« User Created April Fools | Main | Werewolves of London? »

Apr 02, 2004



The One World trend isn't that surprising, considering that a major draw to MMOGs is having your actions affect "the world", and many players will be drawn to the opportunity to be a fish in a bigger pond.

This movement should be accompanied by an increase in instanced zones, especially for guild/friends/personal use. Expect sophisticated random geography/quest generation AI to come on line to relieve much of the stress.

The discussion on an upper limit for communities is quite interesting, especially when you consider that in real life this is accepted a priori by proponents of small towns. However, running a VW on one super-shard shouldn't disallow the formation of smaller intentional communities.


Remember that there are fundamental differences between shards and a connected network of cities, towns, hamlets, &c. Travel, information and commerce exists between these entities (extremely fluidly in the modern world) where as shards (until recently) exchanges only information. Certainly player exchange (or companies that accumulate a lot of wealth on all shards) can enable some travel and economic transfer between shards, it still isn't the same as being able to cross the intervening terrain. When game producers glibly claim that "unified worlds are easy" they aren't really saying that simulating an entire world is easy. What they mean is that moving between defined areas is accomplished in a smooth way and that there are enough defined areas to accommodate their player base -- think UO where you could easily travel between shards whenever you wanted to. So, not really simulating a whole world, but simulating a lot of interesting places on a connected graph. Time will tell whether all the spaces between areas of interest need to exist to make the world real.

Staarkhand hits the core reason why MMORPG companies like shards, though: content creation. It is difficult enough (and expensive enough) trying to create content for a few thousand people that can then be duplicated across shards. Whole worlds populated with unique content requires different approaches, such as careful object and NPC creation approaches (why John Arras will eventually be rich and famous), instantiated spaces (Guild Wars and others) or user created content (Second Life and others).


> This movement should be accompanied by an
> increase in instanced zones

That does seem to be the trend: At once a move toward fewer shards persisted (one-world) and many, many shards (but not persisted).


One big pond vs. many small ponds also leads to fewer players able to distinguish themselves. How many of the leaders of the top guilds would find the same satisfaction in being guild number 20, or 40. Or a sub-lieutenant instead of leader?

Design will need to pay strict attention to this. It's a much more subtle issue than simply supporting 10k players simultaneously with randomly-generated isolated dungeons.


big pond vs. many small ponds...Design will need to pay strict attention to this. It's a much more subtle issue than simply supporting 10k players simultaneously with randomly-generated isolated dungeons.

I absolutely agree with this - but to emphasize the subtle, I don't see this is a trade-off of large simultaneous-player-counts per se, as much as the underlying game (and VW) design. For example, EVE-Online, on the high end, is now boasting 8k simultaneous players, and yet in many respects it feels very much like a small-pond place - more so than other VWs (IMO) with smaller sim-counts...

I think one factor has to do with a design that is able to communicate a meaningful (in game terms) sense of space and distance (in fact I recall a discussion of this over lunch at mud-dev, doh!) without converting it into "dead-time" in game play terms. Distance allows for small ponds within big ponds to develop.


"Are shards on their way out?"

Aside from the discussion of parallel servers, I always find it interesting when people refer to different game servers as "shards" which is inequivocably an Ultima Online term which comes from the fiction they selected to justify their multiple-worlds setup.


Not to dispute my friend Cory's wise words, but I think there's an often unstated assumption that having 10x the number of players in a world means having 10x the content or geography. That's only true if player density remains the same, which is more an assumption about the technology than gameplay I think (even gameplay-wise, players will shy away from the implicit penalty of large gatherings when their framerate drops and latency soars).

Higher density _is_ technically hard, perhaps exponentially so and I wouldn't exactly look to enterprise business solutions for the answer (at least not until my online banking site becomes "snappy", if ever). In may in fact be cheaper to solve the problem of procedurally created universes and questing for now.

The bigger question is why do we want 100,000 players in a world anyway? For larger guilds? Complex societies? Bigger wars? Anecdotally, it seems to me the best "battle scenes" are those where a small number of fairly indestructable heroes deal with an unending stream of relatively easy to kill (but scary) enemies--ideally NPCs. So I have a hard time imagining the fun of a "100k war," with 99k players necessarily being grunts and probably relegated to machine-gun fodder.

If there's a compelling gameplay scenario for 100,000 or even 1M players (what acronum are we up to? Obscenely Multi-Player? Incomprehensibly Multi-Player? Infinitely?), I'm sure the technical issues can be overcome. I'm also sure the content issues will be overcome regardless, since "big games" are just too damn expensive already.

