Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

DeLorean: Using Speculation to Enable Low-Latency Interaction for Cloud Gaming

research.microsoft.com

Post reply on HN
Print view View original

21–30 of 49 posts

Re: DeLorean: Using Speculation to Enable Low-Latency Interaction for Cloud Gaming

#21

Post by frik »

easytiger wrote: post #11

Err, how can they call it that? DeLorean is still trademarked; I know it's a different industry but I imagine considering the inference about the relevance of time to the software and the brand association with time that no lawyer would have approved that. http://delorean.com/ Also I'm pretty sure a lot of games such as FPSs already do this with direction of motion etc and have done for many many years. John Carmack…

It seems like a new trend to re-use established names. For example a lot of frameworks on "Show HN" use common names. There are now two programming languages named Swift. A downside is that it makes the web search experience a bit worse.

The vote buttons on HN are too small and next to each other on mobile devices. And it would be better IMHO if the lowest downvote value would be 1 (and only admins could lower the vote-value below).

Re: DeLorean: Using Speculation to Enable Low-Latency Interaction for Cloud Gaming

#22

Post by snarfy »

There are some games this will simply not work for, like fighting games, where even the delay caused by the LCD rendering each frame is noticeable. The ADC in the LCD is either too slow or has too much processing happening. Many hardcore fighting game fans prefer CRTs due to this.

There is no cure for latency. The network is physical. Instead of trying to abstract it away in a library and assume lag doesn't exist, it should be designed for in the application. The last time Microsoft abstracted the network away, we ended up with asp.net WebForms.

Re: DeLorean: Using Speculation to Enable Low-Latency Interaction for Cloud Gaming

#23

Post by darkmighty »

orasis wrote: post #5

Silly thought: What if the universe actually works like this? Instead of actually calculating the full interaction of all particles, it just optimizes by speculating.

Well but speculation has to follow the same laws, so it's not really an optimization.

Re: DeLorean: Using Speculation to Enable Low-Latency Interaction for Cloud Gaming

#24

Post by darkmighty »

antimagic wrote: post #6
Schwolop wrote: post #2

For those who want to try it out, some binaries and videos are found here: https://onedrive.live.com/?cid=611c3830c430d4e3&id=611C3830C... The user perception results are very impressive - round trip times of 250ms were not significantly noticeably different from the control fat client system.

Not just user perception - they did have some objective measures as well, such as the amount of health lost whilst completing a task, and the amount of time taken to complete a task. Even these objective measures came back as showing no significant difference between DeLorean and the fat client. That's impressive.

It is very impressive. If you think about it though, you realize there's not limit to how good this scheme can be (for singleplayer) -- the limitation is essentially the bandwidth. Now the impressive part is that they can adapt to bandwidth conditions and do it all while keeping the premise of required computations < mobile capacity << actual game.

Re: DeLorean: Using Speculation to Enable Low-Latency Interaction for Cloud Gaming

#25

Post by darkmighty »

nardi wrote: post #18
beachstartup wrote: post #9

interesting that they should use an id game, doom3, because quakeworld implemented something similar: from http://en.wikipedia.org/wiki/QuakeWorld : QuakeWorld's most important feature is its rewritten networking code (for client-side player prediction and delta compression). Player prediction allowed QuakeWorld clients to compensate for high latency, thus allowing dialup users to move around in the virtual world wit…

This is completely different. Every networked multiplayer FPS since QuakeWorld uses QuakeWorld's player input client-side prediction algorithm. This speculative prediction is predicting player input before the player gives it , not predicting the server's response to given input .

Well it's essentially the same algorithm on a different side. FPS clients predict the server state and here it's the server predicting client state. The methods to predict/deal with prediction error/etc are directly applicable to both.

Re: DeLorean: Using Speculation to Enable Low-Latency Interaction for Cloud Gaming

#26

Post by 33a »

I wrote about some related ideas on my blog earlier this year.

http://0fps.net/2014/02/26/replication-in-networked-games-sp...

You can do quite a bit of latency hiding without even having to speculatively predict things by just shifting the rendered state of the world in time.

Re: DeLorean: Using Speculation to Enable Low-Latency Interaction for Cloud Gaming

#28

Post by murbard2 »

orasis wrote: post #5

Silly thought: What if the universe actually works like this? Instead of actually calculating the full interaction of all particles, it just optimizes by speculating.

All the known laws of physics are local* so there is no latency issue.

(* EPR only violates locality if you believe in wave function collapse, but there's no reason to)

Re: DeLorean: Using Speculation to Enable Low-Latency Interaction for Cloud Gaming

#29

Post by murbard2 »

readerrrr wrote: post #7

One cheap way of reducing lag where fast mouse movements are done, would be to render a much wider view, send the entire image to the client and then client uses the appropriate segment based on the latest mouse position. At downvoter: Why would you downvote a suggestion that is on topic?

This is not how it works, you don't send an entire image to the client, you send the coordinates of discrete objects and each client renders locally. And yes, the coordinates of objects moving behind you are being sent too.

Re: DeLorean: Using Speculation to Enable Low-Latency Interaction for Cloud Gaming

#30

Post by rincebrain »

33a wrote: post #26

I wrote about some related ideas on my blog earlier this year. http://0fps.net/2014/02/26/replication-in-networked-games-sp... You can do quite a bit of latency hiding without even having to speculatively predict things by just shifting the rendered state of the world in time.

You can't shift the rendered state without prediction if it's an interactive experience without prediction or rendering parallel pathways because computational bandwidth makes exploring the narrowed state space more efficient than waiting on RTT for interactivity...
Post reply on HN