Re: Vs: Re: [realXtend] Performance statistics
What is the use of 1024 regions if you can't login to any region but one? I have tested grid setup and it still not operational on 0.1.1 The users can log on to any region that is contained on the servers ,They can then Freely Teleport to any other region or simply cross the border entering an adjacent region There is simply a small flash when crossing region borders... This is why I like Realxtend.. the border crossings of the regions .. even in stock modrex... are much better than opensim... In my humble opinion -- http://groups.google.com/group/realxtend http://www.realxtend.org
Re: Vs: Re: [realXtend] Performance statistics
I just tried setting up fresh taiga server from 0.1.1 distribution for the sake of expirement, I set up MySql database (to exclude the possibily of MSSQL problems), created opensim, grid, gridwebdav and os_modrex databases, used configuration wizard to generate config files, (only changed ip address to my machine ip). Run the whole thing using start.bat (empty region). Logged in using 0.42 viewer and select Upload texture from file menu. Nothing got uploaded (texture of size 0x0) and familiar error appeared at grid server log: "OnRequest() falied with Object Reference not set to instance of an object At OpenSim.Services.AssetService.AssetService.GetMetaData(String id)" Any ideas, why this happens? >>Jonne Nauha >>We will propably convert the demo worlds using taiga ~soonish. The main >>problem is that their stanalone db is sqlite and grid cant use that as is. If >>you know how to convert os standalone sqlite into grid mysql give me a shout >>:) the thing here would be a working scene exporter and then importing that >>to the Taiga replacement. I managed to convert assets database from Sqlite to MSSQL, when I accidently setup sqlite driver for assets and found 100 Mb assets stored in sqlite. Conceptually, the conversion is not very difficult (just export sqlite data as sql, change format and upload to MSSQL). Still, takes time to do properly. Scene exporter (with ability to export partial scenes) would be a great thing to have, the one really needed in work >>Matti Kuonanoja >>If you want to test with your own server that should be fine too. Naali 0.2.2 >>will have integrated basic voice UI and you will not have >>to have separate Mumble client installed anymore. Thanks for the info. I'm going to try new voice chat the moment I got working taiga setup. >>Antti Ilomäki: >>Some issues are to be >>expected at this point in development and feedback is very important. It >>would of course be a great help if you could take a look at the code >>yourselves as well and try to find the cause Yes, this is what I'm actually trying to do. But the things go forward slowly, since there are many issues and I can work on one issue at a time. >>Tara Desito: >> We managed to get 1024 regions out of Taiga around the first of april What is the use of 1024 regions if you can't login to any region but one? I have tested grid setup and it still not operational on 0.1.1 Best regards, Denis On 10 май, 18:00, Antti Ilomäki wrote: > The Beneath the Waves server at world.realxtend.org:9000 is also running on > Taiga (and we have some scripting issues there). > > Anyway, this is very interesting discussion indeed and it's great to hear > about everyone's experiences using Taiga and Naali. Some issues are to be > expected at this point in development and feedback is very important. It > would of course be a great help if you could take a look at the code > yourselves as well and try to find the cause, since you are running the > environments and know exactly how to reproduce the errors. > > --http://groups.google.com/group/realxtendhttp://www.realxtend.org -- http://groups.google.com/group/realxtend http://www.realxtend.org
Re: Vs: Re: [realXtend] Performance statistics
The Beneath the Waves server at world.realxtend.org:9000 is also running on Taiga (and we have some scripting issues there). Anyway, this is very interesting discussion indeed and it's great to hear about everyone's experiences using Taiga and Naali. Some issues are to be expected at this point in development and feedback is very important. It would of course be a great help if you could take a look at the code yourselves as well and try to find the cause, since you are running the environments and know exactly how to reproduce the errors. -- http://groups.google.com/group/realxtend http://www.realxtend.org
Re: Vs: Re: [realXtend] Performance statistics
