We have been running our own OpenSim-based grids now for more than a year. I thought it would be a good thing to share some latest comments on how it is going and what some of the issues are. You can visit either the Rezzable Grid or Heritage Key and see King Tut Virtual right now.
Go Check it out, OpenSim works.
OpenSim works, is getting better and will be the future of the 3D Web.
First thing to point out is that we are using OpenSim in combination with Drupal as the core user and even grid admin system. So your own usage may not be the same if you are just working with OpenSim all by itself. We also have about 10 servers cranked-up to support all of this on a cluster sharing 50mb bandwidth. Info on the tech layout here.
In general OpenSim has come quite a long way in the last year. There is still quite some road in front to get it to be a sort of 3D Web Server like Apache. It is very complex–maybe too much so, but nonetheless it works. And it works a lot better than it did a year ago.
The greatest strength of OpenSim is that it is opensource and has a smart enough set of committed developers beating on it. We have been luck enough to work with a few of them and meet even a few more. Because it is opensource we have been able to create our own integrations and services around the core.
The core dvelopers have cleared-up a lot of real problematic stuff like cross-region boarders, assets vanishing, general caching/grey-goo. You can give folders now! The regions are snappy. The build issues are still there on some level, but the workarounds are also better understood. The most painful build issues lately have been with scripting (and not overloading scripting).
The use cases that Rezzable is focused on, try to aim for the what OpenSim does right now and not steer directly into known gaps. So, again you own use cases might trigger different issues.
Below is my list of Top 10 Issues we see with OpenSim —
OpenSim Issue 1: Physics is still weak
Physics engine options are not that good right now. This causes a lot of problems and work-arounds. Shooting, dropping, collisions and driving are basically to be avoided. The physics engine also seems to create a lot of issues with the bounding boxes and sometimes you can’t walk through doors easily or phantom prims get solid-ish.
The obvious solution here is to find a way to plug-in better physics engines. It would seem that OpenSim will support this, but so far I have not heard of anything implemented in the last year that changes the performance.
OpenSim Issue 2: SL Hangovers
OpenSim seems (and we really try to avoid most of this stuff) to have a lot of SL-oriented functions especially around land, parcels that are more interesting to SL-copycat virtual worlds than our interest in 3D online experiences. Net Net is the core code isn’t as performance-oriented as we would like to see it.
The biggest gap vs SL is that OpenSim does not have a commerce feature and of course no $L which is still a trusted micro-currency.
OpenSim Issue 3: Perms are Not There
We did quite a bit of work to implement a fairly basic permissions concept, mainly using Drupal roles. This works fine for Heritage Key, and we also have a separate “sub-grid” for build work. On the Rezzable grid it will require a lot more manual switch-flipping to have quasi-group/collaborative control on access and prim-perms. We have not implemented any commerce concepts (and may not either). I think the whole issue of perms/DRM needs a more complete architecture to avoid the issues beleaguering/demoralizing SL content creators.
OpenSim Issue 4: Lack of Docs
I suppose if you like to read code, you should have all you want with opensource OpenSim–but for those of you, like me, that like something with pictures and some summary, we are out of luck. It is more of a Ouija-board process with the developers to understand what it is, does, might be. And of course things are also changing as new releases flow through–but getting a roadmap certainly would require a crystal ball and some incense.
OpenSim Issue 5: Admin Tools
We have made quite a few tools to track users, manage resources. I think we should even be able to share these (once we write-up what they do somehow). I have seen that there are some basic tools also floating around to make some reports/alerts, but these are quite simple. I would hope to see more user info and logging–especially relating to inventory and server memory.
OpenSim Issue 6: Voice
Vivox apparently has a OpenSim implementation. I have not seen it, but the gang at IBM seems to be messing with it. It would seem to make sense that it works. It is not free, so there is more of a cost consideration than a functional one. We have been more focused on lectures and talk-show use cases and are using skype-to-stream solutions which works fine and also has the advantage of being broadcast to the web.
OpenSim Issue 7: Mesh Support / SL Viewer
Being able to deploy mesh assets is also a gap on the SL grid (although not on Blue Mars). Main points about this are 1) more cost-effective production for sculptural content 2) more realism for objects. I think Mesh will also need careful design and usage so as not to kill performance, not of asset serving, but of the user experience with viewing/streaming. Blue Mars looks great, but has apparently great than 1 gig client with all the meshes inside it. The Mesh topic also hits on the Issue of OpenSim not having a native viewer. All the viewers are basically SL viewers in the first instance. So there is a lot of overhead, random issues and workarounds to get the SL viewer to work — but our guys at Imprudence have done it for us! They rock! And even the new Imprudence code we are using on HK/Rezzable viewers works much better in the last 3-4 months.
OpenSim Issue 8: Concurrency
Concurrency on light interaction areas we think will scale out at 20-25 concurrent on a region. Now, this is a lot different than on SL for a couple of key reasons. First if you are running your own server, a region is less of a cost issue than it is on SL. So you can have more regions each with less content. I think the main focus is on server-loading then. We are running as much RAM as we can get on a machine and hope to see server concurrency at 400-500. The best news about OpenSim vs SL here is that you can have cross-region tps scripted. If you visit our King Tut Virtual you will basically visit 6 different regions on OpenSim that was just 1 in SL. So on a cost-basis OpenSim has better cost per concurrent user than SL by far (SL $300/month for 50 concurrent and OpenSim $350/month for 400+ concurrent).
OpenSim Issue 9: Scripting
We just had to unravel some scripting to get a new project to run with even a target of 10 concurrent. The good news is that the OpenSim modules allow for server-side coding instead of scripting. The bad news is that coding is more complex and of course see Issue 4 about about lack of docs. Scripting will still be very important, but I am hoping we can get a lot done in the code and create modules/integrations that way. We are making more Bots now that so far seem to work nicely. These Bots will access our CMS and have AI features for NPC interaction (= they will do smart stuff with our existing .com content). Again, benefit here is that Bots will be also accessible from the web pages, iPhone etc.
OpenSim Issue 10: Cross-Grid Movements
Maintaining identity, roles and assets is still the biggest isssue holding back larger adoption of the OpenSim platform. We are aware of other work and have our own initiatives in the area. I would think that within 6 months it will be a few clicks to move an avatar with inventor between grids effectively. I can see where a facebook connect will work well and I also am hopeful that DRM issues can be addressed as well. Right now we are testing mechanisms to move avatar between our own grids (ok, that isn’t too sexy, but it proves a point).
So there it is. Stay tuned we are working on these issues as are many other OpenSim project, developers and enthusiasts. I think a year from now it will be even a bigger step forward than over the last 12 months.
