SL Scalability -- Issues to address.
|
Morgaine Dinova
Active Carbon Unit
Join date: 25 Aug 2004
Posts: 968
|
10-21-2004 03:41
Yep, having multiple servers beavering away at the workload of a highly-loaded zone which is currently hosting a popular event is certainly possible.
However, it does require the static assignment of servers to zones broken first, because otherwise it would mean that every normal zone would have to run a high-powered cluster just in case it hosts a popular event, and clearly that would be commercially silly.
Break the static tie between servers and zones and you save a collosal fortune in server costs compared to the static approach, and at the same time it gives you the server-side multi-box horsepower to meet the demands of large popular events in any zone.
|
Hiro Pendragon
bye bye f0rums!
Join date: 22 Jan 2004
Posts: 5,905
|
10-21-2004 05:16
From: Morgaine Dinova Yep, having multiple servers beavering away at the workload of a highly-loaded zone which is currently hosting a popular event is certainly possible.
However, it does require the static assignment of servers to zones broken first, because otherwise it would mean that every normal zone would have to run a high-powered cluster just in case it hosts a popular event, and clearly that would be commercially silly.
Break the static tie between servers and zones and you save a collosal fortune in server costs compared to the static approach, and at the same time it gives you the server-side multi-box horsepower to meet the demands of large popular events in any zone. Again, as I've stated, I agree this would be okay for Linden Lab. Perhaps this could be considered a short-to-mid-term solution until SL becomes more outsourced? What LL really needs then is a tool to see which sims are tortoises, and which are hares.
_____________________
Hiro Pendragon ------------------ http://www.involve3d.com - Involve - Metaverse / Emerging Media Studio
Visit my SL blog: http://secondtense.blogspot.com
|
Marker Dinova
I eat yellow paperclips.
Join date: 13 Sep 2004
Posts: 608
|
10-21-2004 05:33
From: Morgaine Dinova It's the static tiling that doesn't scale for mobile objects --- that's a direct property of the architecture.
You actually stated that "There's no reason why multiple servers couldn't host a single sim, either", which means that either the load handling becomes totally dynamic as I've suggested or else the zone is subdivided into a progressively finer (but still static) tiling.
You didn't say which you meant, but I assumed the former (which is why I agreed with you overall), since the second approach has the same terrible scaling properties as the current grid. I am actually more acquainted with mobile and wireline telecomm, which is my area of study and practice. The mobile telephony is also a grid, in many ways like SL is. You have a series of base sites (cells), with antennas, that influence a certain radius and have a capacity of a certain number of customers (which is usually near 40 ~ 50... something like our SL sims). In order to be "mobile" the user must be able to go through cell by cell without dropping the call, or even noticing the fact. There is tremendus overhead going on between a series of nearby cells all competing to take over your call, depending on relative signal strength to your mobile device. Also, when it is statistically determined that a certain base station will most likely cover more than 30 ~ 40 people most of the time... it's the phone company's duty to further segment that base station into 2 or 3 more with the same capacity, but covering smaller areas. If this strategy could be ported to SL's grid, maybe the sims would be smaller and have enough processing power to handle the overhead messaging that occur during avie and object "hand over" between sims. After all, as much and as little as I can know, there are probably to major ways to impliment the system: 1 Computing grid: as is now, with all it's overhead and processing power intensive "hand overs" 2 A very reduced number of "eternally growing" capacity servers: like the whole grid in 4 huge servers, terabyte mirrored diskdrives and 4 processors each with collosal computing power. As the number of inworld sims, avie population and inventory grows, more processing power, disc and memory are added. Handovers are still around, but less than the first architecture, and maybe the system will statistically be able to handle it.
_____________________
The difference between you and me = me - you. The difference between me and you = you - me. add them up and we have 2The 2difference 2between 2me 2and 2you = 0 2(The difference between me and you) = 0 The difference between me and you = 0/2 The difference between me and you = 0 I never thought we were so similar 
|
Eggy Lippmann
Wiktator
Join date: 1 May 2003
Posts: 7,939
|
10-21-2004 06:10
Somebody hold an event with 160 people at a 4 sim intersection and then post the results.
|
Hiro Pendragon
bye bye f0rums!
