The Inaugral Sinewave Pub Quiz on OSGrid.org

The initial pub quiz (pictured above) was a success, we got a lot of people to show up - many of whom who have never been onto OpenSim or OSGrid before. We’re so happy with the results that we’re going to go ahead and schedule another one - we’re aiming at Sunday, the 19th 26th of October at 9PM GMT (1PM PST) with a Halloween theme. (Sorry, typo! - edit)

The structure of this event was a three part quiz - a picture round, including naming hollywood movies from a still image (some tougher than others), a music round (name the track from a small clip) and finally a ‘google fu’ round with recent history. There were 24 questions in total - our two first place winners getting 19 of them right (and sharing the $500 in prize money).

Q&A was handled by our bot-in-residence Chinzy Quizmaster running the Sinewave Quizbot code, some suggestions for improvements we’ve added to the development roadmap.

We hope to see everyone who participated at the next one - we had a lot of fun running it and want to make this a regular occurance.

As Promised - The Technical Writeup

Technically the Pub Quiz was overall a success - we got a lot of interesting data out of the test (and a 17mb log file), including the fact that OpenSim seems to scale pretty damn well under a tightly controlled environment. We also managed to isolate and fix three critical bugs with OpenSim which have been contributed back to OpenSim already. In addition we have some information about how OpenSim is handling packets internally which I’ll be interpreting this week to try improve general performance with.

We also learnt some other interesting things about running these kinds of profiled events in future which might be useful to other people

• JetBrain’s dotTrace profiler doesn’t work on OpenSim very well - infact the chances of getting a  “completely random MS Visual C Runtime error” while under their profiler goes up about infinity% (ie, from zero to once every minute). We’re going to try another profiler next time around. A slight shame since we wanted to do a memory profile next time vs re-doing the performance one. That being said, if anything’s going to kill a profiler - we’re it, the Mono profilers dont even start when profiling OpenSim.
• For the most part OpenSim seems to behave pretty well these days, however we do need to get some kind of mechanism inplace to handle when things do occasionally crash to somehow get users into another backup region. EventQueueGet makes this now rather difficult, but it might be possible to do still somehow.
• We’ve still got “Grey Avatar Syndrome” showing up on a number of avatars - however we think these are being caused by simply not having an avatar equipped, or having an avatar that was built with local assets enabled. We definetely need to work on providing a nice default avatar in these cases as the simulator should be able to tell the avatar is using a inaccessible asset.
• Windows 2003 still performs significantly better than Linux/Mono it seems. Our servers didnt hit above 5% CPU usage for the entire event, except during one stack heap (which was fixed). Memory use fluctuated a little bit, but for the most part was sitting comfortably around 230mb of RAM plus a little bit of page.