IBM makes a strange choice
The International Herald Tribune discusses IBM’s new game server platform:
International Business Machine’s new video game server blends a mainframe computer with the company’s Cell microprocessors. The result is a server system capable of permitting hundreds of thousands of computer users to interact in a three-dimensional, simulated on-screen world described as a “metaverse.”
The machines will be priced beginning at hundreds of thousands dollars.
Uh, does anyone really think the future of online worlds is in Big Iron?
If anything, it’s on Big Iron now, with dedicated grids and clusters with rigid architectures and seamless environments. We tested UO on 8-way Suns; Shadowbane ran on one of those big Apple server boxen for a while. The future trend is away from Big Iron, just as it is from Big Media and Big Development. The killer app here is worlds that run on part of one box, alongside Apache — not worlds that require dedicated huge expensive hardware.
The article goes on to say:
There are already massively multiplayer games that support hundreds of thousands of simultaneous players, but the IBM system will add an unparalleled level of realism to visual interactions, Meyerson said.
He argued that in addition to gaming applications, this kind of technology could be used to enhance the performance and scaleability of existing virtual worlds like Second Life, an Internet-based service that crosses the boundary between online entertainment and workplace collaboration.
Servers add realism to visual interactions?
Is “realism” what people want anyway?
33 Responses to “IBM makes a strange choice”
Sorry, the comment form is closed at this time.
I am a big fan of distributed computing.
Yet, state consistency and synchronization issues in largely distributed systems is a big challenge. Particularly so when you have zero guarantees on the connectivity, performance and reliability of the peers.
At the end of the day, it’s either fat pipes or Big Iron.
I’m waiting for fat pipes.
You know I’m all for cozy worlds on cheap boxen, but sometimes you really do just want to host the Rolling Stones (with bling hair) and 50k of their biggest fans in One Big Space.
Okay, I’ll bite: The future of online worlds lies in centralized control. If you have Big Iron, you automatically have control of the data, if your framework is distributed, and you don’t have control of the hardware, it’s much harder to maintain control of the data. No control of the data, and you can’t safeguard it against those who will screw with it to gain advantage over their fellow players, leading to the same kinds of problems you get with NWN, Dungeon Siege, and every other attempt to make a distributed framework fantasy world.
So, is Big Iron automatically neccessary? Not absolutely, but the only effort to use P2P principles for the fundamental architecture (WW2O) wound up having huge problems from purely technical causes and gave up, centralizing the servers.
It’s worth noting that there *is* a game operating on IBM’s Grid architecture: Eve. Maybe if TPM catches on and we can count on the client not screwing with the system, the other technical problems will become solvable. But we’re *years* from that, and frankly I’m rooting for the bad guys (I don’t particularly want my hardware to be trustable by the RIAA, MPAA, and Microsoft).
–Dave
Cool things have been happening in P2P networking, if you’ve got a core game and then a cozy community/meta-game around that, you can really make that part of your whole business model/social gameplay mechanic. And its so much cheaper. If you want 3d, persistent spaces, then yeah, this makes sense, but we all know that market is overinflated.
Anyway, IBM is going to do stuff like this because that what IBM does. I don’t think its a strange choice, unless you consider being a nearly-century-old international mega-corporation a strange choice.
So, to address the objections one at a time:
– state consistency and such issues arise because of trying to do monolithic simulations across distributed systems. That’s an assumption that does not necessarily have to arise.
– hosting the Rolling Stones with 50k could actually justify a computer of that cost. 🙂 But again, there’s the assumption that it is preferable to have the 50k in one place.
– I agree on the centralized control issue; but data can be centralized independent of the simulation. Also, it again assumes a single monolithic simulation. Also, don’t forget that stuff like Open Croquet does in fact work and do P2P MMO architecture. 🙂
What does “50k” mean? Avatars or connections?
It would make a great client computer!
I think to understand why IBM/Sun are doing this, you need to see the world from their perspective. They both used to be “the place” to go for a server… and then came Microsoft and Linux and cheap replacable server hardware from Dell.
So, if Sun/IBM still want to be a server company, they know they can’t compete anywhere that Microsoft/Linux/Dell/Etc. are. So they come up with a large machine that only they can build, and then try to find a market for it.
This server has probably been in the works for a few years. However, lately SL and WoW have been getting press, so Sun/IBM marketing got their engineers to say that the same sever (targeted at TV processing) will also work wih virtual worlds… getting a nifty buzzword in their marketing. (Just like everyone had “multimedia” in 1991, and “internet” in 1996.)
Is that cynical enough?
PS – When trying to understand the mind of any corporation, particularly that of IBM or a telephone company, it’s always good to ask yourself: “Would this theory make a good Dilbert cartoon?” If it would, then your theory is probably correct.
[…] IBM makes a strange choice […]
[…] Raph Koster on IBM’s virtual world mainframe: “Does anyone really think the future of online worlds is in Big Iron?” […]
Nothing wrong with mainframe operating systems (though staffing support might be a bit tricky and expensive). They might be more cost-effective and robust, especially in these days of virtualization.
But the Cell processor portion is crazy. No one, except Second Life, wants to render on the server. I haven’t heard anything that would say the Cell chip is great for complex transactional systems.
As far as I can tell, Eve operates on a fairly conventional MMO architecture. There are two “clever” bits that they do do:
1. Since space is huge, it is easy to have a single “shard” and spread people out.
2. Their technical architecture allows multiple (solar) systems on one server and can scale down to one (solar) system on a server (which is good for a space game’s scalability).
3. When they get more than 400 odd folks in a system at one time. Lag and other problems occur. This is beginning to happen more frequently as their are more players in the game interacting in more interesting ways. Player density remains a problem.
I have actually seen nothing that says Eve scales better in terms of maximum players/server than anyone else. I suspect they have a more efficient overall server architecture since they can load multiple star systems on a single server which the standard static shard architecture doesn’t allow.
The whole Eve “single shard” argument is just good marketing.
Doesn’t it depend on what you want to put out there? The question is asked, “Is “realism” what people want anyway?”. The answer is yes, of course they do. Should the idea of a huge world simulation (yes, with fantasy or sci-fi features) be dropped because there’s other options?
Huge worlds that simulate realism have been asked for and drooled over for as long as I can remember. People dreamed about it years ago while playing Dungeons and Dragons. The latest Morrowwind made a PR splash with it. Posters have been asking for it all in between. Yes, gamers want it and you all know this. All the MMO’s advertise a “huge” world. All the MMO’s say “live in our world”, in some form of wording. There’s no question about this issue.
Money, the cost of production, is the obvious reason you Indies don’t want to go (or really, can’t) that route, when you can also sell something other than that. But if all you Indie developers (or even half) got together to make a huge world simulation with fantasy (the much more desired style than sci-fi) world elements, with realistic AI and spawn, with world interaction that’s deep and immersive, with intrigue and mystery and adventure in a game world that makes sense, I’m sure you could get a back loaded deal on the equipment due to the massive following you’d have, together with IBM’s desire to actually sell this equipment and lead a new revolution in design.
That would assume that you guys could agree though, which seems about as likely as, well, peace on Earth and goodwill to all.
Well, it *is* true that each solar system in Eve can and pretty much is treated like a “zone” in EQ1 or CoH. However, the “one server” thing is more than just marketing, all those people are one community, interacting freely (except when more than about 700 of them want to have a fight at the same time). This makes a big difference, there are single player organizations (Something Awful’s GoonSwarm in particular) that are bigger than the whole population of some traditional MMO servers. Eve is about the same size as Camelot was at peak, but where the Camelot population was divided between about 20 different servers (which were themselves split between US and European servers, and subdivided into 3 realms each socially), Eve’s are all in the mix together.
Admittedly, much of this is made possible by both some design slight of hand and by the truly *insane* amount of processing power they have thrown at the problem (something like several hundred high-end blades, and top of the line Solid State Storage Devices for the database). But the social and gameplay effects are very real.
–Dave
People dream of the realism, they think they want it but I honestly doubt realism and incredible server processing power will do much to improve the online gaming space for end users. What it might be able to do is shorten the development cycle for big budget productions, but thats about it.
If you look at the current market and I believe thats where you want to look, rather than desires and wishful thinking, you will probably find that the things people want are such as service, balance, quality and functional incentive structures. (And a nice bunch of other things but lets leave those for now.)
Neither of these require Big Iron investments, they require other things which might be hard enough to get hold of.
Doing things like using the server for processing how the animation system moves vertexes about to make sure that all players view a collission between Player A and Mob B properly is not what these spaces are about. Adding that feature might make some more people talk about your product but it will not pay back the cost of building it (regardless of which tech you use).
As a general principle I would advice against spending Big Money on anything which is abstracted relatively far from the control system of a game, or other entertainment application. Having super simulation algorithms on the server is abstracted farther from the control system than most other ordinary mmorpg features so it’s a dead fish, (or maybe the term Dead Shark applies here?).
The Key Feature of online is other people, that is where you want to invest.
(I could keep on ranting about how some people will develop a hardcore user base around advanced server side simulations which actually might bust my own argument. Establish a niche title in the online simulation space and actually turn a profit on such features but… eww… I’m still suffering mental stress from visiting SL a few days ago.)
[…] Raph Koster argues for a third way (for gaming, anyway), while 3pointD.com has more on IBM’s […]
[…] Raph Koster argues for a third way (for gaming, anyway), while 3pointD.com has more on IBM’s […]
State consistency is a fundamental problem of distributed systems. I don’t see how you can avoid it. Distributed shared memory is already a big challenge. Anything beyond that is a daunting task.
Add latency to the mix and it’s a world of hurt.
Remember the Magid study? Consumers want an immersive and interactive experience. Realism is unnecessary to the actual desires of video-game consumers. Developers and "third-party experts" seem to be unable to understand why the technology used to produce a video-game experience is only relevant to them! Super 8 or Super 35? CryEngine 2 or Unreal Engine 4? Most consumers are unconcerned with the technology used as long as you provide them with an immersive and interactive experience.
There are lots of questions about what gamers want because those Noachian assumptions that “bigger environments, better graphics, faster animations, and ultraphotorealistic cloth-and-hair simulation” ultimately matter to consumers in the long run have led us to nowhere except the remnants of a broken blockbuster business model from the heyday of Hollywood.
I recently coproduced a panel discussion on “What Makes A Next-Gen Game?” (Video available!)
The panel consisted of video-game journalists representing Game Developer Magazine, Play Magazine, Official Xbox Magazine, PSM, mainstream outlets, and Spike TV. After all was said and done about the individual components that combine to form a video-game experience (e.g., graphics, sound, gameplay), the answer to the topic question came down to the value that consumers derive from the experiences provided by video games. That value is not tangible.
Since I’m writing a hefty article on this subject—or why technological innovation and content originality are practically irrelevant to the development of next-gen experiences—that should be published late next month, I’ll leave it at that.
[…] Nick Carr has another post on this, and links to Ralph Koster, who has some ideas of his own.Technorati Tags: sales, […]
The “single world” part of Eve is very different from the technical discussion of a single server.
As far as I have been able to see, Eve cannot handle more folks per box than anyone else in any meaningful way.
Eve has shown something else that is very important. One world allows anyone to play with anyone and has enabled the creation of large organizations… and has gotten some very impressive game play benefits out of that choice.
But we are talking here about servers and technology – and, as I noted above, with the exception of supporting multiple “systems” per server (which is cool and clever), Eve hasn’t made any architectural breakthroughs to support dense player interactions.
So, the two metrics I happen to be interested in today are:
Server Utilization – Peak Concurrent Users / Total Servers
Peak Acceptable Player Density – Peak Concurrent Users / Server – with no perceived “lag” issues (perhaps lag no more than 10% over the lag with Average Concurrent Users / Server)
I think Eve is doing better than most on the first metric, and hasn’t made a notable dent on the second.
Don’t think of Eve and WoW and other classic virtual world designs. Think of Second Life. They have a huge grid of machines to run that. Its the kind of thing Big Iron is good at.
