WolfBaginski Bearsfoot

I'm pretty sure that nobody outside Linden Labs knows how the sim-crossing code works. I wometime wonder if anyone on the inside does.

For Vehicles, at least, the sim-crossing process is complicated. If I read things right, and this is scattered through a whole range of JIRA-entries and old Office Hour transcripts, there's a forced unsit at the start , the vehicle and the passengers are transferred independently, and then they automatically sit back on the vehicle. Since vehicles can have multiple passengers, it does make some sense, but it also explains some of the things that go wrong. I've had my AV dumped at 0,0,0 while the vehicle is eventually returned from some parcel several sims distance. Mesh vehicles seem to take longer to make the passage, which may be part of the problem. I do sometimes wonder if people have got the trick of designing good SL meshes. At the rendering stage, everything is triangles, always has been, but getting the mesh from one computer to another seems to be where things choke for mesh.

I am not allowed to see your Bug report, only the older JIRA, but I would, were I trying to fix this, want to know whether the big change to Physics has affected things.

It does sort of fit with what I heard a few weeks ago, that there was a later version of the Havok code to work through the process. I reckon they need a better-planned layout of test sims. But what do I know: I chose woodwork rather than Latin at school.

As of today, 20th September 2012, sim crossing appears to be tolerable. A test flight with the new THI Avro Lancaster was made, totalling 20.15 km in distance which requires at least 80 transfers from one region to the next. Average speed was 15 m/s, at altutudes generally in excess of 500m. No sim crossing failed, though most showed noticeable rubber-banding. The flight started at about 23:50 SLT on the 19th and continued until about 00:15 on the 20th. In that time there were three problematic events: At about 00:06 I was ejected from a parcel without warning by some sort of Land Security Orb. There was no warning message, and no data logged which identified identified the controller or the Orb or the location of the land protected. There was no obviously visible skybox. At about 00:11 and at about 00:16 the aircraft was de-rezzed and returned to me with a "Parcel Full" message. It is my understanding that this is a known, and very old, bug. Is it worth wasting anyone's time with a JIRA report? On the whole, I can live with this sort of performance. Road vehicles are not so good, with less reliable control and rubber-band events mid-region. I suppose this is a conequence of their dependence on collisions with the ground/Navmesh. There is room for improvement, but this is hugely better than what I saw between the start of Pathfinding testing and the fixing of the Bandwidth bug. Is there a list of known problems? I know that banlines at a region edge are not visible from the neighbouring region, but what else?

So it looks like one on't cross beams gone owt askew on treadle. Pardon?

I didn't see any of the hassles that have been reported, but I have taken to doing something else on Tuesdays and Wednesdays, the rollouts being so disruptive. That's partly the timezone I am in. I did do a bit of test flying of a vehicle, after the rollout was completed, Sim-crossing didn't seem any worse. I managed to make quite a long flight before things went wrong. And then it was one of those no-apparent-reason failures.

The roll-outs this week were slow. 8 hours for 70% of the Grid, 4 hours for 30*. I've been led to understand that they are working on it. I think there have been some improvements.

For the past month or so, the servers have been afflicted by a bug which appeared to send huge amounts of unwanted data to viewers. It was finally fixed this week, rolling out to all servers. Since then I have seen a general improvement in many aspects of SL, such as texture loading and sim crossing, and I infer that the extra traffic was enough to be pushing the Linden Labs internal network to the limits, and dragging down performance. Hindsight is wonderful, but I do wonder why there was not a rollback if the extra traffic was having those effects. The whole business, I reckon, was entangled with the Pathfinding and associated new physics, and i cannot rid myself of the thought that Pointy-Haired Linden was pushing hard to get that deployed, without adequate allowance for the size of the job, compared to the routine code changes. The new shiny of the last year has had an erratic delivery. The hassles climaxed with the new Physics needed to support pathfinding, and the rather disruptive test phase. SVC-8206 was my response to that. I'll settle for a few months of peace, quiet, and bugfixes. Take the time to catch up on the backlog. And, Oskar, if any of your minions has a bright idea for something new, shut them up. Let's see how the JIRA changes work out before anyone tries to do anything too clever.

