Hacker News Reader

Comment by cm2187 | Hacker News Reader

Do you really get significant cost savings? You can't really use hardware from other time zones / continents, because the latency would be very visible. So the hardware has to be reasonably close to the users.

Then I would imagine the problem is that everyone is playing video games at about the same time, i.e. in the evening or at certain times of the week end. So you need to provision for peak usage.

So to me the only benefit to renting is if there are lots of occasional gamers, that do not play every day / every week, including peaks where they would all suddenly play at the same time on certain days (christmas, long week ends, release of new game, etc).

And then of course because the hardware progresses so quickly you need to amortise the hardware pretty fast.

I haven't seen the actual economics but I am surprised it would be much cheaper for consumers.

View on HN

You might be able to make it work if you can get a gaming tier on a cloud platform. EC2 already has GPU focused instances.

I can see things working out, but still expensive, if the hardware also being using for scientific computing and CGI rendering.

Although I'm not sure if gaming hardware can run 24/7.

I first heard about this year ago, and was also amazed that the numbers could work and RDP lag would make it viable. I can't find the reddit post at the moment, but we had a few exchanges where it was explained that latency is not an issue, and availability is not an issue.

This is still a fairly niche market, and while there are certainly evening peak times, younger kids have a lot more free time than you might think.

How well the company hosting this is doing, I couldn't tell you, but the product itself seemed to work and work well, so much so that I have had it on my to-do list to try and experience it for myself.

The speed of light is really fast, Google has fiber connecting all of its data centers, I think viable data centers for game streaming overlap more than you think, it's limited by how much bandwidth Google can spare between data centers.

The reason most services need to be located close to clients is because you want to avoid data transit over the open internet. You want as little open internet as possible between you and the client. Traffic that's internal to Google's networks doesn't have that problem. You can use compute / gpu in US-west and transit that data to you via US-east and the additional latency would be measured in nanoseconds.

With a straight shot, latency from one coast to the other is at least 50 ms [0]. A more typical route is on the order of 80 ms.

That's definitely noticeable for latency-sensitive actions. I recently switched a server from Oregon to New Mexico, and I notice the latency increase with mosh.

Moreover, there's not a lot of timezone difference between the east and west coasts of the US. Going someplace like Europe is more like 180 ms.

I've played games with a ping like that, but a lot of the ping was my wifi. Doubling the latency would not make the game a better experience.

This could work well for certain types of game, those that are a little less latency-sensitive. But in general the latency is still a big issue.

[0]: speed of light in fiber is ~ 100e6 m/s, around 3000 miles between coasts

Exactly. And 60fps (which is not that high for many games) gives you 16.67ms per frame. at a 50ms delay you're already 3 frames behind.

Not all games are that latency depending. If I am playing civilization I probably won't care that much where the server is.

Neither, will you need to rent an expensive gaming rig by the hour, though.

Wow, I never realized my intuition was off by so much. Speed of light in vacuum is still about 16ms between coasts, which is orders of magnitude worse than I thought. I thought most latency came from routers and switches but a significant portion of delay is indeed speed of light.

Reply on news.ycombinator.com