Join date: 22 Jan 2004
Posts: 5,905
|
10-21-2004 06:53
From: Eggy Lippmann Somebody hold an event with 160 people at a 4 sim intersection and then post the results. Buncha grey blob people? 
_____________________
Hiro Pendragon ------------------ http://www.involve3d.com - Involve - Metaverse / Emerging Media Studio
Visit my SL blog: http://secondtense.blogspot.com
|
Eggy Lippmann
Wiktator
Join date: 1 May 2003
Posts: 7,939
|
10-21-2004 08:42
Eggxactly. Someone do the math for me, how long would it take to download 160 avatars worth of textures, and would it even fit on my crappy graphics card with its limited memory? How about the 800 prims worth of fancily textured and scripted attachments? How many full-object updates per second would I be able to download on my 512k DSL from across the atlantic? Why does it seem to be impossible to convince people that the limiting factor in large gatherings is the client? I mean, its not like Philip himself didnt say it a year ago, and its not like the huge amount of work that has been done on the client-side didnt enable us to go from laggy gatherings of 5 people in 1.0 to routine gatherings of 40+ avatars repeatedly being animated. It's not like we couldn't cram 81 avatars in a single server as far back as the birth of avalon. So obviously there must be something horribly wrong with the servers. Yes. It's all their fault. I mean, how could MY computer ever be a limiting factor in scaling SL? No. My PC is perfect. It can take anything you throw at it and not even flinch. If you put 200 humanoid models in front of me, constantly transforming themselves, my crappy 3 year old computer should very well be able to handle it. All we need is to fix the server code and remove all these needless artificial restrictions once and for all.
|
Artillo Fredericks
Friendly Orange Demon
Join date: 1 Jun 2004
Posts: 1,327
|
10-21-2004 08:42
. . . OK anybody here ever play Planetside (Sony Online) ??? I know there's plenty of differences in server and game architecture between SL and PS, primarily the fact that all the land and structures are static for the most part... but I will tell you that as far as I remember, each continent is one server that supports something upwards of 3000 people. In local battles, I have seen as many as 300+ individual players all shooting at eachother, running over people in tanks, laying mines, flying airships, etc. So I think, SOMEHOW, SL servers will eventually be able to handle that kind of load. There's another project that was in the works that is unfortunately dead in the water, called Rekonstruction http://rekonstruction.com/ that supposedly would be able to support simultaneously over a million players at once. Some of you are proposing a "distributed computing" type solution, where the non-loaded servers take up the slack for the heavily loaded ones, ehh? Sounds like a good idea to me, sounds like a massive restructuring project too... I wish a Linden would address these issues directly here with us, cuz it's definitely an important thing for them to consider for a viable and stable SL future! My 2 cents, Arti
_____________________
"I, for one, am thouroughly entertained by the mass freakout." - Nephilaine Protagonist --== www.artillodesign.com ==--
|
Hiro Pendragon
bye bye f0rums!
Join date: 22 Jan 2004
Posts: 5,905
|
10-21-2004 10:14
From: Artillo Fredericks . OK anybody here ever play Planetside (Sony Online) ??? I know there's plenty of differences in server and game architecture between SL and PS, primarily the fact that all the land and structures are static for the most part... but I will tell you that as far as I remember, each continent is one server that supports something upwards of 3000 people. In local battles, I have seen as many as 300+ individual players all shooting at eachother, running over people in tanks, laying mines, flying airships, etc. So I think, SOMEHOW, SL servers will eventually be able to handle that kind of load.
Let us know when Planetside includes a full 3-D Creation interface, extremely customizable avatars, built-in monetary and ratings systems, customized animations, uploadable textures, streaming music, and a full scripting language, all of which integrated into a physics engine.  Eggy, you brought up a good point with Avalon's opening... what saved that was it had good designers who knew how to make things without having an enormous amount of mb to download.
