Yesterday about lunch time the Lindens posted a blog article titled Performance, Performance, Performance. The surprising part, at least to me, is the performance improvements made to the Market Place. I’ve been covering the coming change to Viewer Managed Market Place. But, I don’t recall hearing about Market Place (MP) performance improvements.
Whatever, the Lindens say they are seeing a 70% faster response time and pages load 30% faster.
I’ve been sort of covering the improvements to group chat. News in this area is almost always after the fact. The Lindens have realized that whenever they say anything about changing something they get complaints about how the change is the cause of whatever problem users are having. So, they think they get better feedback on some things if they change and listen before saying anything.
This last week we have been running on new code in the chat servers. The metrics say things are MUCH better. I know I haven’t got a ‘cannot connect’ error when trying to comment in group chat this past week. But, I have also been busy and not in world much either.
I and others have been following and posting about the implementation of the Content Delivery Network. The Linden post reports they are seeing download times reduced by 50%. For those outside the USA improvement is even greater.
There are a few people here and there that report worse performance. A person’s location and ISP can degrade performance. But, most should see better performance or if they are close to the SL data centers, no change.
In a forum post Oz Linden explains what these changes should affect: Another small note about what this does and does not affect…
The CDN (and HTTP in those viewers that have them) improvements for texture and mesh data are improvements to how long it takes to download the the data. Once they’ve been downloaded, your viewer caches them locally so the next time you need that data it doesn’t need to download them at all. What that means is that these improvements mostly affect how quickly you see new things: places and objects you’ve never seen before. You probably won’t see a big difference in familiar places because your viewer should be using cached data for those.
So… now’s the time to try checking out places you’ve never seen before!
The use of pipelined HTTP for the initial load of your inventory when you log in matters every time, since while we cache some inventory data locally we always make sure it’s correct by loading it all at the beginning of your session. (link)
HTTP Pipelining is in place. Viewer 3.7.17-295700 is now the default viewer. It has the HTTP Pipeline and Inventory Fetch improvements. Between CDN and HTTP Pipelining the metrics show an 80% improvement in download time.
SL bloggers have been covering most of these improvements as they have developed. The Linden announcement marks them being rolled out (Tuesday). Be sure you are using the latest SL Viewer to experience the new improvements.
Only a some third party viewers have the HTTP code. It is hard to say which do and don’t. All viewers will take advantage of CDN as it is server side thing, no viewer change needed.
The Alchemy Beta Viewer has the HTTP Pipeline code.
Cool VL Viewer got the code in release 184.108.40.206.
Black Dragon 220.127.116.11 – This version has the pipeline code.
Singularity – I am uncertain, but I’ll say … probably not. Oz Linen says, not. The way they have been reporting working on the HTTP code in Singularity confuses me. All I am certain of, is they have done a lot of work improving the HTTP code in their viewer. But, their last update was early September. That may have been too early to include HTTP Pipeline code. If you are using Singularity, ask in the Viewer Group.
Firestorm – Nope. I don’t know if we will see the code in the planned November release. I am guessing that if they can get it in and it passes their QA, it will be included. But, that is a significant if. The next Third Party Developers meeting is next Friday. We will have seen a release or hear where they are at in their release process at that meeting.
Marine’s RLV 2.9.3 – There is a lot of new RLV code. But, I don’t see HTTP Pipelining mentioned. This viewer gets Server Side Appearance. Something most viewers have had for some time. So, I doubt it has HTTP Pipeline code.
The HTTP Pipeline Viewer in RC form has been out for awhile. I’ve been using it for over a week. There are threads in the forum from others that have been using it. (Thread) In general people are finding it better. I like it.
A thread that Ebbe and Oz are commenting in is here: Performance Improvements.
I am having some minor issues here and there. In the Server Beta Group people were talking about some textures refusing to render and some objects being stuck on LOD 4, the least vertices version for far away render. I haven’t seen that latter one. But, I have noticed a change in when some textures render and how some take forever to render. More on this later.
Whirly Fizzel comments in one post about horizon flickering: This flickering is probably MATBUG-252 , which started with the materials code and will happen on any materials enabled viewer when Advanced Lighting Model is enabled in graphics preferences.
To work around the problem either disable Advanced lighting model under Me -> Preferences -> Graphics or activate the Develop menu in the top menu bar with CTRL+ALT+Q, Develop -> Rendering -> Object-Object Occlusion -> untick it.
Whirly later comments: Anyone suddenly complaining that all rigged mesh is invisible, make sure they are not using the new 14.9.2 beta AMD Catalyst graphics drivers – this driver version will not render any rigged mesh unless hardware skinning is disabled – see BUG-7653
As for the problem of textures constantly going blurry then sharp then blurry over and over until you relog, you were probably out of texture memory – see BUG-2514 for a discussion about that.
Whirly describes one of the texture issues we have been seeing: Funny you brought that up Austin, we had a shoe creator contact Firestorm support today who reported her customers suddenly complaining about the textures taking a long time to change or just remaining grey and never loading when changed via her texture HUD.
I did some testing with her and I can reproduce the problem with the textures staying stuck grey and never loading, or taking a loooong time to load (as long as 5 mins in some cases) on both Firestorm and the default Linden viewer, Second Life 3.7.19 (295700) Oct 20 2014 13:37:20 (Second Life Release).
Her shoes have been on sale a long time and the complaints just started recently and she has had 10 complaints since Tuesday.
I have no idea yet if this is connected with CDN rolling out grid wide but I’m going to get those assets onto Aditi and test on one of the non-CDN regions to see if there is any different behavior.
I’ll file a JIRA issue once I have more info.
Another post by Theresa Tennyson points out an aspect of CDN. Regions that have yet to be loaded into a CDN cache in your area may be slower than usual to load. I pointed out in an earlier article that the CDN servers download the full version of any asset file before sending the first bits of data for the file to you. The older system would send the header, let viewer decide which LOD or image it needed for the current display, request that. Then the region server and asset system would send just the needed part.
While that may seem like it would be faster, that is not true for CDN. For the CDN servers it is smarter to get the whole thing then send you your parts. So, we can see a slow start where little seems to be happening and BOOM it all starts happening really fast.
The good news is once that region and its assets have loaded into CDN caches, no one else will see that delay.
In general people are seeing a big improvement.