Can you or should you do a "seamless" world?

[Trippy]

If I understand the OP correctly, one of the biggest problems of totally seamless worlds that have a lot of variety is your computer's RAM. It takes a lot of RAM to load all the game assets for the entire world at once.

They don't load all the game assets for the entire world at once.

[jasonKomsa]

Yes, I know Trippy. Which is the problem I thought I stated :(

At some point the art must be loaded, hence a loading screen when your character walks into a place that does not have the assets loaded.

If there was more RAM available on an average computer more assets could be stored before the game actually starts, reducing the amount of loading screens seen throughout a play session.

[Yegolev]

You can avoid loading screens in that way if you do something like GTA, where you load the world piecemeal as you move through it.  I'm not suggesting this as an improvement.

[Nebu]

I'm sure that this has been stated a million times in a million similar threads, but a seamless world just isn't realistic in an MMOG unless you have so many shards that you can restrict the population.  Everyone fighting for every resource all the time is what made instancing popular.  There needs to be some mixture of instanced areas and open areas to keep the game playable for a larger playerbase.  If you wanted to limit shards to a set number of online players per world, that could work too... but paying customers don't take kindly to queues.

[Soln]

I believe there's at least 1 professional DBA in the f13 community (me forgets) who can comment, but the main issue is access I think.

[Sairon]

I believe it has a lot to do with scaling. If one player needs to inform the world of his actions at a rate of lets say 1kb/s. Then if there's 100 people standing in the same area everyone of those need 1kb/s from the others, totalling 10 000 kb/s of bandwith, or 100 kb/s per person. Depending on how close they are to each other you would face the same problem with collision detection, if the game has that between players. There's also issues with processing time. If you divide the world into zones you could divide it between multiple servers fairly easy. If you want it seamless I guess you'd have to either go with a really advanced connection system where the server park throws the connection around and share data with each other. Or I guess you could use some form of processor clustering. Of which I'm guessing the later is a lot more expensive to upgrade. Anyway, I'm mostly practicing theorycraft since I haven't done a seemless MMO world specifically  .

[Typhon]

The Hoax/tazelbain argument/dichotomy (ok, tazelbain wasn't really arguing with Hoax, but let's pretend he was. also, I said "dichotomy"!) is the most pertinent one for me.  If you want to be the hero/important a greater amount of the time, the more shards are better (with a single player game being the logical extreme where you are the hero/important the whole game).  If you want to be the hero only if you really deserve it, or simply thrive on being in a game with 25000 of you close friends, a single shard is the way to go.

I guess that it's possible that someday hardware/software might allow for a game to be written where the NPC population was much greater then the PC population, and the PC characters had much greater capabilities then NPC characters, and the NPC character interaction trees were sophisticated to the point of passing the Turing test, so that every player would feel heroic/important for a large percentage of their play time (while still allowing interaction with other players).  A game like that will be the most addicting thing ever created and will change humanity irrevocably (if not completely destroy it).  Course, that's just my opinion.

[Soln]

It's probably not worth mentioning, but one thing about  a "large" world and the ability to manage it I would think depends on how much an illusion of space you can create.  I am pretty sure that part of my FPS problems with EQ2, for example, is that when I have my anisoptric filtering up high, all the mountains etc. render but they are there AS 3d models.  I would've though there'd be some economies if they could be painted with the illusion of 3d.  I wonder if that's not possible or beyond the basic MMO PC realm.  I do know that CCP has a real advantage -- with limitless black backgrounds and wispiness of gas clouds they have a much easier time manking things feel BIG.

[Typhon]

it's all about voxels these days

(poor attempt at mixing the Graduate "ballbearings" reference with a technology that has always sounded very good on paper... you know your joke blows when you have to explain it up front)

[pxib]

It's probably not worth mentioning, but one thing about  a "large" world and the ability to manage it I would think depends on how much an illusion of space you can create.  I am pretty sure that part of my FPS problems with EQ2, for example, is that when I have my anisoptric filtering up high, all the mountains etc. render but they are there AS 3d models.  I would've though there'd be some economies if they could be painted with the illusion of 3d.