_____________________
Hiro Pendragon ------------------ http://www.involve3d.com - Involve - Metaverse / Emerging Media Studio
Visit my SL blog: http://secondtense.blogspot.com
|
Morgaine Dinova
Active Carbon Unit
Join date: 25 Aug 2004
Posts: 968
|
10-21-2004 11:36
From: Artillo Fredericks So I think, SOMEHOW, SL servers will eventually be able to handle that kind of load. Nothing in the area that we're talking about is impossible Arti, it's just slightly harder to devise than a noddy static system, possibly a bit more of a challange, and certainly more interesting. The Lindens will do what several people here are saying is "impossible" (but that's just the inexperience of the regular forum detractors showing), because if they don't then someone else will, and good luck to them. That's the market for you. 
|
Chandra Page
Build! Code. Sleep?
Join date: 7 Oct 2004
Posts: 360
|
10-21-2004 12:56
From: Artillo Fredericks OK anybody here ever play Planetside (Sony Online) ??? It was my experience while playing Planetside that in a massive firefight (say, more than a hundred players), the lag was so intense as to make fighting difficult, and flying vehicles nearly impossible. And that's with player models that aren't customizable and have a very low polygon count. I was impressed with what they were able to accomplish, but I don't think the technology is quite up to the task yet. Even over a fast cable modem connection with a good gaming rig, it's not easy to make a 3D interactive world in which huge numbers of players can co-exist in the same area. It's a lot of data to push over the network, and it stretches the capabilities of even the beefiest CPU and graphics card. It wasn't very long ago (less than a decade) that I was absolutely stunned by a little game called "DOOM", whose graphics are woefully chunky by today's standards, and not really true 3D. It had network latency issues, too, but at the time it was over a modem connection. Machines and networks keep getting faster, and as long as Linden Labs can continue to take advantage of the new technology, SL can't help but become a better and better experience.
|
Marker Dinova
I eat yellow paperclips.
Join date: 13 Sep 2004
Posts: 608
|
10-21-2004 13:50
_____________________
The difference between you and me = me - you. The difference between me and you = you - me. add them up and we have 2The 2difference 2between 2me 2and 2you = 0 2(The difference between me and you) = 0 The difference between me and you = 0/2 The difference between me and you = 0 I never thought we were so similar 
|
Eggy Lippmann
Wiktator
Join date: 1 May 2003
Posts: 7,939
|
10-21-2004 13:59
It's very easy to throw wild ideas into the air when you're not forced to consider the consequences. Let us imagine that LL wanted to hire you to make sure that such a system was put in place. Please elaborate how exactly you would go about it. Explore various use cases. Make informed predictions about what parts of the system would suffer and what parts could benefit. Explore the tradeoffs. I'll get you started: Would you split a simulator symmetrically into equal sized areas, each belonging to a physical machine? It seems like the most logical way to do it, I dunno. Would you disconnect 3/4ths of users, and hand them off to 3 other servers? What kind of load do you think this would impose on the asset server? How long do you think this would take? Would they have to go through a black "loading" screen? What about the builds? Would you then divide the simulator's asset cache into 4 and copy 3/4ths of it to the 3 new servers? Or reload it all from the central asset server? Would you then find all plots that ran along the cut, and automatically subdivide them? What would you do with regard to simulator overlap? Do you even know what the simulator overlap is, and why it exists? What about the problems with collision detection near sim borders? Would you let people fall through their build? How would you ensure the positional integrity of existing builds? Would you go through every prim and readjust its position automatically? What would the coordinate system be like for the new servers? What would you do about objects that are linked? What about script functions that do not work across sim borders? What about scripts that explicitly rely on being in a simulator with a certain name? What about scripts that relied on global coordinates? What would you name the different regions? If I was sending email from a script, and it ended up in another server through this dynamic allocation for 5 minutes, would my server receive messages labeled with a different region name during those five minutes? I stand by my opinion that what you want isn't possible without chucking away the code, and the established user base with it, and all their content, and any attempt to kludge such a system into the current architecture would provide no benefit or even be detrimental. Prove me wrong, please. SL is based on a grid computing paradigm, and while it's easy to run multiple empty regions on a server, you are not going to see multiple servers handling the same overcrowded region. Individual regions are not partitionable. The content they hold is not divisible into independent slices. Agent tracking is not separatable from content handling. So please, Morgaine, do explain exactly how you want LL to go about it. With less handwaving and more concrete data.
|
Marker Dinova
I eat yellow paperclips.
