Fees

Additionally, you failed to address half my points, such as those regarding a CPU spinlock killing the whole cluster. (Mmm grid deepthink). Can you imagine one sim dying, and taking the whole cluster with it?

You also fail to address the implementation costs, let alone those required to maintain such a system (since one computer dying would potentially take down the whole grid, since it would be storing information relating to all sims, if you want to mirror this, this is additional memory thouroughput needed.)

-Adam

/me mumbles somthing about RFC 1149

(For those intrested ftp://ftp.rfc-editor.org/in-notes/rfc1149.txt )

Are you offer solutions or spouting the line "It can't be done"? Nothing is impossible, unless the party in the driver seat has quit before they've even begun..

Have you ever heard of Myricom’s Myrinet? There is the solution to your internal networking problems. 'bout 950/Mbps... Next issue? Lets hammer this out, you give me the problem and I'll give you the solution.

As for internal/external network traffic, the solution for that problem is building servers with two network cards and creating 2 seperate networks. Either virtually or physically so switches don't become maxed.

Just like the Wright Brothers creating the 1st Airplane. You need to breakdown each bottle neck and work out a work around and/or create something new. The task isn't difficult.. The difficult part is getting people who set a goal and work towards achieve it. Those who throw out spats of what cannot be done have already quit and have no motivation to see outside of the box.

Repeat after me... "Nothing is Impossible..."

950mbps is still a long cry short of 30gbps. This would mean a 30 fold memory drop before you even start. Yes it's possible but it is not a good idea because it will be inefficient and unstable.

-Adam

Needing 30Gbps are way off basis.. Your smoking something...

I've been very impressed with SL. Given that they are allowing individual textures, what I've seen of their machine model makes sense. I am unimpressed with sidewalk superintendents who can do it better. If you're so hot, go out, write a proposal, get financing, start a competitor, and offer lower rates than Linden with better services. I'm sure we'll all flock to your new project.

Seriously I wonder if you are taking into account that FTEs are more expensive than servers. That is to say the more cool, specialized and complex your server model is the harder it is to find people to service it and the more you have to pay them.

My understanding is that LL uses a fairly simple server model, and that doubtless makes thier development curve much shallower, allowing them to do more of the cool things we all want faster. It also probably means they have a fairly wide pool of employees, and can hire the best people not the people who happen to know how their obscure architecture works.

If organizing an enterprise like this were as simple as you make it sound, I'm sure we'd all be doing it.

Until your marvelous new startup comes of age, I'm content to pay LL monthly fees for a very well run service that does things nobody else's service does. And I understand that if I want to own a chunk of space on one of their servers, they are kind of going to have to charge me some money for it. It's a fair trade.

Ohh brother, an SL disciple who can't see the forest for the trees.... Making comments with the underlying suggestion that if "it" can be done better, go make it better. Yeah, that would be nice if I had the millions to do it.. Thats just a cop out remark and ignores the issue at hand.

The main jist that I was getting at from the start was that if LL made their design more efficient, the money saved could translate into less burden on players in the form of fees. Is there anybody here willing to say LL's network design of 1 Sim = 1 Server is an unscalable and outdated way of growing a MMOG? Thats the plain and simple that I'm talking about.

1 Sim = 1 Server means that time, money, people, and resources grow exponetially as population grows. After a certain point on a matrix-chart, that design becomes more and more cost prohibitive for LL, who in turn pass on those charges to the users too blinded to see the nitty-gritty details. How much does it cost to a) Buy the Server, b) maintain the server with human capital, c) provide bandwidth for that server, and d) milk the players to pay the bills and make a profit????

My proposal was about redesigning everything so that SimLand exists in a virtual space among many servers all interconnected in a clustered enviroment. The benefits of that setup mean less people need to be hired to maintain fewer machines. Less overhead and a declining cost of revenue on their quarterly reports. Instead of LL selling Private Sims on servers, they would be selling CPU cycles. CPU cycles cost less than new boxes. And the whole thing scales much better than 1 sim = 1 server. Maybe I'm speaking in a foreign language. I don't see why this concept is so difficult to understand. Maybe the disciples of LL just have too much pride to see outside of the box.