Guild Wars (ironically NOT a seamless world) pulls this off spectacularly with a number of goofy tricks:

1. Disneyland geography. Hints of expanse wherever they can be stuffed into otherwise crowded spaces.
2. Steep valleys, sheer cliffs.
3. At a particular distance everything literally turns into a painting. Trees, buildings, dunes, mountains... whatever.

So unless there's something in the way (like a sheer cliff) you can always see all the way to the horizon. Nothing ever fades out of view, it just turns into a painting and recedes into the distance.

[Margalis]

I would've though there'd be some economies if they could be painted with the illusion of 3d.

MS R&D had a go at this in the late 90s, where at a certain distance the 3D model is turned into a 2D texture. Looked like total garbage in even the simplest of demos.

One thing you can certainly do is model the far-off things at lower resolution, or use some sort of Dynamic Level of Detail scheme. It is quite possible to show objects in the distance without taking a huge perf hit.

Another trick is to mark really big or important objects and render them in the distance but leave out other junk. For example if you have a huge mountain with a house on it, render the mountain at a distance but don't bother with the house at all. Essentially you mark the mountain is big enough to render from far away but don't bother with the house. Note that it doesn't have to be just big, you can also do it for things that are more/less important. If that house is really important maybe you want to render it from afar as well, even if it is only 3 pixels tall at that point.

Because of floating point precision in zbuffers and things like that you need to do the rendering in multiple steps but that isn't a big deal.

[pxib]

Looked like total garbage in even the simplest of demos.

...and admittedly, if you actually watch the transitions in Guildwars they look pretty crappy... but it looks good in screenshots, or whenever your eyes stray up from combat and take in the sights.

[Stephen Zepp]

I would've though there'd be some economies if they could be painted with the illusion of 3d.

MS R&D had a go at this in the late 90s, where at a certain distance the 3D model is turned into a 2D texture. Looked like total garbage in even the simplest of demos.

One thing you can certainly do is model the far-off things at lower resolution, or use some sort of Dynamic Level of Detail scheme. It is quite possible to show objects in the distance without taking a huge perf hit.

Another trick is to mark really big or important objects and render them in the distance but leave out other junk. For example if you have a huge mountain with a house on it, render the mountain at a distance but don't bother with the house at all. Essentially you mark the mountain is big enough to render from far away but don't bother with the house. Note that it doesn't have to be just big, you can also do it for things that are more/less important. If that house is really important maybe you want to render it from afar as well, even if it is only 3 pixels tall at that point.

Because of floating point precision in zbuffers and things like that you need to do the rendering in multiple steps but that isn't a big deal.

Only thing I would add/correct is that just about all engines nowadays support one of two forms of LoD, and just about all have a last level of a 2D texture on a two poly surface.

I could possibly see some games deciding not to use the last level of "billboard" detail, but just about everyone I know of does.

Now, you did bring up precision errors in zbuffers, and that's a really critical point because basically the farther your view distance, the less "room" you have for precision in your zbuffer--which leads to really bad "which shape is in front of the other" issues--massive jitter at shorelines since the distance between water surfaces and the terrain underneath is so small it can easily tunnel under your zbuffer precision.

I'd be very interested in seeing (never happen of course) how Vanguard corrected for that--tech, or scene layout, since that's not something I've heard mentioned (shoreline jitter, etc.) in the reviews.

[Alkiera]

The Hoax/tazelbain argument/dichotomy (ok, tazelbain wasn't really arguing with Hoax, but let's pretend he was. also, I said "dichotomy"!) is the most pertinent one for me.  If you want to be the hero/important a greater amount of the time, the more shards are better (with a single player game being the logical extreme where you are the hero/important the whole game).  If you want to be the hero only if you really deserve it, or simply thrive on being in a game with 25000 of you close friends, a single shard is the way to go.