Join date: 13 Sep 2004
Posts: 608
|
Ummmm... wait...
10-21-2004 14:17
From: Eggy Lippmann do explain exactly how you want LL to go about it. With less handwaving and more concrete data. Are we supposed to do LL's Job??
_____________________
The difference between you and me = me - you. The difference between me and you = you - me. add them up and we have 2The 2difference 2between 2me 2and 2you = 0 2(The difference between me and you) = 0 The difference between me and you = 0/2 The difference between me and you = 0 I never thought we were so similar 
|
Morgaine Dinova
Active Carbon Unit
Join date: 25 Aug 2004
Posts: 968
|
10-21-2004 14:38
From: Eggy Lippmann So please, Morgaine, do explain exactly how you want LL to go about it. With less handwaving and more concrete data. Oh sure Eggy, just because you say so, I'm going to waste my time giving you even more level-101 explanations which you then don't bother to pursue if the logic points in the opposite direction to your worldview. I can't even get you to agree on the most elementary premise like the fact that SL needs to scale for mobile objects otherwise Philip's millions won't be able to attend popular events. Fat chance of getting anything at all across to you, because you're not interested in exploring solutions, only in blocking them. So I won't bother. I didn't spend decades gaining the experience to be able to propose those things that I propose merely for your personal entertainment. If you can't understand them, go out there into industry and learn something about scaling. Sorry, but I'm going to be helpful only to people who have a positive interest in creating our future. This isn't a forum of pure technical experts and it's a pleasure to fill-in gaps in people's technical information when they have a desire to learn. You neither have the knowledge nor the desire to learn, as you've shown repeatedly. I would defend to the death your right to squat in a prehistoric cave while the world moves ahead around you. But just don't try to block me in with you.
|
Eggy Lippmann
Wiktator
Join date: 1 May 2003
Posts: 7,939
|
10-21-2004 15:34
It's not a matter of how much experience you have in the real world, but rather how little experience you have in this one. It can be trivial to "scale a database". When your system consists of tiny tuples of data stuck in a table, its an easily parallelizable task since everything is independent and follows a predefined schema. If you have a database of usernames and passwords, you can split your table into 26 boxes according to the first letter of each username. Not enough? Split it into 26*26 according to the first two letters. That's more or less what we have at the U of L here. Well, a simulator is not just a database, it has lots of different components and mutual dependencies. It doesnt follow a strictly defined format, it is designed to maximize freedom. A simulator is not something easily parallelizable, because it is not made up of independent parts. You fail to recognize this. Please refute it. I dont see what "mobile" has to do with "large gatherings", either, they are entirely different problems, and even mutually antagonistic, since allocating more servers to a region will introduce more border crossings to the mobile user. Please elaborate on how you want to scale SL. I know your type, Morgaine. I know your type because I am your type. This is why I have liked you from day one and this is why I insist on arguing with you in this forum  We have everything in common: You know a lot about technology at a high level, but seem uninterested in pursuing more details upon request. You believe that you are someone of great value and gifted with the ability to come up with brilliant ideas. You are very good at marketing yourself. You value logic to the point of obsession, and yet dont shy away from using rhetoric when it suits you (*cough* your whole post was ad-hominem, morgie  ). For better or worse, we have all of these in common  Something you dont seem to realize is that sometimes its not logical to be logical. There are external factors to consider that you fail to account for. The need to interoperate with the vast ocean of non-logical people around you, for starters  The need to realize precisely where the boundary lies between making a suggestion and persistently insulting the hard work of others. The truth remains that you have not presented a viable alternative to the current architecture, and instead insist on presenting some vague guidelines that any non-technical person with the right amount of buzzword baggage could present. You are an engineer, Morgaine, not a manager. You are better than that. So please understand the need to present a more concrete plan, and present it. Regardless what you might think I am not a dogmatic person. I merely evaluate what I read with regards to its feasibility, usefulness, and likelihood of getting implemented. I promise that whatever you post in reply to this will be treated in this purely objective manner.
|
Marker Dinova
I eat yellow paperclips.
