The meeting notes from the Beta Server Group came out. There is some interesting stuff from the meeting. We got great information on CHAT. 🙂
The Lab has been working on what is labeled Server-Maint. This is a group of fixes and features that allow better roll out and problem recoveries. It has taken several tries and roll outs from the Release Channel had to be rolled back each time. The problems seems to have passed and Server-Maint. has made it to the main grid.
This upgrade has been cycling through the Release Channel for 3 or 4 weeks now. This changes touches so many aspects of SL that it is a complex change. While performance of new scripts rezzing in a region has been greatly improved, avatars with hundreds of scripts are still a problem. The Lab was working on that last week and the new fixes are in testing on Le Tigre this week. If they prove up, they will make the main grid next week.
The Lab thinks they have a fix that corrects vehicle problems without breaking the horses. If you don’t have a horse or a misbehaving vehicle, ignore this. This problem has facets in mesh and Mono.
The Release Channel test was giving odd prim counts in the regions. Some people took advantage of the erroneous counts and added more objects. Some probably did not realize there was a problem. When the test was rolled back things got returned. Not a happy situation. The Lab thinks they have the prim count fixed. If it passes testing this week it will roll to the main grid.
How prims are counted is changing to accommodate mesh. The total prim count of a linked set of prims and meshes can get complicated. So, lots of stuff is having to change. But, standard prim counts were not supposed to change.
The number of allotted prims on a parcel should not change. But, mesh can change the number of equivalent prims it costs. So, scaling a mesh up or down can change the mesh’s prim count and thus the number of free prims on a parcel. This is going to cause some confusion.
There is a bug labeled TmieWarp. This is about people getting logged off SL. It is a timing issue that the Lab is trying to run down. They have not found the fix yet. They are still working on it.
Last week the Lab ran a chat test to see if some of the changes they are planning work as intended. Alain Linden has provided a simplified explanation of how group chat works in SL. Paraphrasing:
A group chat is sent to the simulator the region is running on. The sim forwards it to a group chat service. The service tracks who is in the group and online and receiving the group’s chat. As things are now the service must locate every recipient by asking a lookup service. Once located the service sends the message to every sim with a recipient. If the group has 250 members online and engaged the server must do 250 lookups and then 250 sends. This is pretty much done for every line of chat. The lag is in the chat service. The lookup and sims are NOT the bottle neck.
One of the changes coming is to try caching the presence list. Another is to send one message to a sim with multiple recipients and let it the sim make and distribute copies. Unfortunately data shows there is an average of 1.1 recipients in a region, so this 1 message idea doesn’t provide much help.
If a recipient moves to another region and is not where the service expected them to be, the lookup is done and the message resent. But, these changes in small tests have improved performance by a factor of 2… whatever that means. Supposedly we would have half as many lost chats.
This information should give you the idea that if you are in a large chatty group you are going to lose chats. If a number of people are in the chat and on a busy sim, it will be worse.
The Lab has added more reporting channels to the chat service. So, now they can see more data on what their changes are doing.
Last week’s chat test should be analyzed and graphed by the next group meeting, Thursday 6/9.
You may remember at one time XMPP Chat was being considered as a replacement for the current chat system. That idea is still around. The Lab is looking at it as a possible long term solution. They did develop it far enough to realize it had the same problems as the current system. So, it has been placed on hold but not totally scrapped… yet… may be.