[Opensim-dev] Deprecate OpenSim.Grid.InventoryServer and OpenSim.Grid.AssetServer?

Melanie melanie at t-data.com
Wed Jul 8 20:08:49 UTC 2009


Hi,

Justin Clark-Casey wrote:
> But the real question was about your statement
> 
> "But changes are planned as we are moving to more sane protocols."
> 
> source: https://lists.berlios.de/pipermail/opensim-dev/2009-July/006992.html
> 
> Who is the 'we' in this?  What are these protocols?  Why are they more sane, etc., etc.?  This is an entirely different 
> question to generalizing the OpenSim grid servers.  Perhaps they were not meant to be mixed up in this.

"We" is all of us, the project, for one, and Diva and I as the devs 
driving this change, too.

Today's wire protocols are not sane. There is no point in 
transferring ALL the user's inventory to EVERY region visited, just 
to get the root folder ID, which is the only thing needed from that 
potentially HUGE blob.

Just to mention one known bit of insanity.

Another part that is not sane is the user services. They aren't 
natively equipped to handle the concept of no authentication or HG, 
or user levels, or scopes. They mix in data items that don't belong 
together just because Linden did.

Assets were already made RESTful and so the asset protocol was 
preserved unchanged.
The grid server protocol is a lean one and changes will be minimal 
(probably just a XMLRPC->REST conversion if they're not REST already)

Presence is totally insane again. It needs to be ripped out and 
redone, now that we know more about real world demands large grids 
place on the servers.

With the modular architecture, that is a simple as snapping in 
another connector. so if your grid uses a new RESTful gridserver 
protocol, you just use the RESTGridConnector rather than the 
XMLLRPCGridConnector. The service providers and consumers stay the same.

The monolithic servers can't cope with that, so they need to go.

Melanie



More information about the Opensim-dev mailing list