[Opensim-dev] Help and Guidance!

[Justin Clark-Casey]

On 21/02/11 19:25, Umar Farooq wrote:
> Dear All,
>
> Hope you all are good today!
>
> I am a PhD student at the University of East Anglia Norwich, and working
> on dynamic load distribution for achieving scalable and consistent virtual
> worlds. First, of all thanks for all the hard work this community is doing
> for making OpenSimulator a reality as an open source framework. We have
> conducted a survey of the open source frameworks and found that
> OpenSimulator is the best choice for the implementation of our work. I am
> studying the Load Balancer project that was initiated by Mike Mazur et al,
> but we believe in a world without game specific concepts. We (I and my
> Supervisor) are keen to start a project for the dynamic distribution of
> load, using the basic framework and additional work of this community for
> our work that is described below:
>
> Project Description: We have developed and simulated a number of
> strategies and algorithms that dynamically distributes current load of a
> server in to a number of smaller sub-regions (either 4 or 9 based on the
> avatar placement).  We use an aggregation algorithm to get two contiguous
> areas, to avoid resource under-utilisation. This is because some regions
> might have no players at all. We delegate one aggregate to a new server
> while keeping the other with the current server.  The newly selected
> server becomes child of the server that initiated splitting.  Later on, if
> the capacity further increases, sub-regions in megaregions are assigned to
> other servers. However, the levels in the Resource Management Tree are
> kept at minimum by making parent server based on split and not only
> assignment.  The project aims achieving scalability by physical partition
> of the world maintaining the world in a hierarchical fashion, and Time
> Management in a decentralised manner considering only adjacent regions
> sharing physical boundaries (based on inherent properties of virtual
> worlds). We have identified and simulated this work with a number of
> simple scenarios.
>
> What we think:
> 1.	Using a megaregion initially and then splitting it into two different
> megaregions, delegating one to a new server that becomes child of the
> first server.
> 2.	Using OAR files to store the contents and load it one by one in a
> megaregion on the new server.
> 3.	Using serialization concept to transfer the avatars of the regions just
> moved to a new server.
>
> We would be great, if this community help us identify the components
> already developed that might help us and give me a proper direction to
> start with this work (that might be an extended project of Region
> Modules). We are committed to contribute to this community and continue
> this work beyond my PhD work. Thanks a lot for your help and support.
> Best wishes and hope to hear from some you about this proposal.
>
> Umar

Hi Umar - great to hear about your project.  The load balancer project hasn't been maintained for many years old now and 
I suspect that the code is no longer operational.

Much more recent work on performance topics has been done by Intel, though not all this work is currently not 
open-source afaik.  There is a selection of papers at http://techresearch.intel.com/ProjectDetails.aspx?Id=154.  But I'm 
sure that you're very well aware of those since I see both Intel folks and yourself presented at 
http://www.pap.vs.uni-due.de/MMVE10/

Megaregions may well be a good place to start.  However, I believe there are teleporting and OAR loading issues with 
them.  However, these problems may not affect you.  I believe there are Google-able ways to load OARs on megaregions but 
it's not something that I've done myself.

Without having read too closely, I would bet that you'll need additional framework and maybe generalization of existing 
components to do your work.  I would say that we're very happy to accept patches that make OpenSim more componentized 
and open to architectural experimentation.  Maintaining stability is also very good and for me personally, having 
patches that contribute automated tests that confirm this is even better.

Best Regards,

-- 
Justin Clark-Casey (justincc)
http://justincc.org/blog
http://twitter.com/justincc