I don't recall the last time the whole Grid was effectively running on the same server version. It's an odd feeling. I'm pretty sure the bandwidth problem is over. I suspect, from the improvments in such things as general lag and sim-crossing, that there was being enough traffic generated by the bug to interfere with the desired traffic between Linden Labs servers and the customers. Sim crossing is still far from the best I have seen, but there has been a big improvement in reliability. I wouldn't complain if there wasn't new code tomorrow. I sometimes wonder how you can tell what's worked, without some stability in the server code.

It's my understanding that the Lindens will be able to see and comment on a JIRA entry, so posting the link here will be useful to them. The rest of us will be locked out, and that might not be useful. It's possible that threads such as this one will mutate into multiple-JIRA comment threads, and I doubt it will work well.

And maybe I spoke too soon. Either they forgot to update the Grid Status, or the rollout has taken 8 hours, and counting.

It's good to hear that there's finally some progress on this. The long restart delay, an hour or more, makes the whole roll-out process that much more disruptive. The numbers are not as good as they used to be, but they are moving in the right direction.

August was the date for the bug report. Not everybody notices these things, which is why the Linden silence was disturbing. The worst effects could be avoided, if you knew.

The claim that Lindens were "shutting down server development" merely because they shut down the *view* of multiple bug reports and removed interactivity was never logical or justified. You didn't attack that as off topic, however. You have noticed that is in the title for this topic? Unlike your ranting against open source. Which is even more irrelevant when the Linden server code is not open source. Do you know you're posting in "Second Life Server"? You want to shift your inane ranting over to "Second Life Viewer", where the issues of open source code are relevant. Though I have seen plausible arguments that the way Linden Labs runs viewer development is not a particularly good fit with the open source definitions usually uses. Open Source is about collaboration. There is a two-way flow of information and code. And, if you want to turn this into an argument about political philosophy, I've read both Ayn Rand and Kropotkin. I can point you at examples of successful anarcho-syndicalist corporations. All you seem to be doing is extolling the virtues of greed. Which, to be honest, I find a repulsive and self-diminishing basis for an argument. Why should I even bother, when I have already seen how little attention you pay to the words in front of you.

I have just been having similar problems with Firestorm. It may have been a particular attachment, but it did not give problems previously. It's not been an every time problem. The Firestorm support team do not recommend clearing cache unless there is clear evidence of a texture problem. Can't really offer any help other than "keep trying"

Can we take this argument out of this thread, so that those of us interested in what's happening to the servers stand a chance of finding out something useful. I am so not surprised at hearing that you were banned from the JIRA.

I've seen a common pattern with vehicles, not just with the new code. You start a sim crossing, and the view shows a predicted position. The minimap shows you moving. But the location data shows you still in the original sim, with out-of-range position data, and control inputs to the vehicle are dead. It's quite common, and when the sim crossing completes you may get a sudden position and vector change. So, stuck in Sim A, apparently moving through Sim B, what happens when your poisition enters Sim C? Well, you may see Sim C is the viewer, in glorious 3D, but it never appears on the mini-map and when you hit that boundary, there's a viewer crash. Eventually the vehicle is returned, and it's sometimes hard to tell which return was which sim-crossing failure. But, a couple of times. it seems that the vehicle went on in a straight line for 10 or 20 regions before it hit a full parcel or other no-access boundary. It looks to me that the time available to complete the sim crossing depends on the speed and on the distance to the next sim boundary. Which means those near-corner sim-crossings by roads have less margin. There's a terrible junction in the snowlands where three roads meet almost on top of a sim corner. I don't recall one of the four region names, but if you're sailing they have that happen twice, to the north-west of Dex. Thallipoli is one of the region names, or something very like it. From general observations, sim crossings are taking longer to resolve than in the past, it's not something specific to any new code. The rubber-banding is more obvious, even when walking. There may be a correlation between increased delays and increased user concurrency, but I don't have the information to test that. In the end, all I can say is avoid, where possible, sim-crossing close to corners, and go as slowly as possible. The racing sailors must be going through hell.