>>>Matti Kuonanoja:We are reimplementing in-world voice for Naali/Taiga with >>>Mumble > technology. The new voice system is not yet complete but it's already > testable and it should offer better quality and more features than > the > old system did > > That would be great, if I could setup that. Thanks for the info, I > will certanly take a look on this new voicechat system. Could you > please describe what do I need to test this system or may be point to > some place with configuration instructions? Does this work in > standalone mode or grid mode or both? You can find more information about current implementation here: * http://wiki.realxtend.org/index.php/Mumble_integration * http://wiki.realxtend.org/index.php/Mumble_usage_Instruction You can test the audio quality straightaway using Naali 0.2.1+Mumble client to world.realxtend.org:9000 If you want to test with your own server that should be fine too. Naali 0.2.2 will have integrated basic voice UI and you will not have to have separate Mumble client installed anymore. Current implementation doesn't care about regions so users are ably to communicate over region borders. Client connects to murmur server by address it's request after login. This will probably change in future to support more dynamic way to define voice servers. I would appreciate any feedback about the voice system we have currently under development. Especially if you have use cases with feature requirements. Cheers, Matti Kuonanoja ~ realXtend developer -- http://groups.google.com/group/realxtend http://www.realxtend.org
Re: Vs: Re: [realXtend] Performance statistics
> The performance upgrade from .10 to .11 ... was astonishing...A > 152,640 Prim region is definately an accomplishment worth merit because > only SL uses 15,000 prim regions and they would lag out almost as bad > as my 152,640 prim region lagged out ..The high prim counts possible > per region in the latest release .. Seem to also implythat High User > counts are also possible per region Are high prim count regions > like this even possible in Vanilla Opensim ? > We have not been able to test high user counts on our servers yet... We would love to have volunteers attempt to crash one of our regions ..If anybody out there would like to view this grid .. or to help in Taiga testing/development please contact me via private IM .. As you can tell .. I am very excited about Taiga .11... Because of the dramatic increase in server side performance ... IF we can work out the asset issues especially the Rex mesh objects ... Oh hey another error that I get.. that is fairly critical I suppose .. Is that I get the error... Failed to locate a LL cache The config file system.. The way it is set up.. For some reason.. it will not activate theCache(flotsom cache .. glenn tucher cache etc ) When I change the config file to a variant of an Opensim Config file.. It will activate the cache... But then we are unable to Scale up past one or two instances So in short.. with your new config file.. we get the error that cache wont activate .. but we can scale up ..It says that storage of temporary objects is not enabled.. but it seems to work... -- http://groups.google.com/group/realxtend http://www.realxtend.org
Re: Vs: Re: [realXtend] Performance statistics
. We managed to get 1024 regions out of Taiga around the first of april Taiga .11 ... has a dramatic increase in server side operability and performance ... I set up a test region yesterday of 152,640Prim ... To an SL purist it is shocking to have such amazing performance ... With Taiga .10 or with opensim .69... we were only able to get 35,506 prim on a region before it choked out With the new Taiga .11we have been able to get Sci-Sim type performancewhich is simply amazing object not to a reference of an object This error .. is the same one that Denis Mentioned.. we get this error as well.. I see that it is not there this morning .. but we usually have that issue and as he said .. it does Repeat ... failed to fetch asset metadata --- As far as I can tell ..some objects that are uploaded from sketchup or other sources ... They are curropt and these curropt objects will take down a region and the entire databases that service that specific regional instance ... Naali --- wont log into our Grid ... this issue has the greatest downside to it .. as we are not able to utilise... newer services that are being deployed ... Taiga took a dramatic step forward with the latest Release .. at least it did for us .. .10 was choppy .. it was like riding around in an Old Buggy .. that bounces all over the road.. The new .11is an incredible upgrade.. as it is much more like Driving a fine sports car... Another issue .. is the sockets ... it is hard to scale up the instances .. because there are socket issues.. Plus it is not easy to Tie together Various servers that you might have in your HPC cluster ... With standard Opensim .. We were able to tie together all of our servers much easier .. With Taiga ... The regions Appear on the MAP .. but you cannot access them.. You are unable to enter the regions ... It was a challenging task to "partially" deploy our Taiga Grid ( we have not got past the issue of tying our servers together yet ). But it was not an impossible task The new Server set is actually beyond Naali at this point ... In my humble opinion... I am unsure why it is labelled only .11 after such a dramatic upgrade that you just gave us .. Those of you that are having some difficulties deploying your grids ... I am willing to help some of your deployment efforts ... Otherwise.. just hang in there... its a challenging effort ... no doubt about it.. But if we were able to Deploy 1024 taiga regions.. and have kept it stable for a month now... then it is possible Special thanks To Realxtend Core ... as we as Intel .. for making this possible for us -- http://groups.google.com/group/realxtend http://www.realxtend.org
Re: Vs: Re: [realXtend] Performance statistics
Are you aware that you nees to put assets/users etc tables that you have to the grid (default name) db not into opensim db? Sounds to me like the grid servises are going to their own db and not finding anything. For my taiga setup the opensim db is completely empty, i dont think the sim itself stores anything there. It asks the grid services to store/fetch data and thats mainly done in the grid (again default name you might have changed this from taiga wizard) db. Userserver saying its not finding uuid for agent might be that the grid db is empty. Can you verify you have your 300mb dataset in the grid db not is opensim db? The second possibility is of course that your standalone os version is old so that the db schema has changed and just putting your old tables to the grid db wont work. But i would expect it to give error on login/whatever if this was the case. Also for toni and you: I think fish and audi demos are not running taiga. Only thing that is running Taiga 0.1.1 (or 0.1.0 for that matter) is wolrd.ralxtend.org:8002. We will propably convert the demo worlds using taiga ~soonish. The main problem is that their stanalone db is sqlite and grid cant use that as is. If you know how to convert os standalone sqlite into grid mysql give me a shout :) the thing here would be a working scene exporter and then importing that to the Taiga replacement. -- Brgrds, Jonne Nauha Gmail via the Nokia N900 - Original message - > Hello, > > Thanks everyone for the helpful info. One of my problem with realxtend > was that it is sometimes diffucult to get info because it is > considered "obvious" by developers :) > > > > Toni Alatalo: AFAIK e.g. the fishworld demo is now running current Taiga > > > though, so migrations should work? Also some other demo worlds have been > > > migrated to new server versions every once in a while. Could you perhaps > > > provide your problematic dataset so that people who know about the > > > migrations > > > could debug that > > The dataset is over 300 Mb. I can put it online and give a download > link to someone who wants to work on than. > So far I have following exception when starting taiga > > On Request failed with Object reference not set to an instance of an > object at > OpenSim.Services.AssetService.AssetService.GetMetadata(String id) > > This repeats many times > > Then, after server starts no user can login. Server log: > "[CONNECTION BEGIN] We can't user info record for > UUID ..." > "[LLUDPSERVER] Ack timeout, disconnecting" > > When I replace my database name with empty database name in ini files > everything works OK > (I configured databases manually, in OpenSim.ini, OpenSim.Server.ini > and modrex.ini). > > > > Toni Alatalo: In osgrid tests they have reached something close to a > > > hundred > > > (>80?) > but indeed then things become slow. I have heard from osgrid testers > that something like 40 would be usable with 0.6.9, and the Wright > Plaza > weekly meetings host I think typically >20 AVs and run ok? > > Yes, I saw many reports like this on 40-80 avatars existing > simultaneously on OpenSim. The only thing I can say, this does not > work for me. May be they know some magic words or have custom opensim > adaptations or may be this issue somehow related to modrex? > > > > Jonne Nauha : But then again you cant just say "i want my standalone > > > server > > > db to be a grid db" im pretty sure this is not very easy... > > Why