Hell, somebody run out and buy LL some Playstation II's so they can get started building a better mouse trap. 70 PS2s will get you 500 GFLOPS for about $50,000. Maybe with that $8/million dollar investment LL got last month, they could buy 11,200 PS2s and get a massive machine cranking 80 TFLOPS, which would exceed the most powerful supercomputer on the planet which is the IBM BlueGene cranking around 70 TFLOPS.

http://arrakis.ncsa.uiuc.edu/ps2/index.php


What are GFLOPS and TFLOPS?

GFLOPS is an abbreviation for Giga Floating Point Operation per Second. This is a measure of the rate at which a computer can perform floating point operations. A floating point operation is a calculation (such as multiply) involving numbers held in floating point format. This format holds numbers as two parts; a fraction (mantissa) and a power of two exponent. It is frequently used in scientific and engineering calculations where very large and / or very small numbers are required. For supercomputers, the FLOPS rate is an important performance indicator.

Alby, you couldnt be further from the truth. Please read my posts - this will not scale better, it will be costly and unstable, and you appear to have forgotten the contents of my posts before replying to them.

-Adam

I agree to disagree.... Because your explaination doesn't seem to carry throughout the industry, other wise IBM wouldn't have built exactly what I'm talking about for butterfly.net to accomplish the requirements of running a MMOG for thousands instead of maxing out at 50ish... I'm sure in your own mind you've thought up the supporting facts to back up your idea that scaling a network as a clustered setup is impossible. But let me show you the light, maybe a photon might break your tunnel vision.

Don't ask me to critize your view of the world. Ask yourself to explain why what you see as impossible being made possible. If it doesn't scale like you claim, then plus help me understand why IBM's setup for butterfly.net is working in strict contradiction to what you claim to be impossible.


Source: http://www-1.ibm.com/grid/announce_227.shtml

The Butterfly Grid for PlayStation®2 Online Gaming
The Butterfly Grid is a new network gaming environment for Sony Computer Entertainment Inc.’s PlayStation®2 that enables online video game providers to reliably deliver state-of-the-art games to millions of concurrent users. Traditionally, online video games have segmented players onto separate servers, limiting the number that could interact and creating reliability and support obstacles. In the first generation of online games, when one server is down, overloaded, or patches are being installed, game-play comes to a halt. With Butterfly’s breakthrough grid technology, the server interaction is completely transparent and seamless to the user – delivering a resilient gaming infrastructure where servers can be added, or replaced, without interrupting game-play.

The Butterfly Grid is hosted by IBM and powered by IBM Dual Xeon Blade Servers running Linux®. IBM’s WebSphere and DB2 software, along with Butterfly.net’s game servers, gateways, networking software and artificial intelligence systems provide an integrated platform for online game development, deployment and ongoing operations. The Butterfly Grid’s OGSA-compliant, XML-based Game Configuration Specification allows service providers to extend the Butterfly Grid out to the edge of the network, enabling dedicated gaming services and networks, voice communications and single sign-on across multiple titles for potentially lucrative new subscription revenue streams. The grid design is ultra-resilient, offering the potential to support over one million simultaneous players from each facility in a 24/7 environment with automatic fail over capability.

I read the whitepaper on butterfly a few years ago, it's completely different to what you are proposing. Butterfly is basically servers-on-demand. Kind of like how guildwars works - it adds new playfields as demand increases in those area's. They dont share one contigiuos space.

-Adam

But it does what SL needs.. And in the end-all, be-all, thats the goal isn't it? Not fussing about how it works, but getting something that works. And 1 sim/1 server isn't the path of redemption for LL. Grid computing, Clustering, servers-on-demand, whatever the case is, its about supporting thousands/tens of thousands thoughout the game seemlessly without limits if hundreds gather at any one point within the game.