I haven't seen similar circumstance while on other channels. But get about twenty people in a region, and a few physical vehicles, and things went gnarly. At some point, there's no slack in the system any more. And my gut feeling, based on past experience, is that this is happening at a lower load level. And there's plenty of room to argue about what is "expected behaviour". Since I haven't been prompted to do detailed Physics frame-rate tests in the last three years, I don't know how unusual these events might be. I don't know if I have a reliable measurement from the viewer stats window. I later tested, Le Tigre release channel, a gun that fired bullets doing Linden damage on collision. Normal fps figures for the sim were 43-44 fps. The fps drop varied, a lot, which is why I wonder about the reliability of the Stats window, but firing a shotgun sometimes caused a drop to 21 fps. That's a case of rezzing and moving a half-dozen simple objects, but, last year I was in a zombie-maze with a shotgun, and while there was a lot of fuss about script load, this sort of thing might have had the same symptoms. It's an anecdotal gut reaction: the RC code has bust something. I've submitted a bug report. But they only place where people are going to find out about it, and respond to requests for tests, is here. Good luck, Oskar, I think you're going to need it.

Magnum RC channel: I have seen Physics frame rates plummet while vehicles are rezzing and moving in the region, Today it went as low of 0,3 fps.

We didn't break the JIRA It was always burning As the bugs kept churning. We didn't break the JIRA: We just try to use it While the Lindens lose it. You can find enough horribly old and unresolved bug reports to fill out the rest of the song.

Somebody has secretly switched over the compass and the chronometer on the Linden Ship, and they can't tell the difference. So Rodvik is steering us straight for half-past-midnight, and we will arrive on-time at North-by-North-West.

Is SL dying? There is a downward trend on many things, and while I can see how such things as Mesh and Pathfinding can make a difference, they haven't been well done. It's not quite three years for me, and I admit, Jo, I'm not the same sort of user that you are. With the way things are going, I am more and more inclined to give up my Premium status. I paid, and I hoped for some technical problems to improve. I paid, and they didn't improve. Some of them get worse. What, I wonder, do I really get for that little patch of land I get? What do I use it for? What I get from SL are the social interactions, the people I hang out with. I'd be happy without Mesh, without Pathfinding, . I'd be happy if they'd spent time and effort fixing the old problems, which they seem to have not fixed in three years, and which seem worse than I have ever experienced. We're stuck with decisions that were taken a decade ago. SL may have some advantages still, but somebody else might take the idea, and write their own code, and not be tied down by those old choices. SL might be the Visicalc of virtual worlds.

SVC-8124 is a good example. Under the new JIRA system I doubt that there would be public knowledge of a problem which has severe effects on some users, and is effectively a DOS attack on any user who is standing in the wrong part of a region. I just don't have confidence that the Labs has the ability to provide the help to customers, paying or otherwise, that the open access to the JIRA allowed us to provide for each other. OK, I could have missed it, but I never saw a Linden warning us to watch out for excessive bandwidth consumption. And, for those who want to complain about the abuse and the ranting, this makes it even harder to deny the common accusation that Linden Labs doesn't listen and doesn't care. Oskar is exceptional. But I wonder just how he going to be able to give useful information about the bug fixes working through the public Beta process. JIRA references are not yet worthless, because so many being fixed are old-JIRA. IF we cannot read the JIRA entry, which looks to be the new way, all those lists of numbers and titles will start to look pointless. It will depend on somebody geing able to create a JIRA entry with an informative title--possible, but rather unlikely. It's also going to mess up the TPV community, making it harder to see which bugs are down to the Viewer, and which of those are being fixed by Linden Labs.

The impression I have is that the Navmesh is used for all physical objects, vehicles as well as PF characters. Avatars as well? I don't know. There's certainly been enough talk about changes in vehicle behavior. We had to have the Physics change to make Pathfinding possible, but with everything happening at once it's hard to see a boundary line.