I agree that this is one of the major issues.  There is a certain issue of matching up the powerlevel of the PCs with the nature of the gamesystem and the expectations of your IP that have been problematic in the past.  SWG, for example, had a good game system for playing 'random guy in the universe', and the size of the world/shards pretty much supported that.  The problem was that it was advertised as 'Come be the mega hero Jedi!' not 'Come be a moisture farmer!'.  Even when they tried to explain what they expected, players have this expectation from SW-related stuff that they will be playing the heroes of the story, not storm_trooper_0120542.

There is a ratio of world size to PC population that should influence both game system design and your PC powerlevels.  In a 'powerful PC' game, like EQ, WoW, etc, you generally want a very 'hero's journey' plotline, with a goodly development curve, and to support this, you want a relatively high number of players per land area.  The players need other players, because NPCs are either too weak or too busy to help you.  For lower powercurve games, like (original) SWG, you need room for the populace to spread out, so they can do their own thing.  Moisture farming and town construction take up lots of room, after all, and you want the vast wilderness for explorers to wander in, looking for the best place to put said town/farm.  Unfortunately, much of SWG was too flat and terrain had weird effects on movement (no falling damage nor jumping).  I also agree with the person that commented on steep cliffs, rivers and waterfalls, etc, being far more interesting that just endless rolling plains.  However, such objects also take up room, so you need even more space to allow for those, as well as the open spaces needed by the game style. 

UO ran into this issue, where they intended on having you be 'Joe Random', and it was marketed that way, but they had far too large a population for the amount of land on each shard, such that the 'wide open spaces' were all filled with housing.  With the world crowded with people, fighting over mineral deposits, housing plots, etc, you ended up with the same kind of adversarial environment you get on EQ where people are fighting over epic mob spawns.  The EQ problem was one of server population as well, because there were so many people in the 'late game' and the late game was so small, that it was difficult to catch mobs before they were killed.  In early EQ, there were just two big targets, and they were difficult enough that you needed a large percentage of the current level-appropriate population to take them out.  Thus, there wasn't much competition until the mob was dead, and you tried to figure out how to split 3-5 pieces of loot with 70+ people.  Again, this pointed to the 'you are specialer than everyone else' mentality built into EQ.  If you won a piece of dragon loot, you could lord it over everyone else for weeks.

Part of the reason people push instancing is that they want to write quests where 'you are the only one who can save us, $insert_name_here', yet allow all 2500 people on the server to do the quest.  Single-player game quests like that don't work well in a public MMO-space, as you're always arriving just as someone else killed the leader of the bandits and rescued the maiden.  It could work in an Eve-like environment, where the likelyhood of another player happening by that particular location to do that particular quest would be fairly low, however.  The PC/land area ratio really defines how 'special' the PCs can be in the world, depending on their powerlevel.

By powerlevel/powercurve I mean things like the difference in power between a 1 day played and 100 days played character, as affected by exp/stat system, items, resources, etc.

--
Alkiera

[tazelbain]

I don't want one shard because I want to sit around the campfire with 25000 people.  I want one shard because I don't want to have to choose a shard a be excluded from playing with other people players on different shards.  The goal should be to bring as many people together.  In my world, WoW would have 1 NA shard and it would be heavily instatized.

[Margalis]

The problem with LOD stuff in general is that it almost always lags a bit behind what you need. The idea that you can use a billboard really far away where it doesn't really matter anyway is a good one, the problem is in practice it usually switches from a billboard to a model too late, and from a low-res model to a high-res model too late as well.

The MS R&D project I mentioned had some complicated bits, like turning the background into an image on the fly, or something like that. It wasn't just degrading from a model to a billboard, it would take the rendered scene, grab the things towards the back of the scene, turn them into a texture and then use those instead of actual models...or something like that I forget...