#SL Server Update Week 32

Mesh seems to still be the top priority. Two of the release channels are running Mesh Enabled versions; Le Tigre and Magnum. Blue Steel is running a Server Maintenance upgrade. The Lindens have posted in the Wiki what changes are coming over the next few upgrades.

State of the Art?

Main Grid

The upgrade called Improved HTTP Service made it to the main grid last week. One of the items is improved texture fetch. As more users pile on HTTP Texture Get a problem was found in the servers and routers that handle the requests. Lately I’ve seen the time needed for textures to download go up. I’ve also seen more textures simply not download. I’ve also been in crowded places where I’ve changed groups 2 or 3 times to all the avatars to render. I still think changing groups to rez avatars sounds silly, but I see it work. This fix should correct the problem, but I can’t say I have seen improvement.

There is also a fix for Friends List showing Unknown, failed teleports, and not being able to save stuff.

Server Maintenance

This week the Server Maintenance pack was rolled to the main grid. It seems any group of changes bundled for testing is labeled a Server Maintenance upgrade. This one includes fixes for friend permissions being broken on viewers without web profiles. Also, a date this for Spanish language viewers. A fix for not being able to upload sound files. Something about sending postcards via email from the viewer was fixed.

The persistent user-unexpectedly-logged-off problem generally labeled ‘Timewarp’ gets more debugging traps added in this release. They are still trying to track down the cause of this problem. I first mentioned this problem in May then again in June.

The problem is complex because it apparently involves simulator-to-simulator communication. It is more common to be logged off when teleporting or walking into a new region. Whether the problem is in old code or some of the new code that compresses avatar hand-off data or the revised Mono2 script handling is unclear or at least we haven’t been told. I suspect the Lindens have yet to catch the actual problem. Adding more logging to get an idea of what is going wrong is the step one takes when the problem hides too well.

It is interesting that they mention removing some IM code. I know that code was added as the Lab moved toward XMPP Chat. The Lindens said they had not given up on going to XMPP. But, the trials had shown that XMPP was not fixing the problem. More logging and tracking code was added to they could find the cause of group chat lag. Once they understood the problem they revised the existing chat system. Changes made significant improvement in group chat. We still see chat lag, but I can almost always post a comment now and the chat lag is way less. The big exception is the Mesh Volunteers group. The Phoenix group, a very large group, works way better. More on this below.

An apparent high-IM-rate lagging the sim fix is in the package. The original problem was caused by a number of chatty bots lagging a region to a ridiculous degree. That the bots could do that was apparently a problem or at least revealed something that the Lindens decided needed to be fixed. It is often impossible to tell if the Lindens are blocking a griefer problem or improving performance. We may never know in regard to this fix. I know I’ve seen a number of regions with lots of bots and no one else round that appeared to be lagging for no reason.

Release Channels

The current upgrade running on Blue Steel contains upgrades for the new Direct Delivery (DD) system. DD will make selling in the market place easier by removing the need for Magic Boxes. Removing the Boxes will also reduce the script load on regions. It also reduces the need to rent a place in Second Life. I rent a place to set my Magic Box. I wonder what DD will do to land prices. Time will tell.

Mesh Project

The upgrades coming for mesh are no longer labeled Mesh Prep. These are now the Mesh Project.

One of the big changes in the release is how tier is charged. Any objects or link sets that trigger the new Prim Equivalence (PE) Calculation will see a change. Basically, if there is a mesh item in a link-set or it is set to physical the new PE Calc will kick in. This will affect the number of prims seen for prim charges and limits. There has been lots of discussion and complaining in the Mesh Group about how this is calculated and what the actual costs will be. In general, this change should NOT affect existing builds. Basically only new mesh items will be affected.

This also allows one to use large 64x64x64m prims. A Mega-prim is now a prim larger than 64m in any dimension. There is no way to make Mega-prims (prims over 64m) in SL. One has to get a copy of pre-existing Mega-prims. For a time prims over 10m are confusingly referred to as Mega-prims. Eventually Mega-prims will be the name for those larger than 64m. For now we sort of have prims, large prims, and Mega-prims.

You can go to a Mesh Sandbox and build the large prims. Take them into inventory and move them to a non-mesh enabled sim. Avoid editing the size of them in non-mesh enabled regions.

This upgrade has the new physics engine, Havok 2010. Most of the changes from Havok 2010 are technical mad mean little to most SL users. The noticeable change is the 10cm Gap Problem. This was a problem with colliding objects. If you deal with colliding objects you probably already understand the problem. If not, it doesn’t really matter.

The Parcel Return Order changes in this upgrade. This is a fix for the unintended consequence of the PE Calc changes. It is possible to change the PE of a link set drastically by adding a mesh. This can push a region over its prim limit and trigger automatic prim return. This change forces the return process to select the latest additions or changes in the region. So, it should be safer to work with mesh. However, I suggest you carefully test any new link set that you are adding mesh to in a sandbox, not you built out region.

News

Word is things are stabilizing and things are working more as intended. So, fewer roll backs, fewer panic fixes, and more success. We see that success as packages roll to the main grid and new items move into the release channels.

We have a new Linden: Mercille. He is building mesh models for use in The Engine Room, an estate that is planned to be made mostly of mesh objects.

Chat Lag fixes increased capacity by about 40%. More chat is happening and that capacity is being consumed. Stats are showing chat lag is bottle-necking at the Utility Node for the region. It is less group-size related and more related to the number of people who’s chat is routing through a busy node. Presumably one could tp to another region and escape an overloaded Utility Node. There is no way for a user to know which regions are going through which nodes.

Tori and Roads – Oskar pointed out the problem of roads being made of tori. They overload the physics engine and cause vehicles to misbehave. One can use tori to reduce prim count, which deceases performance. Or one uses more prims for the road, which improves performance.

Summary

It seems several older problems already fixes have snuck bask in with the mesh upgrades. Or they may just be scaling issues. Slow rezzing is back. Slow texture download is back. It is hard to tell if the slowness is a real problem or if it is a byproduct of logging code testing and monitoring a number of problems. I know the difference in performance between the new Dolphin Viewer and the new SL Development Viewer 3.0.1 is significant, 25 FPS verses 50 FPS. Inventory response time is significantly better in Dolphin.

Regardless of which viewer is being used texture is slow rendering.

Hopefully, we’ll see things stabilize and debug code dropped out with a corresponding improvement in performance.

3 thoughts on “#SL Server Update Week 32

  1. re. Sim lag (with or without bots):
    We run 3 bots on our mainland region, models and a mute greeter (unscripted). LL seems to be messing with bots. I am guessing this is an attempt to improve performance.
    HOWEVER, in every instance where we have seen Region / Simulator lag is is due to a lack of reserve capacity in the simulator. They are crowding 8 regions / simulator! That works fine so long no one has a successful event. 20 ppl in any one region lags the entire simulator. LL has got to do better at providing the reserve capacity for us to have events and other things that cause intermittent high usage without adversely effecting our sim neighbors.

Leave a Reply to Honza Noyes Cancel reply

Your email address will not be published. Required fields are marked *