not? I have done that before with that old 0.5-release realxtend > server without such problems. I'm now trying to run only one region > simulator, all servers on the same machine and even this setup does > not work. There should really be some easy way to transfer data > between versions and different setups. Starting filling the world all > over again each time new version is released is not really very > pleasing expirience. > > > > Jonne Nauha: And if you do have a big world, why not > go to grid mode and load balance your grid services and sim on > multiple > servers? > > Yes, I agree with that. The problem is I can't even get one region to > work with Taiga. Besides, last time I tested grid mode in Taiga, it > was not operational and since as I understand no one is actively > working on multiregion grid support now, I expect things to be in the > same state. Please, correct me, if I wrong here. Simple OpenSim 0.6.9 > + modrex 100% does not work in grid mode (I have build a version that > *does* work, just to see if it is possible, but have no time to try it > fully, since I have been occupied with single region/standalone > problems) > > > > Jonne Nauha: Was an interesting read. I hope you have posted bug reports > > > for > > > these as it > seems you know where the problems lay. Don't feel shy to post taiga/ > naali > bug reports to github aswell :) > > I have reported one bug to opensim mantis (the one, that I tracked > down to the exact code line where the problem happens), but they
Re: Vs: Re: [realXtend] Performance statistics
Hello, Thanks everyone for the helpful info. One of my problem with realxtend was that it is sometimes diffucult to get info because it is considered "obvious" by developers :) >> Toni Alatalo: AFAIK e.g. the fishworld demo is now running current Taiga >> though, so >>migrations should work? Also some other demo worlds have been migrated >>to new server versions every once in a while. Could you perhaps provide >>your problematic dataset so that people who know about the migrations >>could debug that The dataset is over 300 Mb. I can put it online and give a download link to someone who wants to work on than. So far I have following exception when starting taiga On Request failed with Object reference not set to an instance of an object at OpenSim.Services.AssetService.AssetService.GetMetadata(String id) This repeats many times Then, after server starts no user can login. Server log: "[CONNECTION BEGIN] We can't user info record for UUID ..." "[LLUDPSERVER] Ack timeout, disconnecting" When I replace my database name with empty database name in ini files everything works OK (I configured databases manually, in OpenSim.ini, OpenSim.Server.ini and modrex.ini). >> Toni Alatalo: In osgrid tests they have reached something close to a hundred >> (>80?) but indeed then things become slow. I have heard from osgrid testers that something like 40 would be usable with 0.6.9, and the Wright Plaza weekly meetings host I think typically >20 AVs and run ok? Yes, I saw many reports like this on 40-80 avatars existing simultaneously on OpenSim. The only thing I can say, this does not work for me. May be they know some magic words or have custom opensim adaptations or may be this issue somehow related to modrex? >> Jonne Nauha : But then again you cant just say "i want my standalone server >> db to be a grid db" im pretty sure this is not very easy... Why not? I have done that before with that old 0.5-release realxtend server without such problems. I'm now trying to run only one region simulator, all servers on the same machine and even this setup does not work. There should really be some easy way to transfer data between versions and different setups. Starting filling the world all over again each time new version is released is not really very pleasing expirience. >>Jonne Nauha: And if you do have a big world, why not go to grid mode and load balance your grid services and sim on multiple servers? Yes, I agree with that. The problem is I can't even get one region to work with Taiga. Besides, last time I tested grid mode in Taiga, it was not operational and since as I understand no one is actively working on multiregion grid support now, I expect things to be in the same state. Please, correct me, if I wrong here. Simple OpenSim 0.6.9 + modrex 100% does not work in grid mode (I have build a version that *does* work, just to see if it is possible, but have no time to try it fully, since I have been occupied with single region/standalone problems) >>Jonne Nauha: Was an interesting read. I hope you have posted bug reports for >>these as it seems you