Whatever the original meaning of the quote was, when I hear “render” in this context I think of collision, pathfinding, physics, AI… Some of that kind of stuff can be done on the server.
It’s not so much Fat Pipes that we need, as *really fast* pipes. The speed of light may end up being a limiting factor, here.
[…] Nick Carr has another post on this, and links to Ralph Koster, who has some ideas of his own.Technorati Tags: sales, […]
Morgan, appologies but I don’t think you can see the forest for the trees.
I don’t know what you think they mean by this, but realism (in an exciting world of adventure) is what they are saying they want by these terms. You get these things through a realistic game world. Not realistic in like going to work for eight hours, realistic in like the immersion of AI around them and the interactivity of sitting in a chair. Expand this and you get what players are, have been, and will, say they want.
You’re right in that most players don’t care about the technology behind it, but what does that have to do with it?
That’s on the game developers, not the player’s desires. Lets not argue that these things are important, but they can as easily be used in an immersive, interactive, realistic world as they have been used in the current gamey grind designs.
As far as your Next Gen panel goes, I didn’t watch the video yet because of the time involved. But I am pretty certain that I’d take from this the same thing I always get from these kinds of things. It’s that old tree/forest thing. You guys far too often miss the point. There’s a link missing in your thinking. It’s sort of like magic, adults don’t catch on to the trick as fast as kids do because they have preconceived notions and ideas based on their experiances. In the same way, you folks who have tech know-how have preconceived ideas about making games. And because of this, you simply miss the most simple of things because your mind set is on features instead of the overall picture.
Of course it’s not. It’s entertainment. But what else is this supposed to mean? You certainly can’t provide gamers with that nontangible entertainment without the very tangible means.
There is really no question about it. Players do want realism. They want to sit in chairs, open doors, swim in water, climb mountains. They want food to eat and mean something. They want to be able to pick things up. They want to be able to both run and walk, crawl, to swim under water, to jump accross obstacles, eh hem…to slow down when running up hill. They want weather effects. They want to have pets and train them. They want to swing their weapons, to shoot things, and throw things that bounce off walls. They want to be able to chop down trees, grow crops, sheer sheep, and yes, pick grape bunches or apples from trees. They want to build things, harvest things, collect things, display things. They want meaningful politics, trade, battles, and social structures. They want to dig holes in the ground. They want to push buttons and pull levers, and fill barrels of water. They want rabbits who run from them and trolls who charge blindly into death battle.
Com’on. There is no question about this issue of more realism. No question at all.
Isn’t a more pertinent question: Does anyone think that these Big Iron Worlds will cease to exist?
There’s no point in IBM designing new hardware for home users to run their Web2.0 MySpace3D hosted-on-my-own-machine world on [not where IBM competes directly anyway] (which, conceivably, you could probably already do, complete with green background wallpapers, red text and looping mp3 files harking back to the days of VRML), while there is, and will likely continue to be a demand for bigger iron for virtual worlds and other applications alike, so I don’t see the point in griping about it (other than to complain that a press release is all press releasey).
Consider this though: what sort of computers do you think run the back-end MSN, Yahoo messenger, MySpace web-servers and other services along these lines? I’m going to bet that some of the cores of those systems are powered by some reasonably hefty machines.
Eve’s Jita system regularly has 700 people online, and it’s a bit laggy, but not unbearable (as far as I remember from my last visit). That system is supposed to be dedicated to it’s own machine, but I’ve never heard any numbers from any other game with which to compare. There’s also been some pretty large fleetbattles in systems where the developers pre-reserved the system to a single machine in advance. I think that predictive load balancing is a fairly big factor in Eve (and there’s frequently threads asking CCP to work on a dynamic load balancing system) – it’s fairly easy to end up with a system in Eve that’s usually got nobody in it that suddenly has a 350 man fleet battle and just doesn’t have the CPU headroom to handle it – as the number of subscribers has risen, the average headroom has probably decreased, making it much easier for load imbalances to occur. There’s a difference between not being able to handle X people on a single machine, and not being able to handle X *more* people on a single machine.
