hope of avoiding creating additional bottlenecks in changing things too radically at one time. However, the hope is that the changes will in time result in two improvements:
- It could result in an increase in the number of avatars a region can comfortably support
- As this is an issue at the SERVER level (not the simulator), the changes should help reduce people on regions with few avatars on them experiencing issues as a result of the region being hosted on the same server as one (or more) regions with a lot of avatars on them.
As a result of understanding the problem, the Lab was aware the issue was impacting the Home and Garden Expo even before it had been reported.
Unsuccessful Teleports Impacting Region Performance
[14:20] During investigations into the region issues at Fantasy Faire, the Lab noted that a simulator running a busy region has to carry out a lot of work to determine whether or not someone can teleport into it, which can degrade overall simulator performance.
To combat this, the Lab is going to change the teleport re-try throttle following a failed TP. As viewer-initiated teleports are already somewhat throttled, the change should not affect them. However, it will likely mean that the very rapid retry TP HUDs (aka “TP hammers”) will break or degrade in their performance unless adjusted.
The hope is that by reducing the load placed on a simulator as it tries to deal with too rapid a succession of TP requests which cannot be granted as the region is full, overall performance will be improved and those already in the region will enjoy a better experience.
This change should be appearing in a server RC update soon.
Additional comments on teleport failures:
- A queuing system will not be added, as this is deemed to be too difficult to implement and manage.
- There is no relationship between the size of an avatar’s inventory and the frequency with which that avatar may experience teleport failures. However, the amount of items attached to an avatar, the scripts they are running, etc.
- The Lab can monitor teleport failures in real-time.
Automatic Additional Logging after Region Crashes
[29:27] It was asked if additional logging could be automatically enabled on a region crash. This is something that cannot be done, and Oz’s belief is that doing so would result in an additional load on the simulator during recovery, and so not be a good idea.
Avatar and Object Rendering Cost Investigations
[31:00] The Lab is continuing work in reviewing the rendering cost calculations for in-world objects and avatars, work I first reported on in September 2016. However, the numbers aren’t at a point where any adjustments can be made to the calculations.
Oz Linden marked his seventh anniversary at the Lab this week – so a belated happy rezday to him! Some of us can likely remember his 2010 appearance at the SLCC, when Esbee Linden introduced him to the audience in Boston 🙂 .