Join date: 13 Sep 2004
Posts: 608
|
10-21-2004 16:04
From: Eggy Lippmann I dont see what "mobile" has to do with "large gatherings", either, they are entirely different problems, and even mutually antagonistic, since allocating more servers to a region will introduce more border crossings to the mobile user.
They're the same problem, only one handles more messaging than the other. And of course you'll get more borders by splitting, but each of the servers should have less a region to take care of, thus more processing power to perform all the handing over to the other sim efficiently. And mobile cells - the electromagentic influence a base station has over an area - also overlap, just as I already supposed sims must have to overlap. It's how a grid must be designed in order to provide some form of dynamic crossover in between. 
_____________________
The difference between you and me = me - you. The difference between me and you = you - me. add them up and we have 2The 2difference 2between 2me 2and 2you = 0 2(The difference between me and you) = 0 The difference between me and you = 0/2 The difference between me and you = 0 I never thought we were so similar 
|
Morgaine Dinova
Active Carbon Unit
Join date: 25 Aug 2004
Posts: 968
|
10-21-2004 18:29
From: Eggy Lippmann A simulator is not something easily parallelizable, because it is not made up of independent parts. You fail to recognize this. Please refute it. I dont see what "mobile" has to do with "large gatherings" Eggy, if you only knew the full extent of your ignorance on this issue (and I mean that completely literally, the near-zero extent of your knowledge in this field), you would be truly appalled. I did my PhD in parallelism and concurrency, so I'm not taking any newbie nonsense from you on it. And in industry I was part of a team that scaled a national online service from nothing to 2+ million customers, through numerous cycles of decoupling and dynamic load balancing for scalability, at both network and application level. So, your comments about knowledge and experience in parallelizing and scaling are utterly hollow. If you want to find a lack of knowledge or experience, try looking at yourself. And I'm not going to educate you on any of it because you don't wish to be educated. You just wish to engage in debate to satisfy your totally regressive needs. No thanks. As another indication that you have no interest whatsoever in reasoning about this issue, everyone else who responded to my initial posting in this thread grasped the reason why mobile objects are central to the non-scalability of the static grid, and they even rejected your immediate anti-Morg post (are you stalking me?). Yet you persist in pretending to not understand. I'm not falling for your selective intellectual packet loss, sorry. And furthermore, since we actually discussed this issue when you came to visit me, I know that you're faking your inability to understand. From: someone I know your type, Morgaine. I know your type because I am your type. Not even remotely. Outside of these non-technical posts about our personal disagreement, I use "argument" in the formal sense of logical argument, which has nothing to do with arguing. You in contrast are a debater, and it is pretty much impossible to employ logical argument with you because you refuse to accept where logic leads. We proved that when you came to visit --- every time that the logic went down a path which you didn't like, you would refuse to pursue it further. Furthermore, I even pointed that out to you, but it's like water off a duck's back, a logical point just doesn't stick on you because your interest is in debate. I have zero interest in debate, and every time that you try to engage me in it, it is simply time lost. Like now. No Eggy, we couldn't be more different. I am an engineer trying to find solutions to improve our future, whereas you are a purely regressive detractor --- just look at your posts across the threads, and check out your regressive dismissals of proposals versus any constructive contributions. When you can't find a reason for saying "no", you derail the thread in some other way. All I'll add here is that if you actually think that we are alike then this just highlights how you completely misunderstand the mindset of an engineer. The day that you actually start supporting progress then it might create a glimmer of similarity. A remote glimmer.