So in my mind, it still comes down to design. Can someone pose a fun scenario for 100k people to play?


I've always thought it had less to do with the costs of content creation than the costs of deployment.

Servers have long been able to support larger worlds than what is available, but it has such a significant price tag that it's simply not feasible for the subscription model (unless you want to charge $40 a month). It's a technical limitation not due to technology, but due to affordability.

Until Ultima Online, Lineage, and Everquest, online gaming had only seen limited commercial success, and there was much speculation as to whether MMOGs could be turned into a viable product and revenue stream. The original plan for Ultima Online provided for three worlds and expected the game to be shut down after six months...

With the uncertainty in the MMOG market at the time, the questions about how the subscription model would do, and the excessive development time and costs in mind, it's understandable that everyone chose several small shards over costlier servers which would mean fewer but larger worlds. Being able to scale low-population servers according to how your subscriber base fluctuated is certainly a reasonable model.

Today we see the MMOG model working in action. As we see neverending increases in server improvements and performance, decreases in hardware costs over time, and evolving initiatives to create industry specific hardware like Sun is doing, it will make more and more sense to take a different approach other than the multi-shard concept.


I agree with Avi that density is hard -- disclosure: Avi should know since he was one of the authors of Second Life's rendering engine :-) -- and, in fact, that was exactly the point of my earlier post when I tried to illustrate what was meant by "unified worlds are easy."

As for why you would want 100K people in the same world, it might not be for the big war -- although I think that the assertion that nobody would want to be a grunt is inaccurate -- but out and out, continuous warfare isn't the only game model. The interest of a world like Second Life (like the real world) scales with population, and in fact scales faster than linearly. New York City is more interesitng than Topeka, even if there isn't a war going on.

As for Sun's announcement causing some sea change in the technical limitations of online world building . . . well, color me unconvinced. Simulating more world space is fundamentally about the application of CPUs to the problem at hand. Alogtihmic improvements aside, more/faster CPUs means more space and users can be simulated. Second Life is currently running on a grid of nearly 200 machines, each a commodity Linux 1U pizza box. We are hosted by a high quality colocation facility that provides high quality power and cooling. Despite this, are limitation is not rack space, which bladed servers address, but power, thermal limits, and cost, which blades don't really offer much of advantage of (especially cost, where blades are typically 2 to 3X more expensive than 1U solutions). The latest Intel based systems have particularly severe thermal and power problems.

In terms of the software side, it seems like Sun is competing with a subset of the features of Butterfly.net, Zona, Bigworld, &c, which seems like an interesting decision.


>The bigger question is why do we want 100,000 players in a world anyway? For larger guilds?

One answer is that a single shard or game universe (as I personally I would define a shard as an instance of game universe with different shards being parallel game universes) makes it simpler to provide global dynamic ingame content then several parallel shards. That is also currently one of the biggest limitations when deploying dynamic ingame content as it hard to design such content to be flexible on one side and at the same time have the same result for all the shards. And if you start having different content for each shard, you end with a black hole getting huge staff resources. That is also why here are few examples of shard wide dynamic content (one being UO in 1999-2000).



I often wondered why game companies do not look to mainframes to deploy their games to. While there seems to be a constant trickling of technologies from the mainframe space into the RISC workstation technology space and then to the x86 server space, the trickling is very slow. A mainframe setup has all the power requirements, heat dissipation, etc headaches spelled out (still gotta provide the stuff), it provides unparalleled availability, incredible scalability with ungodly I/O and utilization percentages...

Price, at first glance, seems comparable given the requirements. Is it familiarity with the technology? Something else?


Cory Ondrejka>Second Life is currently running on a grid of nearly 200 machines

How many machines service an individual player?

I ask because back in December, you said that you had a maximum concurrency of over 200. I realise that technically speaking "over 200" could mean anything from 201 to the population of the planet, and that we're 4 months down the line from then, but still, even if you had 2,000 concurrent users that would mean there would be only about 10 per box. That's a great ratio for the players, but it looks a little indulgent for Linden.



We currently support a maximum of 40 users per machine (oh, and the grid that our users have access to is "only" 170 machines or so, with a bunch on deck to handle land expansion. Also, 10 machines or so are dedicated machines to new user orientation, test machines for the developers and liaisons, etc.) Our budgeting is based on customers per machine - not concurrency - and on that basis we're a little over built but rapidly converging to plan. However, you are correct to point out that a result of modeling this way is that the users who are online get a great ratio.

Our current concurrency numbers are approaching 500, so at any given time 1/3 to 1/2 of the world has some activity in it. However, over the course of the day the active machines shift as different residents log in and out. This results in the expected power law distribution, with some machines heavily loaded all day long, while other machines get less use, resulting in an extremely dynamic world and allowing users the choice eiher to live in popular areas or to head for the hills (so to speak). Also, over time you get the expected drift in which machines are the popular ones as trends and fads change within the world.

DS> I will admit to being somewhat unfamiliar with the current mainframe world (having last used VAX clusters a long time ago) but a quick look at IBM's mainframe pricing suggests prices in the $100,000 - $200,000 per machine. Given that the total price for the entire SL grid is less than $200,000, I doubt that a mainframe would make sense. Our grid approaches allows us to easily scale, cheaply replace dead hardware, easily improve the service as CPU speeds increase, utilize the massive redundancy for archiving and backups, &c &c. However, if someone in the know about mainframes can explain why they'd be a win, I'd love to hear about it.


Cory Ondrejka>We currently support a maximum of 40 users per machine

That's still excellent from the user's point of view. EQ would need around 10,000 machines to match it..!

How many concurrent users do you reckon you can support before you need to buy some more hardware? Or are you set for the next 2 or 3 years with this set-up?




True that mainframes cost quite a bit more than a 1U server, and their software does too. Of course you wouldn't be looking at having 200 of the buggers, more like one. These guys are redundant and scalable *within* the system itself; if piceces of it fail that piece goes offline and another one takes over while you replace it, scaling can be as painless as getting an activation key for more CPUs that are there just waiting, or even the same CPUs that are being throttled back.

I wouldn't know the numbers that would make a server consolidation project for 2ndLife viable, but if you can project your revenue stream, computer usage and its cost for the next year or two you might (again 'might', I don't know the numbers) be looking at some money savings and additional reliability and scalability (way more than 40 users per simulator) if you phone up the IBM guys, tell them you want to embark on a server consolidation project and that you need to put together a business case for it, they'll do the work for you and you get to decide if it makes business sense or not.

You did in a roundabout way answer my question and confirm my suspicions. People haven't really looked into the option since they're unfamiliar with it.


Luca>One answer is that a single shard or game universe...makes it simpler to provide global dynamic ingame content then several parallel shards.

That makes sense. I agree with your terminology, btw. A shard is really an "instance" of a VW, much like we've always instanced VWs for single-player or simple multiplayer play. And anytime you've instanced your world into parallel (simultaneous, divergent) copies, it becomes increasingly impractical leverage dynamic content.

However, if the lower cost of content is the main argument for "instancing," than the higher cost of global dynamic content would (in a rational world) eventually tip the balance the other way, and I'm not sure it has thus far. So this may indicate another flaw in the "content" argument, or I may be missing something :)

DS> on the supercomputer issue, I've worked with $1M graphical supercomputers and worked on SL and I don't think supercomputers are necessarily applicable to their problem domain. OTOH, SL has a rocket scientist on board, so maybe he can devise a liquid cooling solution for colo racks. ;)


I look away and suddenly people start talking architectures - that will teach me.

Reading the thread I was having the same thought’s as DivineShadow. A couple of years ago I started to look at this (running multiple virtualised service on a mainframe rather than a set of 1U’s or blades – I also looked at ultra-dense blade stuff too) but unfortunately the project ended before I got to do the detailed biz case but IBM had a good story.

Forgive me for sounding like a IBMer here (never worked for them actually, though I suddenly feel like I’m in a white shirt and dark suit and have a set of golf clubs in the car) but I think there are two options either go Linux on S/390 – where I think you can virtualize a lot of servers, or use their new zSeries that is based on pulling together 20+ machines / instances.

The nice thing about these solutions is that the often provide control across the entire CPU set, rather than having to run 200 separate things, which I know with things like Patrol and OpenView and stuff is not that hard these days but even these tools have their limits and don’t allow flexible allocation of resources easily.

And I certainly would not _buy_ any of the kit. But I do have a background in outsourcing and managed services so I do tend to think that it’s service providers that should be the ones buying tin and cable. Oh and any decent outsourcer would take you existing kit as part of a hand over deal.

But as I say I have not kicked the tires on the new IBM stuff so I can’t personally recommend it – what I can say though is that if Sun are starting to shout about this sort of stuff and if the MMO population projection figures hold good IBM would probably be very interested in throwing a VW on a bit of big tin and seeing how it ran, then again in my experience Sun are always up for playing too – road trip anyone :)


DS> Leaving aside cost of ownership issues -- we have, for example, 1 system administrator who runs our grid, and it doesn't take all of his time -- the advantage of SL's architecture is the fact that we already have the problem of a massive world parallelized. This is normally the difficult part of switching to a grid computing solution and the reason to go for a single, big iron machine. For many of the same reasons that Google uses a zillion commodity machine, SL's world was designed from the start to run on a distributed grid of machines. So, unless the single mainframe could handle the load of our current grid of 200 machines (roughly) then it isn't a cost effective change for us. However, if IBM wants to talk, my email is cory@secondlife.com :-)!

Richard> We're currently expanding the grid based on land demand, which has proven to be much higher than expected since we started selling land. As a result, the grid will keep expanding as fast as users come along and decide that they want land.



I agree, supercomputers are not applicable. But I'm not suggesting they go with a supercomputer. Something like a System/390 (now called zSeries) machine is not a 'supercomputer'. Lets keep in mind that a mainframe like the ones IBM tipically sells (and the ones every bank out there uses) are designed very differently than graphics workstations or clusters. The former being designed with *throughput*, security, reliability and serviceability in mind and the latter for raw compute power. I know no banks out there that have used a large SGI for their back-end, just as I never heard of a Hollywood studio using a System/390 for rendering. There might be a crazy outfit out there, but the capabilities of each put them squarely in different markets.

"SL has a rocket scientist on board, so maybe he can devise a liquid cooling solution for colo racks"

As long as he doesn't design the microgravity fountain pen instead of just buying pencils... :)


For some games the centralized approach might have one additional twist: internet latency. While I have to admire the progress of the internet -I've seen roundtrip times get lower and lower- latency will always be there in one way or another. Distributing and assigning 'simulators' to service a limited internet-topologically-associated segment of your userbase does make some sense for games that need precise timings (there's a term for this I can't remember). Of course, you might come up with some simple methods to improve the perceived synchronicity in these situations, it doesn't take a genius. Then again you might discover someone beat you to the patent office and has "...a chokehold Patent on the Internet Gaming industry" here: http://www.ient.com/patent/patent00.html .
Never fear, you can always join the "aggresive Licensing Revenue Program" (Third slide here: http://www.ient.com/ien_presents_files/frame.htm ). I'm going to start learning Chinese, it seems its going to be the only way to play a game at a reasonable price in the future.


DS>I hope your financial future isn't linked to their ability to enforce 6,042,477 :-) . . . I didn't read beyond the first 10 claims or so, but the prior art on these go back, way back. Netrek is the oldest client-server I can think of offhand. WRT zSeries, the up front cost appears to be ~$500,000, for which we'd have a grid of over 500 machines. While the I/O of the zSeries (especially the CPU interconnects) will exceed the aggregate of our grid, we have an architecture that doesn't need that feature, and our net performance would be superior thanks to 5 to 10X the number of CPUs applied to a problem architected for a grid. That being said, it might be fun to test :-)!



I'm not up for a debate on relative system specs, but I'd be willing to bet that those heat/power issues Cory mentioned are more related to the CPU than the NIC. However, I _can_ make a pretty good case for using high-end graphics hardware on the server side...

Anyway, regarding that so-called 'patent', I expect there's prior art on dead reconing well before 1996, not that seems to matter anymore. Has anyone admitted to licensing this nonsense?


Content for a truly massive world can only be added by the residents. In the real world, the government (the developers) provide some content, but the citezens create the bulk of it. Of course, are we talking about games or virtual worlds? And is there a difference?

As for the single world/shard bit, travel is the key here. If any two points in the world are only a couple of steps away, the distance between them is irrelevant. Travel cannot be free an instantaneous unless we are going to use shards. To illustrate, new york and los angeles are totally different communities. In the average case, people from the two places cannot mingle, but you could travel between them if you wanted. Yet they are both part of the same world.

In real life, you can pay with either time (driving), or you can pay with money (flying). I don't see why we don't have something like this now (except that teleporting as we would need in a virtual world should be proportionately more expensive than flying).

In case anyone important reads this, I'll also toss in that the 'if he can do it, I can do it' mentality is a sinking ship as well. Content that everyone can experience isn't interesting. If just anyone could climb mt. everest, no one would bother.

Realism is the key. People say games don't need to be realistic, but that's isn't true. At the very least, keeping games as realistic as possible and practical, problems that don't exist in the real world (like content and shards) won't exist in our games.



" hope your financial future isn't linked to their ability to enforce 6,042,477 "

It's not. At the same time being on the consumer side I can see companies being extorted with this patent, which damages the business aspect of a game I might enjoy.

WRT zSeries, it might actually save you money... Unsure. You do need to also factor in a System Programmer. Those things require specialized skills to operate day-to-day. That can be a deal-breaker for you.


Cory Ondrejka>As a result, the grid will keep expanding as fast as users come along and decide that they want land

So assuming that the amount of land you need is directly proportional to the number of players you have, does this mean that if SL does become wildly successful and does get the same kind of player numbers that EQ has, you will need 10,000 boxes to support them?



DivineShadow>"a chokehold Patent on the Internet Gaming industry"

A jokehold more like...

http://www.ient.com/patent/patent00.html>The players on each end may be dogfighting in World War II Aircraft as they do in IENT's WarBirds III product.

And as they did in Kesmai's much earlier game, Air Warrior.

http://www.ient.com/patent/patent00.html>IENT's Patent is on using the time information to adjust the graphics of the game on each end of the simulation game loop.

Just as Kesmai did with Air Warrior...



>Avi: However, if the lower cost of content is the main argument for "instancing," than the higher cost of global dynamic content would (in a rational world) eventually tip the balance the other way, and I'm not sure it has thus far. So this may indicate another flaw in the "content" argument, or I may be missing something :)

Well the problem is how we define dynamic ingame content and how it can be realized. One way is trying to deploy the dynamic content through employees (the so called events teams). That type of content has the benefice to be flexible (human driven events) and can be very successful (if you have a skilled team with the right ingame tools) influencing directly the customer retention. At the same time the development of the tools to support an event team and the personal costs of the event team itself can be overhelming.

The other approach is to have some sort of dynamic-static player driven ingame content like factions/guild wars with territory retention (see UO or AO for example), adding static content (players towns for example in SWG) and so on. The problem is that that type of dynamic content is limited, it is complex (it has to be attractive but at the same time cannot be exploited) and it is difficult to judge the impact of the system in the customer population before deploying it.

So one shard = high costs for static content, lower costs for dynamic ingame content (but still newland so high risk when implementing one of the different systems)
Multiple shards = lower costs for static content, higher costs for dynamic ingame content

The problem is simple, dynamic content (independently from the type) is new land with a high risk associated to it. As every risk, it can be high successful or a complete failure.
That is why I suppose the multiple shards model will stay as it is for the next years.


There are a few more 'pros' to shard design - and unfortunately it's still business justification.

Content - As everyone's mentioned, level-based games waste too much content and force player concentrations in mathematically advantageous adventuring zones. Too much crowding, too much content drawing resources that is outside the level range of the average character level.

Customer Service - CS calls due player interactions increase exponentially with population (kill stealing, camp stealing, harassment, etc). Further, more people in the 'community' excentuates a lack of shared memory -- players are unable to recognize and avoid others they don't like. Players on a 'shard' won't call CS as often when they know and can recognize players/groups/guilds they don't like - because they can avoid them.
Keeping 'community' smaller keeps CS calls down. Fewer possible interactions (3000! << 20000!), more 'self-regulation'.

Cost - hardware/architecture costs for server farms are lower in a shard environment. 'One World' design allows many more players to gather in one place than is possible in shard design -and that will tax the hardware (and software architecture) much more. It's more costly to design and build a one-world structure properly. Particularly if we're talking about the difference between 50k online at a time, and 2k. The core hardware of a shard design (player database, etc) doesn't have to ever be upgraded. In a one-world design, more players will stress the shared resources of the servers, requiring costly upgrades and performance tunings, vs simply slapping on another shard.

Scalability - scaling costs with player growth is easy with shards -- Every x000 players who sign up, slap in a new shard. The suits know exactly how much they cost, exactly how many players they need to cover costs, and when they turn a profit. Suits love rigid cost-profit analysis.

Human Cost - the cost of technical staff to install, configure and maintain a 2000 client server-farm with shard-level complexity is high to be sure. But doing the same for 80,000 clients pushes the cost up significantly. Many massmogs seem to be getting away with admins and programmers who've done small/med-network projects, but don't have the (pricey) million-transaction-level experience. With one-world design, a publisher wouldn't get away with such. For x0,000 players, and all the content they'll need, there really isn't an option to bork the architecture, or make performance mistakes.

Redundancy - Shard design also alleviates the problems of the unthinkable. Catastrophic hardware failures, even when recovered from gracefully, lead to longer scheduled maintenance time, and likely a degradation of play. In shard design this only affects the players of a single shard. In one-world design, it affects everyone.

Taken together, not going with a shard-design is almost inconceivable from a business perspective.

And for all that, the only 'Pro' to one-world design: is the game/community intangibles. Not much concrete to justify taking a hardline stance with the fine people who just offered to finance your dream.


Another quick note: one-world design is not inherently cheaper for dynamic content.

Players flock to such content en masse, and will stress your hardware/architecture and create all sorts of lag for each other. (not to mention the usual inter-player personality playstyle friction)

This is already a problem in some games specifically designed around the idea of hundreds of players being very active in a concentrated area (DAoC).

There will be more players to interact with, and more player-events will 'get off the ground' so to speak. But do not underestimate the problems that happen when 500 players show up to a GM event, or a guild 'dueling-night', let alone what's bound to happen with a few thousand.

Such events are hard enough to balance/organize as is (player-run and GM-run). Allowing the probable number of participants to scale by orders of magnitude (10 v 100 v 1000 v 10000) would exaggerate that difficulty massively.


Richard> [when] SL does become wildly successful and does get the same kind of player numbers that EQ has, you will need 10,000 boxes to support them?

:-) Yes, although Moore's Law and other improvements mean that the users per box will go up over time. So, at 10,000 machines we'll have a server farm the size of Google's about 6 - 12 months ago, simulating an area about the size of Chicago. We'll be in multiple colocation facilities, which will mean that some parts of the world will be "closer" to where you live in the real world due to lower latency.


This is all very informative for those of us who don't worry about architecting VWs day-to-day, but I feel like the discussion has focused primarily on the first question "Are shards on their way out?" and not so much on the second "And if so, will we miss them?"

I liked Cory's statement: "New York City is more interesting than Topeka..." Is that always right?

More later...


Gregolas: "will we miss them"?

I avoided that part myself, because it comes down to pure speculation. We don't even have a single existing 'massive' one-world game for a case-study. (Unfortunately I must exclude Eve Online, due my own ignorance of their game and design)

That said, my feeling is that the social trends from watching muds go from 40 to 200, (and newer worlds going from 200 to 2000), would continue. Guilds will become more and more the only 'community' interaction that players will have. There will just be too many 'strangers' to keep track of (outside a handful on a friends list).

I think people will miss the more close-knit feeling of a smaller community, the same way old-timers get nostalgic for our old MUDs where everyone knew everyone else.

As for the New York v Topeka analogy: RL NYC may well have more points of interest than Topeka, but it cannot match the social feelings and interactions of a small town.

And frankly - that's the only difference (to me) that's tangible in deciding one-world vs shard design.

As we've seen above, it's cheaper and easier to do Shards. So if a designer must build a 'New York'-quantity of interesting places -- those places would actually take longer to build for a one-world design*, and fewer people would get to experience them. So for the same investment, a shard could get more points-of-interest.

More appropriately, the One-world would come across as a very crowded Topeka. Whereas the shards come across as relatively empty New York Cities.


From a player's perspective: Yeah. I'll miss the shards. I find there are often too many people playing in most popular game shards already. The idea of competing for resources or being unable to find a relatively quiet area to enjoy the game with a few friends is very discouraging. Zone instancing goes a long way toward addressing this but among the player base I see mixxed reactions. Some people want their world stuffed to the gills (New York) and some people prefer a more open and unpopulated world where you can get to know your neighbors (Topeka). In the long run I suspect this will be more of a design issue than a technical one. A choice developers will make based upon how they want to present their world, which format is a better fit for the style of gameplay, community, and fiction they wish to present.


Does the larger community also invite the vices present in larger off-line comunities? Random violence and such? Does the relative anonymity of the internet, coupled with the layer of game character anonynmity and roleplaying increase the negative behaviors over the positive ones? Or the other way around?


DS>>"Does the relative anonymity of the internet, coupled with the layer of game character anonynmity and roleplaying increase the negative behaviors over the positive ones?"

Very good point I think. We already know that the anonimity provided online inspires people who are fine in every day life to become total jerks (rarely vice-versa). And we also know that, in the real world, violence increases with population density (people per square foot, not number of people total)... so if you were to host all the people who played EQ on a single EQ server (with current world size) would this cause issues of griefing to rise disproportinatlity?

It's all fine and well to say that games with single shards will have to provide more content to accomodate a larger individual player pool (as well as larger overall landmass / play area)... but content in games recently released has been scant as it is... in AC2 there are large area's that are void of anything (including players) except a few low level random mobs... in HZ well over half the landmasses (and notably this is a world that could hold massive numbers of players) are devoid of player nad mobs alike... you're easily able to run in a straight line for 30 mintues without finding a single creature, player, or much of anything.