As such, it all goes back to my original suggestion that LLs current network design is not efficient or cost effective as it grows into the future (assume there is one). A better way does exist, its cheaper than the current model, and the cheapness of this method will reduce costs on LL and in turn hopefully allow them to rework their fee structure of land tiers. Thats why I even said something in the first place. LLs fees are outragous for what you get... And the reason they are that way is due to how their network is designed. I'm just throwing out a different way of doing the same thing for less money which could allow LL to redo their entire fee structure.

Clustering, Grid, whatever is LL's salvation. Not 1 sim - 1 server.

On the bright side, this thread lets some of the largest egos in SL swing their penises around.

Tee hee!

-AP

No, I want my money! Do not take it from me!

*sniffle sniffle*

LOL... Your are too cool.. Hehehehe

I couldn't decide which bit of this thread to write a reply to, the tech side baffle's me, Antagonistic's response was spot on.... but this quote really bothers me....

Surely for somebody who is poor (in RL or SL) then 'next to nothing' IS 'their fair share'? I'd love to see a reduction in tier costs, but for as long as I'm enjoying the SL experience I'm happy to send some cash in LL's direction. Nobody forces you to take on extra charges. If 'land fee's suck' so badly you could just sell up your land, get a $10 lifetime account and pay nothing ever again Alby

My two-pennorth is - as usual both sides of this argument are probably partly right.

It cannot be good, or necessary, to have 40 people in one sim suffering frightful lag, giving up and going elsewhere, when the map shows empty sims all round them. The absolute equation "one server = one geographical square" must be way sub-optimal.

On the other hand, having every server intimately involved in the workings of the entire geographic area and its comms cannot be right either. It likely adds tons of internal communications, and increases the skills needed for servicing, maintainance etc. Obviously sophisticated redundancy and fallback and reallocation on failure could all be built in, but all this could fail too. Big fancy systems do not I think have a good record yet.

Surely, as always, the answer is a compromise. The load statistics of SL do have a particular characteristic of geographic concentration, due to parties, events etc, and these are not matched in the current model. We need averaging over a slightly wider area - like ,say, a 3x3 sim block, or maybe even 2x2 would help a lot.

Load sharing between 4 adjacent servers would add complexity, but surely there must be existing Linux code out there to partly address this without having to re-invent the wheel ?

So my guess is - a first-step small-scale load-sharing facility might be technically and economically feasible, and would help to kill the appalling event-lag. But anything too ambitious too fast would probably be a disaster on both counts.

Looking at the pedigrees of the LL people, I would be surprised if they weren't up-to-speed on all this, thinking and acting fully rationally, with good knowledge of what is possible/economic. Mind you - I could be wrong. I've seen some appalling organisations which looked great from the outside.

I was watching 60 Minutes about Google tonight. The whole atmosphere of the company appears to be built on Innovation. Always thinking of tomorrow as the way to grow instead of focusing on today. Maybe I should go apply for a job at Google. Seems like their company fits my view of the world. Take ideas, turn them into beta operations, and roll with it. Instead of the way 99% of the world thinks, which is take a new idea, make it live up to a high standard, and not until it achieves that goal will it be allowed into the public realm.

Yes, they give you money. You would like that.

Its not worth doing if your not going to be rewarded for your hard work. ' Those who work for free end up getting used for somebody elses benefit.

Yes, Linus Torvalds learned THAT lesson the hard way. Let me tell ya, doing great things for charity just don't pay...

heheheh If only Linus could have received a royality for each copy of the Linux Kernel. He'd be having B.Gates serving him tea at noon every day.. lol

Poor Linus, too Altruistic in a world of Money, Power, and Greed. Noble man, but just like all the others, Famous and Broke. Instead of Infamous and Rich. '

Thanks Linus for the free stuff... Hehehehe Sucker...