#SL Server Updates Week 20

The last two weeks have been hectic for the Lindens. Several updates to the server software did not work as expected and the crash rates went up. It seems those have been corrected and things are returning to the normal cycle of upgrades.

This week’s (today’s) roll out was stopped and rolled back part way through the process. Something went wrong.

Server Room

CERN Server Room - By: torkildr @ Flickr

Snack Release Channel

One of the code updates that is proving difficult to get working is the Mono2 upgrade. You probably remember it is the Mono scripts rezzing in a region that caused regions to stall for 5 to 30 seconds. That fix has been in progress since Mono was released. A couple of times now Mono2 updates and improvements have been rolled out and back. The Snack Release Channel is a special channel made up of just a few simulators on the main grid where the Mono2 changes can be tested.

The Mono2 updates have better tools and error tracking for figuring out what is happening with scripts. These new things are information for the Lindens. But, users get some new stuff too. There are new functions to allow script profiling (llScriptProfiler), so the user can tell how their script is using memory.  Another tells the user the peak amount of memory used (llGetSPMaxMemory) in a profiling run. Another reports currently used memory (llGetUsedMemory), I think, at any time executed.

Other fixes are in how the estate manager tools work. Top Scripts does a better job of reporting which scripts in a region are burning up CPU cycles. Instead of just immediate results the results will cover up to the last 30 minutes.

JIRA SVC-3895 Rezzing Mono scripted object cripples sim FPS – this one is still being worked on. If I understand correctly some improvements to this problem are on the main grid. But, they are not done with it.

A memory monitor tool was crashing sims. That has hopefully been fixed and is in testing.

A couple more sim crashers have been fixed and are being tested.

Since these tests were creating problems the Snack Temporary Release Channel was used to do a preliminary test. More bugs were found and fixed. The Lindens are now confident enough to place the update on Le Tigre.

Le Tigre Channel

In addition to the Kelley maintenance-server upgrades the channel is getting additional changes. Some are currently problems for residents. So, the fixes are welcome.

A couple of scripting commands in the LLSay category are being fixed. They have truncated multi-byte characters, failed to communicate correctly with attachments and sitting avatars, and were incorrectly blocked by region groups stuff.

There has been a way to get around estate bans for some time. That fix is being tested.

For sometime IM’s going to email have not handled replies, which just fail to arrive. That fix is in testing.

Some crash issue fixes are being tested.

The TimeWarp causing people to be logged off is being explored as they look for a fix. It is unclear what all is related to this problem. Some people are seeing their scripts become slow and unresponsive after region crossings. In some cases they fail to cross. In worst case they disconnect from the grid.

Blue Steel Channel

This channel get the pre-mesh roll out updates. I think this is exciting for many of us. Oskar is promising to tell us more soon.

Magnum Channel

This channel gets the Kelley Mono2 upgrades. I have conflicting reports on which channel got this update. I’m going with Oskar’s story.

Region Crossings

These are better and not so better… People are seeing poor performance crossing into some regions. Things are better but in the Homestead regions people are seeing the Time Dilation go to 0.01. Dilation at 1.00 is real time. Anything less is a delay. The region cannot get the simulator to complete all the needed tasks in time and delays updates to the next frame render. In other words you start to see region lag.

Ghost Prims

People are seeing some problems with prims and vehicles going ghost. ‘Unsitting’ from a vehicle may cause it to appear to vanish. It may later return in multiple incarnations and be exceptionally difficult to remove.

Summary

Several things are coming together at the same time. How many of these are affecting the each other to cause current problems is unclear. Things continue on and more solutions are being created as performance is upgraded and new features are added. Growth seems to come only with painful price.

 

3 thoughts on “#SL Server Updates Week 20

  1. Pingback: #SL Server Update Week 34

  2. Thanks for these updates. I’m on Blue Steel, and am having major problems with rezzing the past four days. Especially on the avatar, but with everything in general. My store is now taking 4 – 5 mins for full rez. I’m using Phoenix but switched around to V2 to see if that would help. Thus far, nothing is really helping. Also, people with older systems, such as XP seem to really be struggling. Would love more updates on what is exactly going on, and it’s hard to find specific information. Glad I stumbled upon your site. Thanks!

    • I’ve been seeing slow rezzing, avatars are usually the worst, but shopping malls have had problems too. Part of it is the routers/network and support services servers. A few weeks ago the Lindens saw the problem and started working on it. I am guessing that it has to do with the change to HTTP Get Texture. As more people use newer viewers this loads those servers. Some have reduce the problem by turning HTTP Get off.

      I suspect we will see things change in September when Direct Delivery comes online. We are getting lots of changes with Mesh, web profiles, and direct delivery. As these change the nature of the information flow the Lindens will likely have to change the infrastructure to handle the new bottle necks.

      Whatever the problem, I’ve seen it changing over the last few weeks, worse, better, worse… The viewer used does not make much difference, at least for me. I do try to use a viewer with the new large cache. It helps when I return to a previously visited place. I personally think SL caching is pretty ineffective, but the big cache seems to help.

      Be sure to click in the menu ‘The Stuff’ and look through the Server Related section.

Leave a Reply

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