Is SL truly scalable?

As we have numerous game/business/programming/database etc experts out there, I wonder if the SL model is truly scalable to handle more than a million real users, ones who log on and use the system regularly.

IMHO the three choke points seem to be the asset server, the generally required bandwidth, and the lack of copy pointers (as opposed to copies generating new items).

I would maintain that to be truly scalable,

1. The asset server would need to be distributed to each sim. Each item and av would need to have a home sim. Each sim would also need to carry a list of item keys relating items/av's not in its db but which the client would get directly from the relevant sim. Not all sims would need be owned by the same operator. Given the specs for data in and out of the sim would mean that sims could be programmed to run on minicomputers and cope with much higher loads, a necessary prerequisite for being useful to the corporate world.

2. Bandwidth is not infinitely expandable. P2P between all the clients with AV's active on a sim might better application of bandwidth. The sim would probably still need to follow the action to provide the physics and collision messages.

3. Most file systems employ the concept of soft pointer or shortcuts to other resources such as files. Applied to objects, this would greatly reduce the existing exponential growth as copies of objects are made and become completely independant. Would also solve the problem of how to update objects already sold. Perhaps this could form the basis of a rental model for object distribution? That way the maker/sponsor (corporate) would retain total control of the behaviour/composition of a particular object

It seems to me that some major changes to its architecture are required to make SL truly scalable. Minds much more IT literate might care to comment on this. It might form the basis of evaluating the long term viability of future massive 3d environments.

Over to you guys!

Ed

I doubt it. But everyone loves a bit of 'blue-sky' thinking on a slow day.

No! Please No!

I know in the USA bandwidth is given out like candy but in much of the world we have bandwidth caps and need to *pay* for additional content, or lose all broadband until next billing cycle.


What about objects in my inventory? Which sim are they kept in?

I'm pretty sure rezzed objects are already stored in the sim they are in, which is why they can be duped or destroyed by sim rollback.

Well duh, did it ever occur to you to move to another country?

Eh? I'm not moving to another country cause they have cheaper broadband.

Especially not yours.

My country could have your country any day.

Well that's what I would do if I didn't live in Kyrgyzstan where we have free bagels and fiber optics for everyone.

SL doesn't really work well currently. Textures take too long to load. The name search function only lists a hundred people when you put in "john"; think how many it would be ignoring when the population is larger. Sim crossing isn't seamless at present; lots more sims would means lots more people unhappy with the results of crossing sim borders.

As someone who plays SL to create and play with vehicles, the sim crossings are probably the biggest problem I have with SL next to the slow asset server. It would be nice to be able to fly from sim to sim without dropping out of the sky every time I cross a border.

Huh? You live in Elbonia?



P2

Sarria, if that's the biggest of your problems with SL, then that's fantastic.

Scalability wise, I've been thinking for all of about 45 seconds and had an interesting (albeit the furthest pie from the ground) idea.
Howzbout outsourcing some of the land to some residents? I don't have the fattest connection in the world myself but how much bandwidth would actually be taken up if my parcel, and a few others around it, was hosted on my machine? If the amount of throughput is relatively small (and I reckon I could stream a lot of videos 24/7 and not hit my 3 gig or something bw cap) then why not let me host the land and take the strain off the main server grid. If you had a few people doing that... but then you've got hacker issues and the asset server is still creaking at the seams (anyone had inventory glitches lately? ) so I dunno. Just a thought. I'll get back in me corner and be quiet now.

Ps. It would also mean shattering the current 1 sim = 1 machine paradigm so best just forget it eh?

Seems to me that there is a relatively easy way to reduce the asset-server bottleneck without large scale re-engineering.

Why not have more than one asset server, splitting the assets evenly across them, then when a client or a sim needs the data, it looks at each in turn until it finds the asset. The clients and sims would then record the server name in a smallish persistent cache of asset versus location, hence it would be much faster to find the asset next time... a cache could hold thousands of pin/serverlocation pairs in a few kbytes. After all most people spend most of their time in a few regular haunts.

The overhead of looking at a few machines in the event of an unknown asset, must be far smaller then supposed current overload of the server.

All this would only require a simple bit of extra code added to the sim and client apps.

As i remember it, there are currently two asset servers.

Ah. Are they duplicates of each other, or do they store different assets? If so, perhaps the grid needs 4 or 8 - compared to the thousands of sims out there and hundreds of thousands of users, it's not really a big number...

The real important question we're all not addressing is:
"Is this thread scalable?"

Troy

Is Linden Lab scaleable... after all, with the resident figure skyrocketing, the same amount of staff can't possibly cope, can they?

Lewis