ReactionGrid hosted a 100-200 person event for NetGenEd. The experiences and lessons learned how to host big events technically and socially.
The Good:
The good for ReactionGrid is this event challenged us to try and
manage a large event on OpenSim technology. We did testing we
normally would have not which proved 30-50 avatars could be logged
onto a single for 30 to 60 minutes with little trouble. This led
use to estimate we could handle 4 sims on our grid loaded to 25
avatars each for an event. Unfortunately this estimate did not
include many human factors and the inescapable Murphy's law.
We were able to prove our streaming audio system worked pretty
good grid wide with only a couple of dropped audio sections. Our
use of YouTube video streaming of their .MP4 worked fine when sims
were not loaded but try and change parcel media settings with 10-20
people on a sim, it simply wouldn't work until of course later in
the event when not needed. But this let us know that you can stream
full size video onto a grid using YouTube.
We had estimated likely the event would be 50-100 users, this
turned out to be accurate. However we had more users than we
thought who had not dressed beforehand and this "edit appearance"
group event was something we had not allocated enough power for and
the grid strained because of it. Again the upside is we will know
best practices for our next event.
We were able to record the audio stream of presenters, we were
able to capture some video, and we were able to interview Don
Tapscott the event speaker. In addition we were able to
announce the NetGenEd winners to people worldwide.

The Bad:
ReactionGrid tested like mad pre-event so why did we only get
20-30 users stable for the event? Especially after successful 40-50
users tests? We made some assumptions which did not pan out.
We assumed we could get an early start on the user staging process.
A power outage at one of the schools ruined this plan mid-stream,
aka Murphy's law. We also assumed our avatar tests simulated
properly what a "real" avatar would do on a sim. However when we
tried to enforce the same "no fly-get seated asap" rule we
practiced with it didn't work on any level with real users.
For this reason login was chaos. Once the kids, teachers and
others started to pile in at the same time we could no longer
handle the load. Our best was 30 or so avatars attending the
event far short of the 100 minimum we had hoped to accomodate.
The Ugly:
The ugly was simply the failure of not accomodating all 50-100
participants. We achieved only about 30% of this target and
those had a rocky time. I call this "The Ugly" not because we
didn't put real effort into trying to make this work smoothly but
because we didn't do enough real practice and testing before trying
such an event with live users under a very narrow event time
window. We simply cannot expect all users to do as we think
they will and as such instead of entire classes logging in we
should have had a few representative kids and teachers from the
schools participate until we properly tested more. The Ugly
happened simply due to not enough pre-event testing and
practicing.
On the upside, we won't ever be able to learn how to accomodate
100-200 users if we don't try, and possibly fail, to do it.
This event was invaluable to us learning how to make this work and
the next large event will benefit from today.
Summary:
So did we learn today OpenSim cannot handle large events? No, we
simply tried to go for the gold and allow all avatars to be within
sight of the stage for presenters. Had we gone with a
multigrid event with chat bridge we likely would have come much
closer to our 50-100 concurrent users goal. We also learned
we need not 1 hour before events to stage users but 3-4 hours
especially when dealing with people around the world under a very
tight window of time.
This is not a reflection on the attendees merely a best practice
forming for large events. We achieved 25-35% of our
estimated true attendance over the events course, we streamed to
dozens reliably and we recorded video which means we can now setup
the event sim to show the 3 videos that were winners 24/7 along
with the video capture and audio capture of the event being
available on the sim for anyone to see and hear. In addition to
live events the real payback of virtual worlds is persistent event
elements long after the event which usually accounts for a larger
number of views than did the live event.
I want to thank the attendees, my ReactionGrid team and
especially Trevor
Meister & Vicki Davis the event organizers.
While we cannot call today's event a qualified success we did get
enough materials to recreate it persistently & we are very
happy to have hit 20-30 users on one grid node as opposed to
the 6 max we got when we started ReactionGrid. Things can
only improve from here if we keep hammering away as we are
dedicated to doing!
Kyle "G" aka Dr_Manhattan
We recommend that you discuss this article on Think, but if you really want to you can leave a comment right here as well: