Server based game augmentations. The transition to cloud. Really possible?

[Billy Idol]

As Gradthawn says, a server can be anything. On Amazon's elastic cloud, one EC2 CPU equivalent is basically one 1.6GHz Opteron core. You specify how much you're going to use it: light, medium og heavy load. Only if you state heavy load is the entire core dedicated to you. In the light case they probably oversubsribe the core by a factor of 4-10.

Amazon's new instances are basically equivalent to Core i7 3Ghz cores, with 3.25 times the performance of an EC2 CPU equivalent. That means you can have 13 EC2 equivalent "servers" running on a single quad Core i7 blade/server, oversubscribe this by a factor of 5 and a single physical server is 65 virtual ones. A single 1U HP DL120 with a E3-1240 Xeon and 32GB RAM can do this for less than $2000.

Cheers

Exactly. So 300.000 is up to now just a number. We need more details for sure...

 

[patsu]

I am not surprised by that number. MS already own a lot of sites. They can certainly reserve some for gaming use. Operationally, they need standby servers anyway. It’s one of the reasons Amazon introduced EC2 (They use the servers themselves for fault tolerance and want to spread the cost to other companies).

I remember MS already deployed a lot of edge servers because they don’t want to use CDNs like Akamai. I think Sony use others’ CDNs largely.

EDIT: Oh yes, they can count virtual servers too; but they do have huge server farms.

 

[Billy Idol]

So at the end, we basically know nothing about the cloud, just lots of PR bla bla?

 

[Gradthrawn]

So at the end, we basically know nothing about the cloud, just lots of PR bla bla?

I wouldn't say that. I'd say we know for a fact MS has proven that they can be good at it, and we know they have a very hefty infrastructure in place for providing hosted services or just simply compute if you want to look at it from that perspective. In other words, I don't doubt, from an infrastructure side, that they can do whatever it is they say they plan to do. Whether its used or not, whether its useful or not, is a different story. But they're in an excellent position to put it to the test.

 

[patsu]

So at the end, we basically know nothing about the cloud, just lots of PR bla bla?

Hmm.... I wouldn’t put it that way. Their cloud technologies are definitely for real.

It depends on what kind of services they want to roll out. They already have XBL authentication, party management, game save sync services, etc. right ?

They may have R&D on super large scale games and gaming services.

EDIT:
It’s one of the reasons I’m curious about Playstation Home 2.0. Whether Sony will kill it or morph it into something else, together with Gaikai and SOE. Actually, I’m hoping MM will be in the loop too. LBP and Home were introduced together under the “UGC” heading.

 

[Scott_Arm]

Looking at Microsoft's data centers for Azure, they house an astounding number of servers. I still think 300,000 for Xbox Live alone is not realistic. But it could be a huge number of servers. They also seem to have smaller content deliver centers that mirror content from the large data centers, so it depends if those are being included as well.

Edit: For example, the data center in Chicago can house hundreds of thousands of servers, most likely. If you look at the news about Azure, even in the last few days, they're building data centers in Australia and China, on top of the ones they already have in the USA, Europe, Asia, South-East Asia. I can find locations for 8 centers located in those regions, plus the 2 that are destined for Australia and however many they are putting in China, not including the smaller content distribution sites.

Edit: Orleans is basically a "job" type system for cloud computing. It looks interesting. Totally avoid the dedicated VM per client model. The Halo team did some kind of testing with it last year. Thanks for sharing that bkillian. http://research.microsoft.com/en-us/projects/orleans/

In Orleans, grains are atomic units of isolation, distribution, and durability. A service is constructed from grains running on servers in a data center. An external request, from a client of the service, is sent to a grain for processing. A grain can concurrently invoke the operations of other grains through asynchronous messages. Grains internally are not parallel; they process a request fully before handling the next one. Multiple instantiations of a grain, known as activations, process multiple independent requests to a service in parallel. Orleans creates multiple activations of a grain to handle simultaneous requests, increasing system throughput, reducing queuing latency, and improving system scalability.

 

[DieH@rd]

Microsoft Xbox Australia spokesperson Adam Pollington:
“It’s also been stated that the Xbox One is ten times more powerful than the Xbox 360, so we’re effectively 40 times greater than the Xbox 360 in terms of processing capabilities [using the cloud]. If you look to the cloud as something that is no doubt going to evolve and grow over time, it really spells out that there’s no limit to where the processing power of Xbox One can go. I think that’s a very exciting proposition, not only for Australians, but anyone else who’s going to pick up the Xbox One console.”
http://stevivor.com/2013/05/microsoft-xbox-australia-on-some-of-todays-lingering-xbox-one-questions/

What a wonderful PR.

 

[expletive]

I wouldn't say that. I'd say we know for a fact MS has proven that they can be good at it, and we know they have a very hefty infrastructure in place for providing hosted services or just simply compute if you want to look at it from that perspective. In other words, I don't doubt, from an infrastructure side, that they can do whatever it is they say they plan to do. Whether its used or not, whether its useful or not, is a different story. But they're in an excellent position to put it to the test.

Agreed. Also, not every request sent to the cloud is a compute task, a lot of the requests will end up being database calls for (compute intensive) results that were pre-computed. So the number of PFLOPS is only a part of the equation when we are talking about "servers."

 

[Kb-Smoker]

Microsoft Xbox Australia spokesperson Adam Pollington:
“It’s also been stated that the Xbox One is ten times more powerful than the Xbox 360, so we’re effectively 40 times greater than the Xbox 360 in terms of processing capabilities [using the cloud]. If you look to the cloud as something that is no doubt going to evolve and grow over time, it really spells out that there’s no limit to where the processing power of Xbox One can go. I think that’s a very exciting proposition, not only for Australians, but anyone else who’s going to pick up the Xbox One console.”
http://stevivor.com/2013/05/microsoft-xbox-australia-on-some-of-todays-lingering-xbox-one-questions/

What a wonderful PR.

Only 40x? That seems really low.

 

[Daozang]

Pretty much every game I've ever worked on could have used more cycles for AI.
Not so much the decision making that's cheap pretty much regardless of the mechanism used, but the construction of the world view.
Having to use rays instead of volume queries being the obvious one though not the only one.
And of course cost scales linearly with character count.
Having to drop to very low fidelity simulation relatively close to the player.

Of course I'm nor sure that moving all that to the cloud is worth the added complexity, and I'm not sure it results in an obvious fidelity improvement to the player either.

Games will use as many CPU cycles as you give them.
For whatever reason people seem to assume that console games are GPU bound, but IME many games are CPU limited far more often than being GPU limited.

I see.
And I can see how a multiplayer game can benefit from more cpu cycles regarding AI.
But why wasn't this implemented in existing mmo's?
The infrastructure is there (relatively), and each server can assist in computation of this kind.
I'm in no way knowledgeable in this area, and I'd like to understand what are the barriers that stopped us from doing it before.

 

[astrograd]

Right. Except in the real world, buying/renting and running servers costs money, so servers are going to be distributed to a few-ish, or as few locations as possible to reduce running costs. Also, they will be distributed according to population density, so if you live out in the boondocks there won't be anything close to you.

Sure. And? Over time that infrastructure improves across the board. In the meantime, sucks for those ppl. Again though, we aren't talking about anything that relies sensitively on the networking itself. We are talking about maybe hundreds of ms of delay here for wiggle room, depending on the task and implementation.

Wow, memory gains. HOW MUCH? Lol. I can't imagine skeletal animation data being very weighty. There's 8 gigs in these consoles, I think they'll manage.

Depends. It could be considerable if you move from canned animations to physics-based animations in the cloud. These machines have lots of memory for sure, but the fidelity of complex physics interactions influencing the way the environment and the many objects embedded within it animate could likely add up. Sure, you could just store that on the disc and RAM. But why would you do that if other aspects of the game already will use the cloud?

Anyway, is it worth risking screwing up your entire gaming experience if you have a laggy connection or the internet is having the conniptions that day? All it takes is one bad router along the way for you to not be able to reach your cloud server. Yes, the internet is supposed to route around troublespots, but guess what, it doesn't (always) do that. And if the trouble is at the endpoint there's no possibility of routing around it anyway.

The games would require a connection to play them. Plenty of games already do this. We even know of a couple next gen ones that have been announced. Does this mean some SP games would require a connection? Sure. And nobody would give two shits about in the real world.

I'd rather say, WHY. What's the gain here? Nothing.

This kind of post makes me think you don't actually understand what is being discussed. There seems to be a miscommunication somewhere. I recommend re-reading the thread perhaps. I'm not sure why the answer to your question here isn't blatantly obvious.

The more processing you move into some nebulous cloud, the more hardware has to be kept running somewhere else, costing loads of money.

Not costing the user anything extra. Not at first at least. XBL's price is staying the same based on their interview questions btw. MS is footing the bill for the infrastructure. That's their investment into the platform and the future, which they see as being in the cloud. I'm sur ethey will want to be the first to offer true, server-rendered cloud gaming akin to a Netflix model where subscriptions subsidize their imminent expansion. And in that model, my bet is ppl would be thrilled.