So content in many games is hurting as it is... (and lets not get started on the whole "players should provide content" argument) with all the players in one world I suspect new ways to generate content will need to be developed.

Warhammer Online (www.warhammeronline.com) is trying some interesting experiements with dynamic mob movements and "nesting" as it were.... I can't wait to see how it plays out.


Will we have in the future a chao theory, AI life, weather forecasting, simcity, simlife, and simfarm of an infinitely massive multiplayer online game on a supercomputer grid network? I say "yes."

Will me miss the "shard" system? I say "yes" and "amen" to that too.

Both will have their place in the pantheon, but one will be more favored at different times.

One trend pointed out by Julian is the "Unreal Estate Boom." There is just some instinctual need to own and individualize property. Put up an empty virtual Topeka or New York and people may start to occupy it and shape it in their image.



There are two losses -- one is the loss for researchers of discrete social iterations of architecturally identical environments. (I think Dev-Researchers might miss this too, though most "shards" are not differentiated. Cf. TSO cities.)

In a switch from shards, we don't lose Topeka -- we just gain NYC. In sharded VWs, we see 5-50 iterations of Topeka. In a shardless VWs, we'll see Topeka, Boise, Jackson, etc., etc., and we'll see NYC and LA.

Another loss is the loss of the parallel world conceit, which is interesting. See, e.g., Borges, Sliders. Do not see Robert Heinlein's Number of the Beast, which is terrible, even for Heinlein:

The book is an embarrassment; it is unremittingly awful... the first I've wanted to shut in a lead-lined cupboard... My (fairly) humble view is that the book says nothing and says it very badly. "In literature," Auden observed, "vulgarity is preferable to nullity, just as grocer's port is preferable to distilled water." "The Number of the Beast --" manages to combine vulgarity with nullity, giving us a species of denatured grocer's port which makes you thirst for some good honest gruel (E.C. Tubb) or even meths (Harlan Ellison).

Also, we may lose the "Mortal Combat" metagame option: battles (not necessarily just combat) of avatars from different shards.

I personally like this game idea :)



why using mainframes is dumb (neat article): http://blog.topix.net/archives/000016.html

4k in one shard and counting! :)


Bruce> 4k in one shard and counting! :)

w00t . . . congratulations, Bruce!

Yes, we watch Google very closely. GoogleFS is very cool and has actually been a major topic of conversation at Linden over the last week.

Sure, SL has 3 orders of magnitude fewer servers, but the general philosophy and approach are very similar -- define the problem to allow a distributed solution and then throw commodity machines at it.


My two cents, as someone who runs a game that focuses on smaller shard size as a primarily design decision. We could grow Meridian 59 into a huge game world that could support a lot of people. Ignoring the fact that we don't currently have enough subscribers to fill such a world, it would take away a lot of what makes Meridian 59 the game it is.

First, to respond to Cory's assertion that "New York City is more interesting than Topeka." This isn't always the case. Perhaps I'm interested in learning more about my ancestors, and I know they lived in Topeka. Or, maybe I live in Topeka and want to find people to board game with. Topeka still exists, and people still willingly (I assume...) live there, despite the marvel that is NYC.

Although, Cory's comparison is not really inaccurate. People who live in small towns give plenty of reasons why they like smaller towns: less crime, friendlier people, etc. A lot of the same thing can be said in games. Meridian 59's community is much tighter knit than other games. With smaller shard sizes, you really get to know people. If a new name shows up on the "who" list, you know this person is almost certainly a new player, not just someone that was never in the same area as you. It's also much easier to administrate the game.

Especially in a PvP world, a smaller shard eliminates a lot of problems. For example, a PKer can't just kill someone then run off to some other part of the gameworld never to be seen again. Well, until s/he wants to terrorize newbies again. It's much easier to hunt a PKer in a smaller world. Also, having lower peopulations of people in a smaller setting means that battles happen more frequently, and you don't have to worry about your clients choking to death on a large number of models. Guild vs. Guild combat in Meridian 59 runs just as fast as being a lone in the wilderness.

But, the biggest reason why shards shouldn't die is because of their use in game design and community management. Each shard can be different in interesting ways. Meridian 59, for example, has the Sacred Haven shard which prevents all PvP. Some people enjoy Meridian's gameplay, but don't like fighting other players. With a single game world, it would be virtually impossible to offer the same setup since you would have to deal with PvP players harassing non-PvP players, etc. Other games have done similar things with shards, notably with different "PvP rulesets". I think shards are very underused aspect of game design, and too often developers lock-step each shard into remaining identical in order to make life simpler.

