Wednesday, May 02, 2007

« Rik's Picks for May 2-6 | Main | OPEN FORUM: HIGHLIGHTS FROM CORY LINDEN'S TOWN HALL »

THE WORLD AT A PLATEAU

Tat_presents_concurrency_001

I had an inkling that meaningful population growth in Second Life had gone flat, these last couple months, so I contacted New World Notes demographics expert Tateru Nino to do the research and to draw up a chart of maximum concurrency over the last 150 days. 

"All the signs seem to indicate that the grid basically hit capacity," Tateru confirms. "As concurrency climbs each day, performance decreases, lowering it past the threshold of what Residents find acceptable."

Maxcon150800

The plateau began in mid-April, when the total number of Residents in-world during peak hours reached 39,000.  And while growth in new sign-ups has continued to increase at a rapid clip since, surpassing 6 million total just a couple days ago, the steady growth of maximum concurrency finally began to slow on February 25, the foothill of the plateau; from then on, it's been idling between 36-39K.  "Since that time," says Tateru, "we have spent 88.86 hours at or above that level, as reported by the Linden Lab data feeds."

When Tateru last concerned herself over demographics in this space, it's worth pointing out, she was writing about the scalability trouble from two million accounts.  The world had reached that point last December.  Four months later, four million more accounts created.  This is irony at its most brutal, for during a period when some were openly speculating that Second Life was over-hyped because it had few active participants, the world was actually being throttled by mass enthusiasm.

"Middle of the day is worst," Tateru adds, "Say 10am to 3pm [PST]."  That is also the time frame when European and American activity intersect most, and Second Life's in-world population is at its most international.  And in this way, a brotherhood of nations come together every day in a single place, to create and to be convivial, and to clobber the living bejesus out of several thousand servers.

Tat_presents_concurrency_004

Which is what was meant by "meaningful population growth", for while several million can enter the world at any time, in theory, the true choke point at the moment is just 40,000 people wide.  By one industry rule of thumb, total active users are calculated by multiplying the maximum concurrency by ten-- meaning, on that view, 400,000 active Residents.  (And surely several hundred thousand more who might join them, if they could endure the aggravation.)

"Like any good or service," Tateru speculates, "[SL] has passed the sweet-spot of quality-delivered for time/money invested. As that gap widens, people increasingly decide to shorten their time in-world, or spend their time doing other things. Essentially it doesn't necessarily indicate a drop in active users (although that seems to have plateaued as well for the time being) but more of a drop in user hours. Minimum concurrency has also leveled out significantly...

"[P]ractical growth in activity seems to have stalled," she concludes, "because the platform cannot presently scale to meet the user-experience demands being placed on it."

I asked about this plateau effect with the Lindens, of course, and will run any reply in an update.  Presumably, it'll be a topic of conversation at Cory Linden's town hall talk tomorrow; perhaps an upgrade to new server facilities will stanch the bleeding.

As it happens, at any rate, I asked Tat to present her findings in-world, but that took some doing, because I couldn't upload the chart graphic (it displayed as a blank gray texture), and when I tried to teleport over to her location, the severs choked on the request several times.  And so shooting several screenshots took far longer than it normally would.

"You begin to see what I was referring to about degraded quality of experience, yes?" She inquires sweetly. "You feel our pain, yes?"

"Uh, yeah."

Tateru Nino nearly claps her hands with glee.  "At last!"

Chart, to be sure, by Tateru.  See more demographic studies and other musings at Ms. Nino's blog.

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d8341bf74053ef00d834fe291053ef

Listed below are links to weblogs that reference THE WORLD AT A PLATEAU:

» Growth has slowed down from Bart in SL
When you take a closer look at the key metrics of April or make diagrams out of them like here by Tateru Nine, who has been interviewed here about the growth, you can only come to one conclusion: growth is still there, but it has stalled. What are the ... [Read More]

Comments

One thing I don't get. If I count each day on the graph as a little square - and that seems correct because of the 7 day peaks I get if I do - the graph ends around May 28th...

Wow. I hope that spatial VOIP and a new physics engine and sculpted prims don't increase the network load or computational complexity.

Wait...

If it is not providing more features than are currently provided, the new physics engine might actually use fewer resources than the current one - but I wouldn't put money on that.

Spatial VOIP and sculpted prims don't add (we are told) any measurable load at the server end.

Right on schedule. Thanks for the charts.

Tateru: "If it is not providing more features than are currently provided, the new physics engine might actually use fewer resources than the current one - but I wouldn't put money on that."

I'd really hate to see the mob react to Havok 4 with no new features enabled.

"Spatial VOIP and sculpted prims don't add (we are told) any measurable load at the server end."

Oh, pshaw.

Nah, actually, I can believe that with sculpties. Server end, sculpties are just another prim - the only additional load is loading the mapping texture, and those can't be larger than 64x64 anyway.

Agree. Sculpted prims are - according to what I recall reading - limited such that they do not exceed the current prims in load. Hence (as Aliasi mentions) the small sculpt texture.

I'm not sure how the VoIP system work, so no clue how it affects server load.

And wrt Havok 4, one would hope that the code has gotten more efficient as well as better. We are, after all, comparing Havok 1 to Havok 4. I suspect for a lot of SLers, a ~better~ physics implementation is all the "new" feature they need. Now if only it would migrate from the test labs over to Beta...

I see the grid broke through that peak today (6 May) to set a new record at 41K-42K.

At long last, yes. And again 7 May. Looks like a new plateau at this stage, though it's a bit early to call.

Is LL slowly raising the bar as changes are implemented to the backend?

Post a comment

If you have a TypeKey or TypePad account, please Sign In