What OnLive purports to do is deliver a virtual game system to wherever you want to access it. All the graphics processing is done in the cloud (a fancy word for a random computer inside a data-center) which means no expensive hardware is needed to use it and sessions can easily be transferred across any device – pause a game on your TV tonight and continue playing in your browser at work during lunch.
Cool! So why hasn't anyone done it yet? The reason is that there are some important technical limitations, namely the speed of light. If the graphics processing isn't done in your home, then a tonne of data has to be moved great distances. To do that fast enough to not appear as though the game is lagging is hugely difficult. If you press a button to shoot your gun and two seconds later a bullet comes out you aren't going to last very long in most games. Our previous experience developing online applications has given us an understanding of dealing with internet latency and its limitations for response time.
With that in mind we put on our berets and sat down with a pen and paper to see if we could bust this myth.
The OnLive micro-console and controller for TVs.
Response Time
Our first stop was to analyse what response time would be necessary to make a seamless gameplay experience. A general rule of thumb on this for computer interfaces has been around since 1968. Any response up to 0.1 seconds (100ms) is considered instantaneous by most people. Anything above that and the user will notice the delay – it will feel like the computer is doing the action, not you.
This follows on fairly well with what you experience with most gaming systems. Gamasutra has a fantastic article on the response time for recent games. Here are some numbers they found for response time for various games:
- PS3 system menus - 50ms
- Guitar Hero III - 50ms
- Ridge Racer 7 - 67ms
- Virtua Tennis 3 - 67ms
- Ninja Gaiden Sigma - 67ms
- PixelJunk Racers - 67ms
- Genji: Days of the Blade - 100ms
- Tony Hawk's Proving Ground - 133ms
- BlackSite: Area 51 - 133ms
- Halo 3 - 133-167ms
- Skate - 167ms
- Grand Theft Auto IV - 167ms
- Heavenly Sword - 117-300ms
"This is rather a long response time, and correlates with people reporting the game being sluggish and unresponsive. The delay in firing the gun after pulling the trigger is quite noticeable," they said.
The 300ms response in Heavenly Sword is noticeable too, as shown in this video Gamasutra produced.
In conclusion less than 100ms would be the sweet spot, but up to 160ms is acceptable.
Network Latency & Bandwidth
The two important factors that need to be considered when moving the console into the cloud is the latency (time it takes for a signal to travel from your TV/PC to the remote computer & back again - also known as ping) as well as bandwidth for the data to be transferred.
Typical network latencies are well documented. They depend on several factors including the physical length of the cables as well as the number and speed modems and routers in-between. On cable and DSL connections they are generally less than 100ms to a computer in the same country. In their GDC press conference OnLive CEO (and David Hewlett look-alike), Steve Perlman said the service will work with a data-center up to 1000 miles (1600km) away. Typically, you would expect up to 30ms latencies for this distance.
On the bandwidth front, the issue is video compression. You can't send a raw HD signal over the net without compression otherwise the bandwidth usage would be too high for most households. Video compression helps reduce this data, however traditional solutions have been rather slow – 500ms or more latency to compress HD video. This appears to be the area where OnLive has made a considerable technical achievement. They have custom built a network card that can compress HD video in less than 1ms. This almost wipes out the impact that compression has on the latency and has significant implications for other video processing applications to boot.
Finally, there will be some additional latency to consider at the client end involved with decoding the compressed video, but this could be expected to be less than 5ms on most machines.
The OnLive dashboard.
The Result
It's possible. Just.
Moving the console to the cloud under OnLive's architecture could be expected to add under 40ms to the response time. This is 25% of our maximum acceptable response time (160ms) and 40% of our "good" response time (100ms). Clearly some games could get away with this additional response time without too much trouble, but some of the high-end console games would have problems.
OnLive appears to be using entirely PC hardware in their architecture. High powered PCs (and optimised game code) could reduce this latency even further. From the analysis above plugging in existing consoles to the OnLive network probably would not work for many games, even if the manufacturers agree.
Questions Unanswered
The main question we have from looking at their current architecture, is how do game states get preserved? From the demos they've shown, games appear to be resumed instantly, but even in cloud computer, some time is required to boot a game into a playable state on the server. It would be impossible to keep machines in a state ready to play all previously paused games when this is released in a public beta. There must be some delay when dropping back into an existing game.
Rolling out OnLive to other countries is also going to raise some interesting challenges. Given that Australia is 4000km wide you would need to roll out at least 3 data centers for full coverage. Australia's high data costs and bandwidth caps would also come into play. If you played 60 hours in a month on a 5Mbs connection, you would use about 100GB – more than most bandwidth caps in Australia. It would require significant infrastructure changes before OnLive could be rolled out here.
Brag Clips of Mirror's Edge anyone?
The Cool Stuff
If it succeeds, OnLive could have some significant impacts to not only the way we play games, but also the web as we know it. Here are a few salient ideas:
- It could revolutionise the way we play online games. Being able to see anyone else's game in progress and jumping right in transparently is a huge step, not to mention the ability to play a game on any PC, Mac or TV (and no periodic hardware upgrades!) is a holy grail for gamers. It could mean the end of console wars as we know it – third party publishers would love this system, not only for its ability to effectively drive out piracy, but also because it takes away the cost of developing for multiple platforms.
- It could revolutionise virtual worlds which are currently hugely limited in their ability to download high & render high res textures on-demand. Imagine a version of Second Life where you could create and visit worlds rendered with the fidelity of Gears of War.
- Finally it could revolutionise the web. Websites currently rely on your computer to display their graphics and respond to your mouse clicks. If this could all be performed on a server, web content could be delivered in a HD game-like environment regardless of the device you are using.
My main worry is how much this will cost Australians.
Loading...
