Not a lot is happening in sim development right now, well … that is exciting anyway. The work of fixing bugs and tweaking performance continues and keeps Lindens busy. There are a few interesting tidbits.
The Morris region in ADITI got a new version of the LSL Optimization Project last week, week 35. Scripts should run faster there. Andrew is looking for any data people may gain experimenting with scripts there.
Andrew Linden is looking at the Lab’s auto-build service and considering how to move server development over to it and VS2010. There are advantages to using VS2010. I’m not doing any serious software application development, so I have no idea why it is better, other than the general newer is better thing.
But, there are several compile issues that will be improved by moving the server development over to the auto-build process. Things like using newer libraries and upgrading the server’s OS kernel are dependent on moving to auto-build. So, the move to VS2010 may just be a side effect.
Simon Linden is making some performance upgrades. Those are in early stages. So, we won’t see them reach a release channel for a couple of weeks. These improvements are better load balancing between region servers and utility servers, if I understood correctly.
SVC-6972 and the TIMEWARP SVC-6689 are being worked on. It is looking like the problem has been tracked down to something in the Linux/Debian kernel. So, the fix may be updating all the Linden servers to a newer Debian version.
Related are problems with the simulator server I/O. A busy region/sim can I/O block on another region. Clean up on those problems is in progress.
Being considered is the concierge service that assigns regions to servers. Region crossing are faster between regions in the same colos (co-locations) and better between regions in servers in the same rack, and best between regions in the same server. Assigning a balance of busy and idle regions will improve performance for busy regions. Figuring out which region is ‘busy’ is not simple. A region lagging might be considered busy, but how does one move a live region once it starts lagging? So, it becomes a matter of deciding what makes a region laggy and providing that information to the concierge as it is starting regions.
One also considers traffic patterns. Those regions with lots of teleports between them need to be in the same server or hopefully the same rack and definitely the same collocation. That criteria has to be integrated with the busy factor.
Plus there is the challenge of creating something that can’t be gamed. Once information is available on how to get a region into an ideal location, people will try to get a better location for their region. So, the influence of bots and loads of scripts being used to fake out the concierge has to be factored in.
The HTTP Texture Get is being looked at. Simon and Andrew have both seen the problem of HTTP Texture Get (sometimes called Fetch). Some documentation from the Phoenix team is in PHOE-2707. It seems some ISP’s are unhappy with the number of connections that SL uses. In some cases user modems cannot handle the number of needed connections. Other ISP’s have problems with the use of the HTTP protocol on ports other than the standard port 80.
One of the fixes the Phoenix Viewer Team has come up with is to reduce the maximum number of simultaneous HTTP Get connections allowed. I think they changed it from the default 16 to 8.
SH-744 – Avatar clothing change invisible to all inworld, but visible to me. This seems to be a problem related to HTTP. Some part of it was/is a network problem. But, it is not just a router problem. Part of it is the HTTP pipeline in the viewer. It seems to get blocked and have problems.
The Lindens are looking at the HTTP Texture Get system. Andrew is anticipating getting assigned to the team that will try and save the HTTP system.
Summary
While things are rather mundane work continues and progress is being made. Some of the big problems are taking time to fix. Steps for the fixes are in progress. We may not see much change on big problems like TIMEWARP, the 5 minute region slow down, rez surges, and others for the next couple of weeks.
The problem of defining “busy” regions is very complicated. A region might, for example, be nearly vacant except on the 2 times a week when the little club there holds events and 40 people routinely show up.