5 ms·

Cloud Computing - Streaming Video Games

10 points by pkaler 18y ago

swombat 18y ago
The elephant in the room of this whole OnLive games thing is, why the heck are they bothering with video games? If they have a technology that's capable of remoting games (which are some of the most demanding consumer applications imaginable), then the whole "Linux vs Mac vs Windows" debate is moot. If it can remote 3D, fast action games, it can remote whole OS's with ease. That's surely a 100x bigger market than games. That, to me, casts huge doubts on their claims... it's like someone claiming they've solved cold fusion and are using it to power really strong, long-lasting flashlights.
- teamonkey 18y ago
  You can already remote OSes with ease, and have been able to for years. VNC, X11 etc. It's not particularly fashionable, but PC hardware is so cheap you might as well buy a new one. You can easily spawn an instance on EC2 and connect to it, but those machines don't have state-of-the art GPUs. The problem they're trying to solve here is that a new gaming PC can cost a couple of thousand dollars, and you have to keep it on the bleeding edge of expensive hardware to play the latest games. It's typical that even the most avid gamers simply can't keep up with all the consoles and PC hardware.
  - swombat 18y ago
    Most OS remoting solutions I've seen are only just usable. I can remote into my Mac Mini into the living room and even that doesn't feel quite right. A 3D shooter, with this kind of latency, would be almost unplayable. And that's just to my living room! If they can get games to play well on the average DSL connection, they can get OS's to feel 100% local, not laggy, like the average VNC session. I'm sure there'd be a big market for that kind of technology, outside of gaming.
- liuliu 18y ago
  the thing is, OS is in dominant position on the market for not very long time and it was born purely for abstracting the hardware, to make it accessible for upper layer application and have no other functionalities. As all things moved to the cloud, where the OS is running should not be an important question anymore. The real question should be, where the application is running, where my data stored. Get the OS clone running in remote does not serve any benefits for us. You still have to install MS Word in remote and upload your file to the cloud which make no difference to our current status. However, the cloud gaming service solve a more practical problem we will face and we don't have to update our GPU anymore. All the cloud service should solve a practical problem (storage, data management, collaboration) otherwise, there is really no need for it. I hope it will explain why most Cloud OS services now fail.
Angostura 18y ago
I've been thinking about the OnLive business model and the difficulties with the technology/bandwidth and this is what I reckon they are going to do: Sell it to DSL/Cable companies. I sorry I just don't believe that the system will work if it is hosted in some nebulous cloud somewhere, and I suspect that the ISPs will go bonkers about the amount of traffic that it generates - if the ISPs aren't getting a payback. Obvious answer - the ISPs sell it as a service to their customers and colocate the equipment at their exchanges.
- bitwize 18y ago
  Better yet, sell it to restaurants, bars, etc. for those completely noddy trivia and keno games. That's about the only place I can see it really succeeding. I don't want to play even an RPG with laggy, video-compression-artifacted graphics.
greendestiny 18y ago
As complete an analysis of the difference in latency between OnLive and a local game as I can be bothered making is: - player input to server: ping/2 - game engine + rendering: same as local - video compression: worst possible is 1/framerate - first video bit back to player: ping/2 - whole frame back to player: worst possible 1/framerate - decompression + display: worst possible is 1/framerate So assuming video compression, decompression and transmission are all as slow as they can be thats a total of ping + 3/framerate. So they can rightly claim a perfectly smooth 30 frames a second performance, but the latency could be ping + 100ms worse than a local game. The best case latency approaches the ping. I doubt the compression and decompression is much faster than 60 frames a second, although the transfer time could be quite low in the case of a connection good enough to do HD sending SD. So a more realistic estimate might be ping + 2/60 or ping + 33ms, where a good ping is ~20ms(?) to your ISP or well peered server location. So assuming things work quite well you're looking at >50ms latency. I think that'd be a fairly unsatisfactory experience.
- corysama 18y ago
  OnLive claims to have compression time down to 1 millisecond. I'm not sure how long decompression takes, but it would be more accurate to say "the latency will be ping + 1ms + decompression time worse than a local game". Anyone have stats on latency for decompressing 720p video? BTW: Local game latencies for 30fps games are often (30ms/2 average time between controller input and controller sampling) + (30ms gameplay&physics processing) + (30ms CPU-side graphics) + (30ms GPU-side graphics) + (30-90ms TV signal processing) >= 135ms. I had a chance to play Mirror's Edge through OnLive at GDC. They were claiming a 50ms ping to the server for that setup. I choose that game because I am very familiar with it and it is a very demanding. The extra lag was noticable, but still playable.