You can talk up the economic challenges all day long. They know more than you and they made the investment in spite of what you have convinced yourself are impractical challenges to the business model. They have much experience in this area. They know what they are doing on the economics side.

This is supposed to be a tech thread either way. Let the bean counters worry about costs. I think Shifty was moreso hoping to run through an informed discussion about what is possible from a programming pov in order to enhance games via the cloud computing model.

 

[patsu]

I see.
And I can see how a multiplayer game can benefit from more cpu cycles regarding AI.
But why wasn't this implemented in existing mmo's?
The infrastructure is there (relatively), and each server can assist in computation of this kind.
I'm in no way knowledgeable in this area, and I'd like to understand what are the barriers that stopped us from doing it before.

Hmm... If the MMO game needs to update AI for the players and NPCs collectively, they are already done on the servers. It helps to prevent cheating, and is generally easier to do "centrally".

If I read things correctly, the difference here is someone like MS may factor out common game services and run them for the games (via API calls) on the servers.

Secondly, they want to use this for all games... hence, always-connected.

 

[Lycan]

Depends on what they use the servers for specifically.

e.g., For game streaming, Gaikai deploy servers in 300 data centers in US alone:
http://www.joystiq.com/2010/03/11/gaikai-will-be-fee-free-utilize-300-data-centers-in-the-us/

They probably want to be as close to the customers as possible for more reliable and low latency work.

I'm not that inclined so pardon my ignorance, but does a data center correspond to one server?

 

[patsu]

A data center provides physically secured space, connectivity, electricity and cooling to the servers. So they can host racks of servers for one or more customers.

 

[patsu]

Only 40x? That seems really low.

Yeah, IMHO, “X times more FLOPS" is the wrong way to look at it (or market it).

Looking at Microsoft's data centers for Azure, they house an astounding number of servers. I still think 300,000 for Xbox Live alone is not realistic. But it could be a huge number of servers. They also seem to have smaller content deliver centers that mirror content from the large data centers, so it depends if those are being included as well.

Edit: For example, the data center in Chicago can house hundreds of thousands of servers, most likely. If you look at the news about Azure, even in the last few days, they're building data centers in Australia and China, on top of the ones they already have in the USA, Europe, Asia, South-East Asia. I can find locations for 8 centers located in those regions, plus the 2 that are destined for Australia and however many they are putting in China, not including the smaller content distribution sites.

Edit: Orleans is basically a "job" type system for cloud computing. It looks interesting. Totally avoid the dedicated VM per client model. The Halo team did some kind of testing with it last year. Thanks for sharing that bkillian. http://research.microsoft.com/en-us/projects/orleans/

Sounds similar to Apple’s Grand Central Dispatch framework. They are both based on the Actor parallel programming model. It is not mutually exclusive from the VM model though. The servers may still be virtualized in VM, and then request for gaming services from the backend, which may or may not be virtualized too.

 

[eastmen]

Only 40x? That seems really low.

thats almost 50 teraflops . don't think thats low

 

[MfA]

It was bullshit when Vince said it before the previous generation launch, it's still bullshit now.

 

[Kb-Smoker]

thats almost 50 teraflops . don't think thats low

.240 Tflop x 40 = 9.6 Tflop

But really is just non sense.

 

[JB9861]

How would an online Xbox One running BF4 stack up vs. the gaming PC that ran BF4 awhile back during DICE's SP reveal? If say 300,000 cloud servers were up and running, then what things could we see pop up in favor of the Xbox during a technical comparison? Or would only 1st party titles offload to the cloud? I'm confused on how useful cloud computing is for games like Crysis, God of War, Battlefield, etc.

 

[Scott_Arm]

Yeah, IMHO, “X times more FLOPS" is the wrong way to look at it (or market it).


Sounds similar to Apple’s Grand Central Dispatch framework. They are both based on the Actor parallel programming model. It is not mutually exclusive from the VM model though. The servers may still be virtualized in VM, and then request for gaming services from the backend, which may or may not be virtualized too.

Yeah, I just meant they don't reserve a VM per client. Early in this thread there were people suggesting they'd have to do things like dedicate cores per user. That is not the case. The client load is distributed across the VMs running the Orleans application.

I'm not sure how beneficial cloud computing will be, or how often it will be used, but considering people seem to believe Gaikai will be rock solid and universally available, I don't see how some people believe cloud computing on Xbox One is a pipe dream. Instead of processing an entire game in the cloud with low latency controller input and low latency video output, you're processing a small piece of a game but with less latency sensitive requirements.