Is one sim per server feasible ?

Don't know enough about hrdware to contribute on that aspect, but I have done my own little investigation which may be of interest.

I'm a somewhat cynical type (haven't you noticed yet) and I am also suffering badly in my home sim from crippling time dilation 24 hours a day caused by the nearby brothel (main reason I'm so grumpy at the moment).

So I decided to test whether two sims on the same server really were independent, ie not affecting each other's performance.

I found which sim I was sharing a server with. I set up simultaneous time dilation logging. I compared the two graphs, looking for correlation, ie does a bad dip in one cause a bad dip in the other.

First thing I found was that the huge time dilation hit once per hour when the backup is taken did indeed happen at the same time in both servers. Both are being backed up, it seems, at the same time.

But apart from that I could NOT see any correlation in the fine detail of the graphs. As my sim crawled along the time-dilation floor with its 36 campers and prostitutes, the other sim went it's own way independently. I felt ashamed of my mistrustful nature.

BUT oddly, and this is harder to be sure of, when my poor sim was paired with a very quiet empty one it seemed to cope better, and when it was paired with another busy one things seemed much worse (for both). This seemed quite definite, but is so unlikely that I think I must be deluding myself on inadequate data. I had about 8 different sim-pairings to go on.

It doesn't seem possible for there to be a general effect over the long-term without specific short term effects being visible as correlation in the graphs.

Putting this longer-term thing aside unless someone can suggest a mechanism, I have to conclude that the Lindens are right. The two colocated sims on a class 3 server do look truly independent in performance, based on the absence of short-term correlations in the time-dilation graphs under bad, bad conditions.

I wonder why they back them up simultaneously ? Wouldn't performance take less of a hit if this processor load was staggered ?

This would make sense if the server applications required resources unfairly. What I mean by that is, if one sim requires a huge amount of resources to deal with particle effects, automated scripts etc - it may ask for or require far more resources than is "fair" just to keep it running at normal speed. Sort of like, the more you ask for the more you get, the less there is to go around.

That may sound really untechnical, and it is, I haven't really thought of how that may work. The reason I suspect it though, is because I saw exactly the same thing on my own C2D machine. I ran two intensive applications, one more demanding than the other, which then seriously impacted the other application, reducing performance to half. There was no way the bus was saturated, it may not be hyper transport, but it is quad pumped. All the while, the more intensive program ran as if the other one wasn't there. Affinity was different, priority was the same.

I would suspect poorly optimized code over insufficient hardware. I was really tempted to get myself an island. I would even love to set up my own community, but would not want to have to deal with all these issues. If along came a resource hogging club and messed up what I had created, I would not be best pleased.

We need something more than just stating this. We need to get evidence to see if it is true,

I believed it was true. I set out to get convincing evidence. I logged dilation on partnered sims simultaneously and looked at the graphs. Apart from the synchronised backups every hour, I could see no evidence of any interaction from minute to minute at all.

I was amazed.

Try it for yourself.

Use this site to find who you are paired with. Be patient and keep checking, it needs someone with an attachment to accidentally visit the right sims and update his database. Help everyone by wearing the attachment yourself.

Once you know who you are paired with, act quickly - the pairing may only last a day or two.

The second phase is to log the time dilation. For this you need
this other site . If the servers of interest to you don't happen to be being logged, you'll need to IM Chromal for a logger (copyable). Put one in each sim you want to log, yours and the other. Only you can decide whether you ask permission, or hide it away. It probably only needs to be there for a few days. People can be touchy. Permission is best, but you can't waste days waiting for it, or your chance will be gone. Some non-techy people may think you are trying to steal their soul.

A day later you have your two graphs, on Chromal's site. Snapshot them and compare.

Remember to keep visiting each to be sure they are still paired. Either look in "Help/About Secondlife" whilst in the sim, or get the Neighbours attachment and wear it to update the site.

Amazing. The graphs show no sign of them interacting. Yet, like me, maybe your preconceptions are so strong, you still try to deny it. Maybe there is a hidden, overall effect, not apparent in the graphs ?

But I just can't think how such a thing could be. One graph falls through the floor. The other doesn't flicker.

It's no good. I think I have to admit it. Lindens are right. At least the two on a class 3, they really are not interacting.


If you would rather not be forced to admit this - don't do the test...........

Nobody answered my question.

I'll take a look tonight and post more on this after work.

I think Ellie is talking about making the comparisons on Class 3 servers, I'm talking about making them on Class 4 servers. 1 sim should not effect the other on a Class 3 server, or atleast the impact of what one sim does should be minimal to the other except when writing simstate. On a Class 4 server, which is where my point is, this is not the case. Performance is worse, since it's 2 sims per processor.

Yes absolutely, Rhyph. My tests were on Class 3. Is it established and admitted that on class 4 a cpu is shared ?

Certainly the method I used (Chromal's dilation loggers and website) can provide solid evidence if you are right. Can we identify a couple of sims colocated on a class 4 which happen to be being logged right now, and, and go have a look ?

Otherwise all we have to do is find a couple of colocated sims under stress, drop in Chromal's loggers, and we could settle this issue of alleged interference on a class 4 for once and for all. Hmm - guess we need to identify three of the four sims to be certain of getting two sharing a cpu (if they do) - or examine several pairs. Take a bit longer but perfectly practical.

Here are the tools for such an investigation :

Max Case's website to find sim colocations
Chromal Brodsky's website to log and graph sim dilations