10 points by pkaler 18y ago

The elephant in the room of this whole OnLive games thing is, why the heck are they bothering with video games? If they have a technology that's capable of remoting games (which are some of the most demanding consumer applications imaginable), then the whole "Linux vs Mac vs Windows" debate is moot. If it can remote 3D, fast action games, it can remote whole OS's with ease. That's surely a 100x bigger market than games. That, to me, casts huge doubts on their claims... it's like someone claiming they've solved cold fusion and are using it to power really strong, long-lasting flashlights.

teamonkey 18y ago
You can already remote OSes with ease, and have been able to for years. VNC, X11 etc. It's not particularly fashionable, but PC hardware is so cheap you might as well buy a new one. You can easily spawn an instance on EC2 and connect to it, but those machines don't have state-of-the art GPUs. The problem they're trying to solve here is that a new gaming PC can cost a couple of thousand dollars, and you have to keep it on the bleeding edge of expensive hardware to play the latest games. It's typical that even the most avid gamers simply can't keep up with all the consoles and PC hardware.
- swombat 18y ago
  Most OS remoting solutions I've seen are only just usable. I can remote into my Mac Mini into the living room and even that doesn't feel quite right. A 3D shooter, with this kind of latency, would be almost unplayable. And that's just to my living room! If they can get games to play well on the average DSL connection, they can get OS's to feel 100% local, not laggy, like the average VNC session. I'm sure there'd be a big market for that kind of technology, outside of gaming.
liuliu 18y ago
the thing is, OS is in dominant position on the market for not very long time and it was born purely for abstracting the hardware, to make it accessible for upper layer application and have no other functionalities. As all things moved to the cloud, where the OS is running should not be an important question anymore. The real question should be, where the application is running, where my data stored. Get the OS clone running in remote does not serve any benefits for us. You still have to install MS Word in remote and upload your file to the cloud which make no difference to our current status. However, the cloud gaming service solve a more practical problem we will face and we don't have to update our GPU anymore. All the cloud service should solve a practical problem (storage, data management, collaboration) otherwise, there is really no need for it. I hope it will explain why most Cloud OS services now fail.

Angostura 18y ago

I've been thinking about the OnLive business model and the difficulties with the technology/bandwidth and this is what I reckon they are going to do: Sell it to DSL/Cable companies. I sorry I just don't believe that the system will work if it is hosted in some nebulous cloud somewhere, and I suspect that the ISPs will go bonkers about the amount of traffic that it generates - if the ISPs aren't getting a payback. Obvious answer - the ISPs sell it as a service to their customers and colocate the equipment at their exchanges.

bitwize 18y ago
Better yet, sell it to restaurants, bars, etc. for those completely noddy trivia and keno games. That's about the only place I can see it really succeeding. I don't want to play even an RPG with laggy, video-compression-artifacted graphics.

greendestiny 18y ago

As complete an analysis of the difference in latency between OnLive and a local game as I can be bothered making is: - player input to server: ping/2 - game engine + rendering: same as local - video compression: worst possible is 1/framerate - first video bit back to player: ping/2 - whole frame back to player: worst possible 1/framerate - decompression + display: worst possible is 1/framerate So assuming video compression, decompression and transmission are all as slow as they can be thats a total of ping + 3/framerate. So they can rightly claim a perfectly smooth 30 frames a second performance, but the latency could be ping + 100ms worse than a local game. The best case latency approaches the ping. I doubt the compression and decompression is much faster than 60 frames a second, although the transfer time could be quite low in the case of a connection good enough to do HD sending SD. So a more realistic estimate might be ping + 2/60 or ping + 33ms, where a good ping is ~20ms(?) to your ISP or well peered server location. So assuming things work quite well you're looking at >50ms latency. I think that'd be a fairly unsatisfactory experience.

corysama 18y ago
OnLive claims to have compression time down to 1 millisecond. I'm not sure how long decompression takes, but it would be more accurate to say "the latency will be ping + 1ms + decompression time worse than a local game". Anyone have stats on latency for decompressing 720p video? BTW: Local game latencies for 30fps games are often (30ms/2 average time between controller input and controller sampling) + (30ms gameplay&physics processing) + (30ms CPU-side graphics) + (30ms GPU-side graphics) + (30-90ms TV signal processing) >= 135ms. I had a chance to play Mirror's Edge through OnLive at GDC. They were claiming a 50ms ping to the server for that setup. I choose that game because I am very familiar with it and it is a very demanding. The extra lag was noticable, but still playable.