know where the problems lay. Don't feel shy to post taiga/ naali bug reports to github aswell :) I have reported one bug to opensim mantis (the one, that I tracked down to the exact code line where the problem happens), but they seem to have many bug reports and few time. The other issues, I'm not sure why exactly they happen. Sometimes I don't even know if the bug can be seen in "pure" opensim without modrex and conseqenlty don't know where to report it >>Matti Kuonanoja:We are reimplementing in-world voice for Naali/Taiga with >>Mumble technology. The new voice system is not yet complete but it's already testable and it should offer better quality and more features than the old system did That would be great, if I could setup that. Thanks for the info, I will certanly take a look on this new voicechat system. Could you please describe what do I need to test this system or may be point to some place with configuration instructions? Does this work in standalone mode or grid mode or both? Many thanks to all, Denis Tarasov -- http://groups.google.com/group/realxtend http://www.realxtend.org
Re: Vs: Re: [realXtend] Performance statistics
On Fri, May 7, 2010 at 10:03 AM, Denis Tarasov wrote: > Problem 5 > realxtend Voice chat is not operational with multi-region setup > Solution: Fix the code voice chat server code (we only have partial > solution so far) We are reimplementing in-world voice for Naali/Taiga with Mumble technology. The new voice system is not yet complete but it's already testable and it should offer better quality and more features than the old system did. I hope this will help you in near future if you have problems with voice chat and you can use consider to use Naali as the viewer. Cheers Matti Kuonanoja ~ realXtend developer -- http://groups.google.com/group/realxtend http://www.realxtend.org
Re: Vs: Re: [realXtend] Performance statistics
Hello, Seems like most of you problems are just opensim based not that much about rex? For a heavy server UDP will be slow and you need to do many reconnects to get the world etc. We have made some big improvements in Naali for HTTP assets. We can now fetch textures and meshes in Taiga 0.1.1. Its a shame that you cant set Taiga up with you db. But then again you cant just say "i want my standalone server db to be a grid db" im pretty sure this is not very easy unless you have your whole grid on the same machine, and even then it prolly requires some expertiece. And if you do have a big world, why not go to grid mode and load balance your grid services and sim on multiple servers? Was an interesting read. I hope you have posted bug reports for these as it seems you know where the problems lay. Don't feel shy to post taiga/naali bug reports to github aswell :) Best regards, Jonne Nauha realXtend developer http://www.realxtend.org/ http://www.evocativi.com/ On Fri, May 7, 2010 at 10:03 AM, Denis Tarasov wrote: > Hi, > > We are basically working on similar project - virtual education > environment and have some expirience working with > realxtend-based worlds. If you want to build a large realxtend-based > world, be prepared to confront seriuos issues. > The specter of problems is different from version to version. We are > currently using OpenSim 0.6.9 RC1 + approriate modrex, all > compiled from sources on 64-bit windows server + MSSQL Server as > database engine. Overall database size ~ 300 MB > > Hardware: Dual CPU SMP Xeon Quad Core -based Server, 8 GB of RAM > > The setup is multiregion standalone, but almost all content is > now contained within single region. We uploaded ~ 250 meshes (polygon > count 900-3, average 7000), ~ 100 textures and about 1-minute > length 50 sounds. > > Problem 1. > Most of the time, the server uses 50-150 MB of memory, but peak memory > consumption could easily go over 1.3 Gb. When this happens 32-bit > version crashes with OutOfMemoryException. 64-bit version is > incompatible with 32-bit ODE physics. Recompling ODE for 64 bit on > windows produced weird physics behavior. > > Our solution: - compile the program with /LARGEADRESSAWARE flag > > Problem 2. World never loads fully. Only about 50-60% of all meshes is > visible in 0.42 viewer, with Naali, the situation is even worse. > Sometimes you also have few prims missing and rectangular holes in the > ground. This is due to the way LLUDPClientStack is implemented in > 0.6.9. There is also a bug there (see problem 3). Older > implementations (Opensim 0.6.7) are better in a sense that 90-95 of > all meshes will eventually be uploaded to client, provided you set > ReliableIsImportant=true in opensim.ini. But 0.6.7+modrex is unstable > when it comes to handling of large number of avatars simultaneoulsy > (see Problem 4) > > Possible solutions: > a. distrubute world viewer to users with all meshes already placed in > cache (improves situation a lot, but not solves the problem completly. > Increases client size, make difficlut to update your world) > b. Use old Opensim 0.6.7 > c. I have read that new Taiga server supports http assets downloading. > This might improve situation (only when using new Naali viewer). When > I tried to test, this, I was unable to run Taiga with my database > (example installation with empty database works well, but useless). > With our data set, it throws multiple exceptions on startup, have > problem with logging users in, and a number of other issues. > > Problem 3. The code that handles resend of lost UDP packets in opensim > 0.6.9 has threading-related bug, that lead to server continuolsly > throwing "Index out of bounds" exception (every second of so) when > user is logged in. This also makes Problem 2 worse. > > Solution: We have implemented a custom workaround, that seems to work > > Problem 4. > The maximum number of avatars that can login to the region is 16. > Anything more makes world absolutely unusable. Every five minutes > somebody is thrown out, and every 15 minutes the server crashes and > must be restarted. The maximum number of avatars that can work > comfortably is 8. > > Solution: I don't know any solution here yet, we are working on it. > Any help from anybody would be greatly appriciated > > Problem 5 > realxtend Voice chat is not operational with multi-region setup > Solution: Fix the code voice chat server code (we only have partial > solution so far) > > I hope, this summury could help you. Good luck with your project. > Feel free to contact me if you have any additional questions. > > Best regards, > Denis Tarasov > > > > On 7 май, 09:07, "Mikko Liukkonen" wrote: > > Thx, Keep me updated, I will appreciate it. > > > > BR, > > > > Mikko > > > > >>> Toni Alatalo 6.5.2010 22:49 >>> > > > > Mikko Liukkonen kirjoitti: > > > > > We are implementing realxtend based learning game during this fall. I > like to enquire if somebody have made some performance tests and cou
Re: Vs: Re: [realXtend] Performance statistics
Denis Tarasov kirjoitti: Problem 2. World never loads fully. Only about 50-60% of all meshes is visible in 0.42 viewer, with Naali, the situation is even worse. Sometimes you also have few prims missing and rectangular holes in the ground. This is due to the way LLUDPClientStack is implemented in 0.6.9. There is also a bug there (see problem 3). Older This is fixed in current Naali and Modrex releases (the modrex in Taiga 0.1.1), with the usage of http for asset downloads. In our tests so far it works really well now, all meshes and textures come reliably and quickly. For more information, see Jonne Nauha's recent post to the -dev list. The first solution, which requires running a command on the server side to associate a URL for each asset, works against Naali 0.2.1. The other solution to use an Opensim CAP to get the baseurl for assets on that particular asset server is implemented in current Naali develop and is going to be included in the 0.2.2 coming after next week (we do time based releases every 3 weeks now, after each dev sprint). a. distrubute world viewer to users with all meshes already placed in cache (improves situation a lot, but not solves the problem completly. Increases client size, make difficlut to update your world) This has been done here too, Naali even supports that as a feature so that you can add a content pack to your client install. Was done before the work on getting the http asset pipe was finished, earlier this year. c. I have read that new Taiga server supports http assets downloading. This might improve situation (only when using new Naali viewer). When I tried to test, this, I was unable to run Taiga with my database (example installation with empty database works well, but useless). With our data set, it throws multiple exceptions on startup, have problem with logging users in, and a number of other issues. Bleh, should have read the post fully before started replying, of course you knew about this already :) Well, at least there was some specific info about in which versions it works and how. We haven't worked too much on migrations yet, one reason is that we are also using the dotscene import to bring the scenes in from how they have been exported from e.g. Blender. AFAIK e.g. the fishworld demo is now running current Taiga though, so migrations should work? Also some other demo worlds have been migrated to