Talk:AWG Scalability through per-resident subdivision of the Grid

From Second Life Wiki
Jump to navigation Jump to search
  • In various places you refer to user resource allocation proportional to land acreage. You need to rephrase those references to be proportional to prim allocation, as that corresponds to a slice of CPU power and bandwidth. The association of those with land acreage is purely artificial, because land devoid of prims and devoid of people uses no CPU nor any bandwidth, but only disk storage. Since the cost of disk storage for the relevant land representations is effectively zero, 3rd parties will undoubtedly offer vast acreage at very low cost. As a result, LL's acreage-based pricing system can be expected to alter under pressure of competition, ie. to similarly decouple acreage from prim allocation. As your design is looking ahead, it would be best to couple it to something that represents CPU and bandwidth more directly, instead of the very weak and temporary proxy of land acreage. Morgaine Dinova 14:50, 24 November 2008 (UTC)
  • It sure is sad that suggestions such as yours are totally ignored by LL. They need this kind of redesign badly, yet don't even realize it. Morgaine Dinova 14:50, 24 November 2008 (UTC)
  • Thank you for your support and remarks, Morgaine. I'll be updating this proposal around a more pertinent measure of elementary grid processing. As for LL going on with the arbitrary geographic allocation of resources, well, I guess it just makes sense for their business model of being a glorified colocation service, as it appears to make their internal accounting easier and leverage their position vertically. They're probably not ready to become wholesalers of grid processing, and I think their profit is still significantly found in the brokerage of smaller land units despite the Land Store, Estate management features and OpenSpace sims. In any case I'm content to know that there's at least one virtual world project which is following this P2P, owner-subdivided decentralised model of grid simulation with the added bonus of being open-source and free.--Jesrad Seraph 11:21, 17 February 2009 (UTC)
  • I find that even prim count makes an arbitrary processing power measure, and I think LL agrees to some extent seeing as how they made experiments with over-primmy islands with 18000 prim count and more. The thing it aims to rate is really the user experience with complex virtual worlds, so it is really QoS that matters ultimately. With an open source server codebase anyone can tweak the constants to get more prims on the same server, at the expense of experience / QoS. So the prim count, just like the ratio of sqm to prim, is an arbitrary rule of thumb for decent performance and experience, and should remain what it is: a guideline.--Jesrad Seraph 11:54, 17 February 2009 (UTC)
  • Solipsis adds an interesting twist on the model: places that are immediate neighbours in the topology (whatever this topology is) are caches for the content of other immediate neighbours, so that an offline client does not remove any content from the Grid and, because visiting "foreign" clients on average comme from all other parts of the topology, they help balancing the load in crowded situations. This is a clever way of caching content and balancing load by topological proximity instead of just geographical proximity like I suggested.--Jesrad Seraph 12:14, 17 February 2009 (UTC)