After reading the comments on the recent article about OnLive's lag issues, I felt compelled to educate the the dtoid community. It seems to me like consensus among the posters is that OnLive is destined to fail, and given the fact that the service isn't out of beta yet, it is almost certain that none of them have tried it for themselves. So before anyone else decides to join these travelling fortune tellers I am going to do my very best to explain why this just might be the next evolution of the video game industry.
The article references a blog post where OnLive's founder Steve Perlman is defending the service against the informal preview of a blogger from PC Perspective using the beta credentials of a friend. The PC Perspective post itself was very critical of the lag that the blogger experienced, stating that the games were almost unplayable. For those that don't care about the technical details, suffice it to say that in its current form OnLive was designed to be played from the same server that the beta account was originally registered to. This restriction is necessary for OnLive to gather statistics about people's use and the performance of their system. So if you stop now and take nothing else from this blog post, please keep in mind that the experience of PC Perspective is not an accurate assessment of the way the system will work once it is out of beta and available to the general public.
For those of you that have decide to stick around I will now attempt to describe how OnLive hopes to deliver on their promise of instant access to all the hottest games. As Perlman noted in his blog the biggest hurdle that OnLive has had to clear is that of the limitation of the speed of light. Through a little bit of research into human psychology they have learned that longest amount that can pass from when someone performs an action to them seeing that action and accepting that they were responsible for that action is 80 milliseconds. That means that they have 80 milliseconds from the moment a user pushes a button to get that input to their server, compute the next frame, compress that frame, and get that frame to the user. So for a user at that 1000 mile limit the round trip time that it would take light to go back and forth between the user and the server would be about 10 milliseconds, which is highly optimistic. First, although much of the backbone network in the US is fiber optic most end user connections are not, so the packets won't travelling at the speed of light to the end user. The other factor is that their is a lot of switching that happens as a packet makes its way to the user. If I remember correctly from the presentation that Perlman gave at Columbia University they are assuming a 25-30 millisecond window for the round trip packet time.
The problem gets even more complicated when you realize that ISP's try to route as much of their traffic through their own network to avoid switching fees (fees charged by other ISP's for switching the packet onto another ISP's network.). This means that someone in Las Vegas attempting to connect to the data center in California might be connected via Seattle because that is cheaper for their ISP, but it does nothing to help keep OnLive's round trip packet time down. So OnLive had the brilliant idea get their internet connection from 10 of the top ISP's in the country and then optimize their servers to find which ISP gives them the most direct connection to the end user. So given that their biggest limiting factor being the speed of light, they have spent a lot of time optimizing the rest of their algorithm to hit into that 80 millisecond timeframe.
They have used a number of novel approaches to achieve this. First unlike typical internet connection using TCP, which request lost packets be resent from a server, OnLive uses UDP. This basically means that if a packet gets lost they don't really care. So a frame here of there is going to have a lot of missing data. Their algorithm is designed as a feedback loop, so maybe if certain parts of a frame didn't make it for any particular frame they will attempt to make sure that in the next frame those errors aren't present. So unlike other online video services, they are just trying to make sure that in motion the picture looks good. And all of this has been designed into a piece of hardware that allows their servers to do this compression very quickly.
At this point I feel that I am just rehashing points that Perlman recently made at a college demonstration of OnLive. If you really want to understand how this works, I would suggest watching the whole forty minute video. And to all of those that still think that this is going to fail, that is why you are never going to be rich. It takes a bold idea, with the right amount of innovation and desire to make it happen. Steve Perlman has a bold idea with the potential to reshape the video game industry as we know, and he has spent many years developing an innovative solution that I predict will wow even the most staunch critics.