I don’t know about IBM’s hardware specifically, but I don’t see any reason why big centralized servers would go away. Many companies host their websites with a hosting company anyway, not on their own box with Apache on it. The hosting companies can share redundancy, CPU, disk, and bandwidth between many sites to get economies of scale that wouldn’t be possible if any individual site were on its own box.
I’m not sure why the company providing the software wouldn’t want to also provide the hosting. It’s a lot harder to get somebody to pay you for something if they have to download and install before they can even try it out. It seems much more likely that the company building the software would sell it as a service and host ALL the worlds, probably with a teaser level of service that didn’t cost anything at all. All world configuration and content creation can happen via custom clients or the web regardless of who does the hosting.
Of course all this assumes that there’s a viable market for selling reusable software for small virtual worlds. I don’t think that’s clear at all.
Just a couple quick comments, since I am on the road:
– Realism and Big Iron have little to do with one another. The thing precluding realism isn’t CPU, it’s will. Realism isn’t driven by deep complex physics sims, it’s driven by plausible behavior.
– The pressures driving current Big Iron worlds are not, barring Second Life and maybe Eve, pushing towards monolithic hardware. Current architectures are driven by commodity hardware for a variety of reasons: power consumption, heat the racks, ease of expansion, ease of contraction when the user surge fades, ease of migration to new commodity hardware as processor upgrades happen, etc. There are a myriad of things that go into the equation. Once upon a time UO ran on Sparcs, and it migrated off for very good financial reasons.
This is a close cousin of the fact that there are many many pressures driving sharding which have little to do with the max capacity of clusters.
The cost and scalability argument against mainframe is valid.
Yet, in the same way people don’t build their own power plants to power their PC, no one says developers should own and operate their own mainframes. Such data centers can be managed by third-parties with huge economies of scale.
I think people love white-box distributed systems because it’s cheap and it matches the requirements of instancing.
Which leads me to my last point…
Yes, “Realism isn’t driven by deep complex physics sims, it’s driven by plausible behavior.”
But, “plausible behavior” is certainly not driven by instancing.
(How lovely to see that wizard come back to life 3 seconds after being defeated, just because another player wants to complete that same quest.)
So, I dream of the EXTREME MMOG, which is made of:
– A plausible 3D world
– A plausible time dimension and progression
– Plausible circumstances
– Plausible behaviors
n players enter the world and live in it until their characters die or the world era ends.
All players whose characters are alive must be present or the game cannot be played.
So, basically, you throw n players into a biotope and see how they interact, evolve and reach the end of the world era.
Yeah, try to sell that to a publisher…
That’s not quite true… There are many people around the world that build their own power generators, such as solar batteries and windmills, for cheap renewable energy that cannot otherwise be acquired due to stringent export laws and scrimpy incomes.
Oh, yeah… I like this quote from this article:
I think ultimately you are correct, when we achieve realism people will ignore the ~40 year journey, and all its blood sweat and tears and start to demand abstractions.
But we’re not there yet. Until then, there is a risk of standards. Once you’ve experienced it, you won’t really want anything less. WoW is so far removed, graphically, from Orcs vs Humans.
Is this the direction virtual worlds are going? I think IBM is shooting for at least one type of virtual worlds that I haven’t heard anyone really talking about yet: virtual worlds for the enterprise. Great for augmented reality environments where the physical workplace is mapped, datacenters are annotated with data layers, and remote workers can teleconference just by walking into a room with a virtual camera and a physical viewscreen.
Heh, when I was at Sony Online, I was thinking about writing a proposal for a similar virtual world. I discussed the idea with a few people in the training department, and I think Raph, too. I was told that virtual world wouldn’t fit the corporate culture.
Well, it does make sense to use cell processors on the server in a computing center if most of your load is off-hours and you get the time cheap… Think: business-use during the daytime, entertainment-use in the afternoon, batch-work at night. Think: moldable worlds with physics (even line-of-sight is costly if the topology keeps changing). The cell might be costly to develop for, but you obviously have a lot to gain from vectorizing your algorithms (assuming IBM can achieve good yields/excellent performance compared to Intel).