No sławuś znów wyszedłes na durnia gry lokalnie maja nawet 100ms laga
a podobno 20ms jest niedopuszczalne


Editorial: Analysis of Onlive's claims on performance
ShareThis
Posted by: NJ5
03rd Apr 2009, PC, 1,146 views

Onlive has generated a lot of excitement due to its potential impact on the gaming industry. Here we take a close and slightly technical look at performance claims from Onlive's founder and CEO.

If you are an avid reader of gaming news, it's pretty likely you have read a few articles about Onlive. If you haven't, I suggest you take a quick look at articles like this one or this one. In a nutshell, Onlive is an upcoming service which allows for playing games running at a remote server in Onlive's headquarters, with the controller inputs and screen images being streamed through the Internet at 480p-720p resolution and 60 frames per second.

Right from the day the service was announced at GDC 2009, I and many others have had serious doubts that a service like this could really take off in the short/medium-term, mainly due to low speeds and traffic caps of most Internet connections, even when we restrict our attention to developed countries.

Having said that, in this article I am going to skip these general concerns (which are bound to become less important with time) and focus on the concrete claims from Mr. Perlman (the founder of Onlive) in a recent BBC article. The most interesting claim is in this quote:

"The round trip latency from pushing a button on a controller and it going up to the server and back down, and you seeing something change on screen should be less than 80 milliseconds.

"We usually see something between 35 and 40 milliseconds."

When I read this tidbit in the article, it was pretty clear that it's at best a highly optimistic claim, perhaps based on a best-case scenario which few gamers can expect to see. Unfortunately, a closer look raises doubts that these response times are possible at all in this day and age. Keep in mind that response time refers to the amount of time it takes from a button press to the on-screen reaction to that button press (for example, the firing of a gun after a fire button is pressed).

In this very interesting article from Gamasutra, a digital camera is used to measure response time of several PS3 and Xbox 360 games with accuracy down to 1/60th of a second. Here are some of their results in milliseconds:
Software Response time
PS3 System menus 50 ms
Guitar Hero III (Xbox 360) 50 ms
Ninja Gaiden Sigma (PS3) 67 ms
Halo 3 (Xbox 360) 133 - 167 ms
GTA IV (PS3) 167 ms

Comparing this with Mr. Perlman's claims of 35-80 ms response time, this makes Onlive's claimed performance look either exceptional or unbelievable (depending on your perspective). For me, it is very hard to believe that a game running on a local console can have the same or higher response time than achieved with the same game running on a remote computer streamed through an Internet connection from potentially hundreds of kilometers (and several Internet router hops) away.

This is especially true when you notice that the above results are adjusted to discount the effect of a Plasma TV's response time, which by itself is 33 ms according to Gamasutra's research (keeping in mind that TV response time is not the same as the panel's response time, which refers to how long it takes to switch a pixel on after the information has already passed through the TV or monitor's pre-processing circuitry).

If this is not convincing enough or you want a more technical analysis, ask any PC gamer what he/she considers a good "ping time" (network response time) for online play. The standard for good online play is usually a ping time of less than 50 ms, with less than 100 ms being decent enough. However, as the same PC gamer will tell you, ping times vary significantly with time. This is known as jitter, and it must be taken into account in any service running through our unpredictable Internet.

In the specific case of streaming a real-time game through the Internet, if a service like Onlive does not guard against jitter (in other words, if it relies on network response time being constant), the gamer will see jerky framerate, meaning video frames will come at irregular intervals.

The solution to the jitter problem is to buffer one or more frames in local memory before they get transfered to the TV. An undesirable side-effect of this buffering is that total response time will increase, since it introduces an additional waiting period before the response image shows up on the screen. If a single frame is buffered in memory (which will guard against some but not all of the jitter), this will introduce one frame of delay, or 17 milliseconds if running at 60 frames per second. Add this to an assumed ping time of 30 (by all standards a very good network response time), and we're already near the 50 ms level, before accounting for any other sources of lag such as the computation of the game logic, physics, video rendering, compression, transfer and decompression or the response time of a typical LCD/Plasma monitor or TV.

Conclusion

For the time being we should take Mr. Perlman's performance claims with a grain of salt. The claimed performance numbers of < 80 ms response times are at best optimistic given the latency and jitter of today's Internet connections. It is clear that Onlive's engineers have spent a lot of effort in reducing the response time of their system, but there are a lot of factors out of their control.

Nevertheless, there are plenty of reasons to be excited about Onlive if you have a great Internet connection and don't mind the image degradation resulting from the video compression necessary for real-time HD streaming from the Internet. It is expected that Onlive or a similar service will truly take off at some point, but that point seems farther away than Onlive's claims seem to imply.

Jan Strybyszewski pisze:

No sławuś znów wyszedłes na durnia gry lokalnie maja nawet 100ms laga
a podobno 20ms jest niedopuszczalne

1. To polska grupa - pisz po polsku.
2. Do sławusia pisz na priva albo nie crossuj z komputerowych.
3. W dupie mam on-live. Ja to stanie na nogi to *wtedy* mogę się temu przyjrzeć.

Dziękuję za uwagę.

--
Pozdrawiam - Greg McRazy
Zdanie poniżej jest prawdziwe
Zdanie powyżej jest fałszywe
GT: Sylvetka

Dnia 03-04-2009 o 16:21:49 Greg McRazy <a.czy.to@wazne.pl> napisał(a):

Jan Strybyszewski pisze:

No sławuś znów wyszedłes na durnia gry lokalnie maja nawet 100ms laga
a podobno 20ms jest niedopuszczalne

1. To polska grupa - pisz po polsku.
2. Do sławusia pisz na priva albo nie crossuj z komputerowych.
3. W dupie mam on-live. Ja to stanie na nogi to *wtedy* mogę się temu  przyjrzeć.

A ja sie temu przyjrze, jak bedzie dostepne w Polsce, czyli za jakies,
powiedzmy, 20-30 lat :).


--
....::-- _GomezCore Technologies_-- ::...
..DRUM.&.BASS..TECHNO..HARDCORE..NOIZE.
..INDUSTRIAL.AMBIENT.DARKWAVE.IDM.JAZZ.
..POWER.ELECTRONICS.DEEP.HOUSE.ELECTRO.