p2p Texture Sharing a possiblity???

Just wanted to ask if there would be a way to decrease the load on the servers by having clients in the same sim linking up and sharing textures from client to client instead of just the Server... That way everyone that has rezzed everything in the SIM already can start sending those textures they have rezzed to other's just coming in... I think this would really speed things up... I mean think about it...

You get a TP to a club... When you get there.. Everyone there has almost all the textures rezzed and your the only one requesting them... BOOM about 20 people starting throttling you textures as well as the server... Within seconds you have rezzed the entire area regardless of SIM Lag or texture size... Make it sort of like a Bit Torrent style p2p sharing technique... Seeds and Peers and the like... I mean it works for Linux and everything else... And the coding would public domain so no overhead costs... Just wanted to get some idea of feasibility... I'm also cross posting this in the Linden Answers Forum... If you think this is a great idea and would like me to post it in as a Feature Vote let me know...

Heratiki Turnbull

I believe this has been mentioned before. I think the problem is that the asset server needs to be rebuilt from the ground up so that assets can be managed properly (such as deleting unused textures).

See I had a thought that it might need something like that... This is the reason I suggested it... The asset server (or server farm whichever it is) seems to take the most abuse when it comes to normal everyday runs... But if everything was shared then the speed itself would exponetially increase... Those of us with a faster connection could do all kinds of things to help slower connections get what they need...

Now I had plenty of time to sit and think about this yesterday and came up with a couple downfalls... There are always some... Say with this process implemented everything was going fine... Then someone decides... Hey wouldn't be great if someone corrupted all the items in my cache so that whatever I send out is corrupted... That one is fairly easy to work around as you can do MD5 Sum checks on everything sent out and in which inherently would probably slow things down again... The only other thing I could think of is if someone came up with an undetectable Virus that was somehow laced into their cache that was being sent out to others... Again MD5 Sums would probably catch this but yet again it would slow things down a bit... So there are the dangers I can think of...

Just like all p2p when you put the ability in someone else's hands there will always be a "bad guy" out there... But it's still feasible... Anyways... I'll stop my ramblings... Still checking to see if anyone would be interested in the idea...

Heratiki

Except the individual sims are alleged to have a local texture cache which should supercede direct reads to the central asset server.

The main confound to the theory "the bottleneck is texture loading from the asset server" is that they should first come from client disk cache then sim cache and then only the central asset server, yet you can be in one place, healthy sim, no network traffic and the client isn't even attempting to load missing textures.

I think this points to the renderer. Peer to peer asset swarming would help if and only if the client actually knew what it should be loading, otherwise, you'd be adding a layer of non-trivial code that would likely never be called; best to fix the cache or "interest list" before optimizing that which hasn't been demonstrated to be defective.

But I clearly have no idea what I'm talking about.

As well as the cache and interest list, there's the fake-TCP-over-UDP implementation. The texture got sent but because there's no congestion control the computer's input queue was full so critical parts got dropped. We know this happens because that's the reason for the inventory problems downloading the initial avatar a week or so back: since they're not maintaining a reliable stream the client doesn't even know that a texture was sent and lost.

You know, I think that's gotten worse recently. My router went kaput (again) and my ping times were normal (under 200) yet I was very obviously dissconnected (editing an object to find the owner resulted in a blank space).
I think they've been trying to fix EVERY bug except the all important ones: connectivity, asset management, and texture loading.

Some would speculate that they have been deemed unfixable, or "broken by design" as it is called in the field - put another way, the system design was never expected to have as much load as it is experiencing and things like the "slow-render problem" and the newer "vendors scripts failing to deliver" are merely symptoms.

When faced with an insoluble problem, you can either admit it, blame it on someone else, or try to draw attention away from it.

Oooh, look, a peacock!
(images not permitted in this forum, so just imagine a peacock here)