new server versions every once in a while. Could you perhaps provide your problematic dataset so that people who know about the migrations could debug that? The maximum number of avatars that can login to the region is 16. Anything more makes world absolutely unusable. Every five minutes somebody is thrown out, and every 15 minutes the server crashes and must be restarted. The maximum number of avatars that can work comfortably is 8. Solution: I don't know any solution here yet, we are working on it. Any help from anybody would be greatly appriciated In osgrid tests they have reached something close to a hundred (>80?) but indeed then things become slow. I have heard from osgrid testers that something like 40 would be usable with 0.6.9, and the Wright Plaza weekly meetings host I think typically >20 AVs and run ok? Did you look at the Intel work on this? They target 50,000 or something users in the same place :) , and seems that with their client manager idea that is possible, but not a short term solution (hopefully they have some sort of solid prototype by the end of the year - first test was iirc in last August running 400 text-only test clients from the same computer). Denis Tarasov Thanks for the info about the results in your testing, hopefully the prob with your data and other possible probs with current modrex versions for you are solved soon so you also get to test the http assets etc. ~Toni On 7 май, 09:07, "Mikko Liukkonen" wrote: Thx, Keep me updated, I will appreciate it. BR, Mikko Toni Alatalo 6.5.2010 22:49 >>> Mikko Liukkonen kirjoitti: We are implementing realxtend based learning game during this fall. I like to enquire if somebody have made some performance tests and could share the results with us. Folks at Intel did some profiling and optimizations last year to Opensimulator and are now continuing to work towards allowing large scale mass events etc., they gave a nice presentation in a workshop recently and the material is on-line: paper athttp://vw.ddns.uark.edu/X10/content/ARCHITECTURE--Designing-Extensibl... and the presentation with more about the future plans inhttp://vw.ddns.uark.edu/X10/content/Extensible%20Virtual%20World%20Ar... Otherwise the opensim-dev mailing list and sciencesim, osgrid etc admins are knowledgeable about this. Realxtend is not used as much for heavily used things yet, and the server side performance is probably quite identical 'cause modrex doesn't introduce much additional computing. BTW we
Re: Vs: Re: [realXtend] Performance statistics
Hi, We are basically working on similar project - virtual education environment and have some expirience working with realxtend-based worlds. If you want to build a large realxtend-based world, be prepared to confront seriuos issues. The specter of problems is different from version to version. We are currently using OpenSim 0.6.9 RC1 + approriate modrex, all compiled from sources on 64-bit windows server + MSSQL Server as database engine. Overall database size ~ 300 MB Hardware: Dual CPU SMP Xeon Quad Core -based Server, 8 GB of RAM The setup is multiregion standalone, but almost all content is now contained within single region. We uploaded ~ 250 meshes (polygon count 900-3, average 7000), ~ 100 textures and about 1-minute length 50 sounds. Problem 1. Most of the time, the server uses 50-150 MB of memory, but peak memory consumption could easily go over 1.3 Gb. When this happens 32-bit version crashes with OutOfMemoryException. 64-bit version is incompatible with 32-bit ODE physics. Recompling ODE for 64 bit on windows produced weird physics behavior. Our solution: - compile the program with /LARGEADRESSAWARE flag Problem 2. World never loads fully. Only about 50-60% of all meshes is visible in 0.42 viewer, with Naali, the situation is even worse. Sometimes you also have few prims missing and rectangular holes in the ground. This is due to the way LLUDPClientStack is implemented in 0.6.9. There is also a bug there (see problem 3). Older implementations (Opensim 0.6.7) are better in a sense that 90-95 of all meshes will eventually be uploaded to client, provided you set ReliableIsImportant=true in opensim.ini. But 0.6.7+modrex is unstable when it comes to handling of large number of avatars simultaneoulsy (see Problem 4) Possible solutions: a. distrubute world viewer to users with all meshes already placed in cache (improves situation a lot, but not solves the problem completly. Increases client size, make difficlut to update your world) b. Use old Opensim 0.6.7 c. I have read that new Taiga server supports http assets downloading. This might improve situation (only when using new