Are all sims truly created equal?

we've been told quite a number of times that all sims run on identical hardware. And for quite a long while i have had no reason to doubt this.

Some recent observations (and incredibly poor performance) in my home sim of lusk have begun to have me doubting this.

Now, before i get started i just want to reassure everyone im not writing this because lusk has a absolutely horrible average simfps (it *does* however... it never cracks 200 when empty, and with 3-4 people is easily in the 50-60's) That was merely the original clue that got me doing abit of snooping around and abit of thinking.

The basic crux of the problem im having is that lusk specifically, seems to knuckle under load *much* faster than the other servers do.

SimFPS aside.. the specific measure i'm going to be comparing is the 'run agents' timeslice of the server, versus the number of agents in the sim. All things being equal, more or less... You would expect 6 agents in sim X to run roughly the same amount of time as 6 agents in sim Y. Since they all run the same basic software package, and ostensably are running on the same hardware that at least would seem to make sense.

Lusk however, seems to flagrantly run afoul of this idea. Take any of the neighboring sims, say rizal.. and pack 5 avatars into it... the 'run agents' timeslice is about 0.5ms... or roughly about 0.1ms per active player in the region. Dore as well seems to run about the same time.. as does olive... roughly something on the order of 0.1ms per active player.

Then wander over to lusk, with 4-5 of your friends... you'll notice the 'run agents' timeslice is pushed up to somewhere between 2-3ms. Doing the math that gives you something on the order of .5ms per active player... or that lusk is spending roughly 5x the amount of real-world time processing each player as the neighboring sims are.

Now given that the software package each server runs is identical... (and that several lindens we have talked to weren't able to find anything 'wrong' with the server) the natural conclusion seems to be that somehow, in some way, lusk is inferior to the others.

well it *IS* one of the oldest sims.... dating well back into the beta runs....


I guess what im asking for is two fold... for other players, have you noticed other sims taking such wildly differing amounts of time per player on the 'run agents' timeslice?... and for the lindens... if the hardware on lusk is identical to the hardware of the surrounding servers... why is our poor lusk so *so* much slower at running everything?

eltee, I think there's something to what you're suggesting. I mean, pragmatically, it would be difficult to build the exact same system (motherboard, CPU, RAM) for eighteen months, anyway, just 'cause of the way the market works. I've been supposing the new sims being added are P4s of some flavor...

It's hard to be sure. You look at the max sim FPS on a sim when there are no avatars and you see some interesting trends that are, apparently, linked to in-world region. Compare the color/beach/moon sim FPS scores on the graphs to see what I mean.

I'm pleased to say I've seen Lusk hit its highest sim FPS in recent memory with 1.3 servercode-- 400 FPS. Yay!

well i think it probably best to reserve judgement till after people get a chance to go back in an turn all their stuff back on if the update hiccuped it etc.. also i don like comparing just simfps at least not for this thread.. since so many different things can affect it...

Yeah, true. I wish I could get at all of the stuff in the Simulator status under the client's ALT-1 popup with LSL calls. The only thing sim FPS seems to really show reliably is when a Simulator is hurting, but it's pretty obvious when it gets to that point.

This adds evidence to a suspicion I had as soon as I learned the distribution model for SL: that cycles are not going to be fairly distributed around the world, be that by sim or just be user experience or what-have-you.

The one-server-per-sim idea is straightforward, but it is rather simplistic seeming, compared to true invisible load-balancing as will exist in the MMORPG Wish, built on ICE distributed processing middleware.

With one server per sim, even if you have the same exact hardware for each server, cycles are wasted. Certain sims (new areas, undeveloped, watery, etc.) don't use their potential, while others strain under the load.

Linden Labs should look at some kind of add-on system, similar to Apple's XGrid, which distributes server load dynamically to relatively idle clients without requiring an entire middle-ware solution. It does require the servers to run the same software, but that's no big deal in a homogeneous environment like the SL server farm, I suspect.

One computer per sim is better than the alternative from a damage control point of view. There are things the average user can do in-world that can bog down any computer or cluster that Linden Labs can reasonably afford, so it's better to have one sim slow to a halt than to have the entire world stop.