In the end, shards aren't leaving. And, if they did, they would be missed. Sometimes I don't want the hustle and the bustle of the teeming masses surrounding me. Sometimes I just wanna head to a laid-back place where I have my good friends to chat with, and my familiar enemies to hunt and fight.

My thoughts,


Brian>Although, Cory's comparison is not really inaccurate.

The shard as small town dynamic works because you can't transfer your investment out to the next shard. There are plenty of social institutions that are shard-like, but that still allow limited transfers, e.g., four-year colleges (you can usually transfer course credits), companies and firms (your experience transfers, though you've got non-competes to limit that somewhat), and towns like Topeka (you can move most of your financial assets to NYC, but you can't port your social capital -- which can be a good thing if you were a griefer in Topeka!).

I'm not down on shards at all (I think they're fascinating and like you said, underutilized) and I appreciate your points about lag/rendering problems for one-iteration VWs. But Topeka is a place you can leave freely. Places you can't leave freely are North Korea and Turkmenistan.


I am 'down on shards' a little bit, by personal preference. Shards clearly have prominent placing amongst the available and desireable options. But I think a large, integrated VW can be done - and the games I most want to make nearly require them.

Yes, shards are convenient for easily allowing alternate rulesets of essentially the same game. Yes, they make overwhelming financial sense.
Yes, they neatly sidestep some thorny design issues.

However, my personal preference is along the lines of greglas' keen insight:
In a shardless VWs, we'll see Topeka, Boise, Jackson, etc., etc., and we'll see NYC and LA.

We can address design problems with design solutions. And we don't have to lose out on the small community feel - nor the big-city wonder.


Brian> But, the biggest reason why shards shouldn't die is because of their use in game design and community management.

Shards aren't the only way to accomplish this. SL, for example, allows regions (the land simulated on one of the back end machines, 256x256 meters) to be either PG or Mature, with different Community Standards applied depending upon which one you are in. Also, regions can be either Safe or Unsafe. In Safe regions, you don't have a life bar, you don't take damage, can't be killed, and is against the Community Standards to trap or impede other users (in a physically modeled world, there are *many* ways of doing those things). In the Unsafe area, you do take damage, can be killed, and it's OK to run around shooting each other.

Being able to offer users different experiences is important. However, shards aren't the only way to do it.

Also, the small town vs. big city discussion: small towns *and* big cities appear in the real world, despite our (apparent) lack of sharding. I strongly believe that single, connected worlds are superior because -- if the world is properly crafter -- they allow the users the choice of whether they want Topeka or NYC in a way that shards do not.


Thanks for the technichal discussion, guys. I was able to slog through enough of it to understand why the shirts would want to go with a multiple shard system when setting up a game.

As far as the Topeka/New York City discussion, I have an observation: A person can deal with only a certain number of personal relationships before they are maxxed out. Someone can't maintain a higher number of meaningful personal contacts just because they live in a bigger city. Forget Topeka, I'll bet there are people in Dimebox, Texas that have more friends than a lot of people in New York City.

In real life, living in NYC is interesting because it is scaleable; you have different relationships with your wife, your neighbors, your co-workers, and the stranger who sits next to you at Shea Stadium. The real challenge to any designer who wants to pitch 20,000+ users into one shard is not hardware. It's how to design a game structure that does not degenerate into social chaos.


Cenn wrote: “Just a question that may be better answered by the people here: why are you seeming to "dismiss" the single world idea?"

Ever since starting Eve-online, i have loved the single world idea, and eve is reaching 10k simultaneous users in a single world (no shards). “
There was an interesting thread about the topic on Mud-Dev (even if at the moment the archive is unavailable). Also on Terranova there was a discussion about the topic.

Lets look at Eve-Online: they were able to manage up to date a concurrent user population in a single shard of 12256 Users. I assume they have currently around 40k to 55K subscribers. The had several technical issues when they achieved the 8k to 9k concurrent users mark. Mostly they were software bottlenecks that were solved with an upgrade of the hardware (there were several posts about the topic, see for example eve-i). It means larger infrastructure costs, more content to provide and in the long term it will be necessary in any case to split the world. But while it is easier to fill a small shard with new players, how long would it take to populate a shard with hundreds of systems like Eve-online is with enough players to make it a stable shard?

The comments to this entry are closed.