|
Morgaine Dinova
Active Carbon Unit
Join date: 25 Aug 2004
Posts: 968
|
10-21-2004 20:16
Marker, I did a sizeable stint at a national cellular provider during the time that they were building and launching their brand new GSM network. I helped create graphical applications which processed the engineering data on RF densities and transit latencies and a million other parameters gathered by the fleet of sensing vans. We weren't the data analysis team, but we worked with them to get the tools doing what they wanted, so it was pretty interesting. When we got our first warbly end-to-end service packets, it was like "The Eagle has landed".  Pretty exciting times for us during development and launch. Cells don't help us here though. Cellular phone technology does have some similarity to statically tiled systems and does share some of the same issues, but there are no cells in a dynamically organized architecture. It wouldn't be right to call a dynamic server cluster a cell, because that would be like having a single phone cell covering the whole country, so the analogy breaks down. Smaller-sized cells are just another way of talking about static resource tiling in this context, because they partition the resource supply in such a way that it cannot reach areas where it is needed. My local sim serves me fine when I am at home in Kamba, but it cannot help reduce the load that I cause when I visit Gibson, to use my previous example. It has been partitioned off, it is no longer available for dynamic assignment, and it represents wasted resource when its home avs or objects move elsewhere for an event. The resource isn't able to move to where the demand for processing is, and that means that the completely inevitable growth in event attendance when you expand the population is not matched by a corresponding growth is server resources. Hence the simple summary line of " A statically tiled grid does not scale for mobile resources." --- maybe it's not totally obvious what it means initially, but at least it's concise, and most people seem to have understood. That's why the item about zone decomposition by someone else a few posts back completely failed to address the scaling issue. In fact all it did was to derail the argument by misrepresenting it, which was uncalled for if one considers that the point about mobile objects had been explained a dozen times on this thread. Regarding "cell handover" in the static grid .... When you eradicate static assignment of machines to zones, you no longer have physical zones left at all, instead they get virtualized into just another parameter in the object processing. You no longer have to pass anything from one zone to the next, so as a side effect the problems with transitions between zones disappear as well.
|
Jack Digeridoo
machinimaniac
Join date: 29 Jul 2003
Posts: 1,170
|
10-21-2004 20:49
Eggy's right. Morgain, if your talking about putting SIM's on a big mainframe so as each one draws more cpu as it needs, what make & model # ?
_____________________
If you'll excuse me, it's, it's time to make the world safe for democracy.
|
Lordfly Digeridoo
Prim Orchestrator
Join date: 21 Jul 2003
Posts: 3,628
|
10-21-2004 20:54
I AM SO SMART MY PANTS ARE BACKWARDS
Argh.
_____________________
---- http://www.lordfly.com/ http://www.twitter.com/lordfly http://www.plurk.com/lordfly
|
Hiro Pendragon
bye bye f0rums!
Join date: 22 Jan 2004
Posts: 5,905
|
10-21-2004 22:19
From: Morgaine Dinova Regarding "cell handover" in the static grid .... When you eradicate static assignment of machines to zones, you no longer have physical zones left at all, instead they get virtualized into just another parameter in the object processing. You no longer have to pass anything from one zone to the next, so as a side effect the problems with transitions between zones disappear as well.
I currently work at a network center at a cellular company which everyone would recognize the name. While I agree the cell site analogy is weak and really not applicable, the fact is, Morgaine, you can't make all machines unassigned if you expect SL to expand past LL. If you have all machines unassigned, then you have a sharing of servers by all providers. It's a simple matter of real world ownership. Let me lay it out in a series of steps how I see SL growing. (Granted, this is my opinion based on my analysis alone.) 1. Linden Lab, in the short to mid run, can do whatever shared resource optimization they desire since they own 100% of the servers. 2. At some point, Linden Lab's server farm grows in a pace that will lead to it become impractically huge to manage by one company. 3. As Linden Lab strains to continue to manage all the servers, companies and organizations will demand the flexibility and ownership of their own servers to best provide service to their visitors / customers on the Metaverse / SL. 4. Linden Lab makes the decision to lease out / open source the server code to 3rd parties - the same people as in 3. This results in the decentralization of LL as the primary content server. 5. As Linden Lab is no longer the primary content server, shared resources between companies becomes improbable. "Joe's Autos" server does not want to share resources with "ABC Bank". Each want their own dedicated resources that they will be happy to pay for to ensure visitors to THEIR section of the Metaverse are served. They frankly don't care if other parts of the Metaverse are laggy.And that's the crux of things. When the servers become decentralized away from Linden Lab, the natural consequence is that the responsibility for keeping lag low in the Metaverse will fall on the individual server owners. Ergo, shared resources are improbable. Sharing resources is actually a move toward centralization, not away from it as you assert.