Naali viewer). When I tried to test, this, I was unable to run Taiga with my database (example installation with empty database works well, but useless). With our data set, it throws multiple exceptions on startup, have problem with logging users in, and a number of other issues. Problem 3. The code that handles resend of lost UDP packets in opensim 0.6.9 has threading-related bug, that lead to server continuolsly throwing "Index out of bounds" exception (every second of so) when user is logged in. This also makes Problem 2 worse. Solution: We have implemented a custom workaround, that seems to work Problem 4. The maximum number of avatars that can login to the region is 16. Anything more makes world absolutely unusable. Every five minutes somebody is thrown out, and every 15 minutes the server crashes and must be restarted. The maximum number of avatars that can work comfortably is 8. Solution: I don't know any solution here yet, we are working on it. Any help from anybody would be greatly appriciated Problem 5 realxtend Voice chat is not operational with multi-region setup Solution: Fix the code voice chat server code (we only have partial solution so far) I hope, this summury could help you. Good luck with your project. Feel free to contact me if you have any additional questions. Best regards, Denis Tarasov On 7 май, 09:07, "Mikko Liukkonen" wrote: > Thx, Keep me updated, I will appreciate it. > > BR, > > Mikko > > >>> Toni Alatalo 6.5.2010 22:49 >>> > > Mikko Liukkonen kirjoitti: > > > We are implementing realxtend based learning game during this fall. I like > > to enquire if somebody have made some performance tests and could share the > > results with us. > > Folks at Intel did some profiling and optimizations last year to > Opensimulator and are now continuing to work towards allowing large > scale mass events etc., they gave a nice presentation in a workshop > recently and the material is on-line: > > paper > athttp://vw.ddns.uark.edu/X10/content/ARCHITECTURE--Designing-Extensibl... > > and the presentation with more about the future plans > inhttp://vw.ddns.uark.edu/X10/content/Extensible%20Virtual%20World%20Ar... > > Otherwise the opensim-dev mailing list and sciencesim, osgrid etc admins > are knowledgeable about this. Realxtend is not used as much for heavily > used things yet, and the server side performance is probably quite > identical 'cause modrex doesn't introduce much additional computing. > > BTW we are working on a learning environment for school kids using it > too, pilot testing etc. will probably be done during fall in that > project as well. There's a bit old/earlier news announcement about it > athttp://maxping.org/business/real-life/realxtend-virtual-worlds-powere... > (maxping seems to be offline now though?), more info coming late this > month / early july probably an
Vs: Re: [realXtend] Performance statistics
Thx, Keep me updated, I will appreciate it. BR, Mikko >>> Toni Alatalo 6.5.2010 22:49 >>> Mikko Liukkonen kirjoitti: > We are implementing realxtend based learning game during this fall. I like to > enquire if somebody have made some performance tests and could share the > results with us. > Folks at Intel did some profiling and optimizations last year to Opensimulator and are now continuing to work towards allowing large scale mass events etc., they gave a nice presentation in a workshop recently and the material is on-line: paper at http://vw.ddns.uark.edu/X10/content/ARCHITECTURE--Designing-Extensible-and-Scalable-Virtual-World-Platforms--Bowman-Lake-Hurliman.pdf and the presentation with more about the future plans in http://vw.ddns.uark.edu/X10/content/Extensible%20Virtual%20World%20Architectures_Slides%20(Lake).pdf Otherwise the opensim-dev mailing list and sciencesim, osgrid etc admins are knowledgeable about this. Realxtend is not used as much for heavily used things yet, and the server side performance is probably quite identical 'cause modrex doesn't introduce much additional computing. BTW we are working on a learning environment for school kids using it too, pilot testing etc. will probably be done during fall in that project as well. There's a bit old/earlier news announcement about it at http://maxping.org/business/real-life/realxtend-virtual-worlds-powered-future-school-planned-at-oulu,-finland-.aspx (maxping seems to be offline now though?), more info coming late this month / early july probably and we can of course talk about it too if you're interested. > Mikko Liukkonen ~Toni -- http://groups.google.com/group/realxtend http://www.realxtend.org -- http://groups.google.com/group/realxtend http://www.realxtend.org