_____________________
Hiro Pendragon ------------------ http://www.involve3d.com - Involve - Metaverse / Emerging Media Studio
Visit my SL blog: http://secondtense.blogspot.com
|
Eggy Lippmann
Wiktator
Join date: 1 May 2003
Posts: 7,939
|
10-22-2004 02:50
I support progress, but you just feed me non sequitur and ad hominem attacks, and claim decades of experience instead of replying to concrete technical questions. Instead, I would like to hear a concrete plan that adresses not only technical issues but social and business ones. Show me the numbers. Estimate performance increases for several use cases and mention what tradeoffs must be performed to achieve it. If you hold a phd, then surely you must understand that scientists put a little more effort into their critique of other people's work than your repeated "OMG THIS IS TOTALLY UNSCALABLE!! PLEASE MAKE IT, YOU KNOW, SCALE! ESPECIALLY FOR LARGE GATHERINGS AND MOBILE OBJECTS, OK??? THX!!!" Anything else will just end up putting you on the same level as the average unproductive forum whiner, regardless of your education and experience.
|
Hiro Pendragon
bye bye f0rums!
Join date: 22 Jan 2004
Posts: 5,905
|
10-22-2004 03:03
Eggy, I hope you're not talking about me, since my post is right above yours ;^P
_____________________
Hiro Pendragon ------------------ http://www.involve3d.com - Involve - Metaverse / Emerging Media Studio
Visit my SL blog: http://secondtense.blogspot.com
|
Morgaine Dinova
Active Carbon Unit
Join date: 25 Aug 2004
Posts: 968
|
10-22-2004 03:38
From: Jack Digeridoo Eggy's right. Morgain, if your talking about putting SIM's on a big mainframe so as each one draws more cpu as it needs, what make & model # ? But Jack, that ridiculous idea came from Eggy, not from me.  A mainframe is the least scalable implementation that there could possibly be, and it just shows how totally unfamiliar Eggy is with the concept of scalability that he even proposed it. Or with economic viability. It's his usual straw man approach --- invent some idea which wasn't proposed so that he can knock it down. The hardware of a scalable architecture for SL would be no different to that used currently, ie. just a collection of Linux boxes, subdivided into 3 or 4 load-balanced clusters or pools to deal with different aspects of it. LL seem to be doing that already in the database area.
|
Morgaine Dinova
Active Carbon Unit
Join date: 25 Aug 2004
Posts: 968
|
10-22-2004 04:23
From: Lordfly Digeridoo I AM SO SMART MY PANTS ARE BACKWARDS Argh. Hehe, I know what you mean, and there's some truth in what you say.  I'm used to living in a very logical world, where we do have huge intense arguments about how to make things work, but those arguments are always either about fact or measurement or issues of logic, and objections are almost incontestable because they can either be measured or deduced in an objective manner, they're not opinions nor advocacy. Everyone is a winner in technical arguments because they are not debates. If one avenue encounters a technical showstopper then everyone has learned something new, and then we're better prepared to focus on the next possible solution. But one thing is a given: everyone wants to find solutions, not just block them again and again and again. It's a matter of progressive worldview, and it usually goes with the territory for engineers because they build new things as a way of life. Unfortunately that's not what happened here, and when faced with non-technical nonsense and willful misunderstanding and repeated derailing of arguments from one person for thread after thread, I've resorted to fighting fire with fire. I don't enjoy it, and I apologize to the rest of you. Yes, my knickers got bunched up more than usual. I guess I just got tired of hearing that things were "impossible" when any half-decent CompSci course teaches them and when clients have been paying me very nice money for using this approach to help them scale. It's pretty sad that we've got to this, but it's not where I wanted to be, it was forced on me.
|