Hacker News Reader

Comment by smoldesu | Hacker News Reader

Wifi works fine for it, honestly. As noted elsewhere, the biggest problem is latency-based. Using WiFi will probably add ~30-50ms of latency to your setup, which is bad, but also consider that both the PS5 and Xbox Series X have over 100ms of input lag alone. Once you do the math, Stadia over Wifi isn't any less responsive than playing on console. As long as your Wifi connection is half-decent and you aren't playing some Chinese MMO, your experience should be fine.

View on HN

> Using WiFi will probably add ~30-50ms of latency to your setup, which is bad, but also consider that both the PS5 and Xbox Series X have over 100ms of input lag alone.

This is absolute nonsense. Average end-to-end input-to-rendered-frame latency is half that at 60fps and more like a third of that in 120hz mode in Gears 5 on the Series X https://www.neogaf.com/threads/xbox-series-x-input-latency-r...

Lots of similar results abound. Here’s Multiverse on the PS5 averaging 31ms end-to-end latency at 60fps: https://twitter.com/noodalls/status/1552942081014132738

“100ms of input lag alone” (before the engine even has the input? c’mon mate) is a preposterous idea.

If your WiFi connection adds 30-50ms of average latency you seriously need to stop living in a microwave oven or upgrade to something after 802.11b.

With an Intel AX201 WiFi chip connecting to a UniFi6 Lite my average latency is 1ms to the router.

Even a cheap and somewhat compromised WiFi chipset/antenna can do way better than 30-50ms latency. Here's an original Pi Zero W connecting through the same AP hitting the local router:

64 packets transmitted, 64 received, 0% packet loss, time 156ms rtt min/avg/max/mdev = 2.293/6.699/63.116/7.364 ms

This is while I've got streaming music playing on my phone on WiFi connected to Bluetooth headphones, streaming video on the WiFi TV, and other wireless network traffic happening.

>64 packets transmitted, 64 received, 0% packet loss, time 156ms rtt min/avg/max/mdev = 2.293/6.699/63.116/7.364 ms

A 63ms delay on some packets can totally ruin the user experience even if the average is relatively low... and this is the best a near-optimal (in the real world) setup can do.

If you left your test running longer than 64 packets, I'm sure you would see at least a 100ms max. GP wasn't far-off with their assertion that running on WiFi can easily add 30-50ms of latency. Their wording was a bit loose though.

That 63ms packet was one packet. It's way out of the standard deviation. You don't point to the max and state that's the typical experience of the network.

If we only looked at max values then Amazon isn't 2 day shipping it's like 1 month because some delivery sometime took that long, even though like 95% of deliveries are within two days (just an example, don't know what their actual rates are)

And on top of that, that's with an extremely cheap chip with a compromised antenna inside a network closet with a TCXO literally right on top of the antenna running on 2.4FHz. My 5GHz example with a nice AX chip and decent laptop lid edge antenna never experienced any pings higher than 1ms.

> You don't point to the max and state that's the typical experience of the network.

The test uses 64 packages which is a tiny sample size for streaming use. If we blindly scale that up you would hit the max several dozen times per minute.

> If we only looked at max values then Amazon isn't 2 day shipping it's like 1

A better example: Imagine your tongue just stops moving once every minute and then tries to catch up with what you where doing, imagine this will happen for the next two months. It is just a minor annoyance, right?

A relevant ping test should also be sending packets far more often than once per second (more like once per frame at a minimum), and should be sending larger packets than the default for ping.

Here you go, 1 thousand byte payload ping sent every 0.02 seconds on my Pi Zero W.

  ping -i 0.02 -s 1400 router.home.lan
  . . . lots of pings . . .
  6203 packets transmitted, 6203 received, 0% packet loss, time 128938ms
  rtt min/avg/max/mdev = 2.023/3.397/43.075/1.785 ms, pipe 2

The previous ping had the Zero W way overloaded, I realized that after. Single core running with a load average of 2.8, still managed to be decently interactive and have the previous pings. This ping was happening while having 1 minute load average of 1.2. So only kind of overloaded ;)

This kind of latency is achievable on a $10 board with only 2.4GHz Wireless N. Suggesting WiFi adds 50ms latency on average is extreme hyperbole.

Again, max ping of 43ms. The worst kind of latency is the unpredictable kind that can't be compensated for. The reaction speed of a casual gamer will be somewhere near 180-250ms. If we take the upper bound, occasionally the casual gamer's input will take 17.2% longer to register.

So if even 99.9999% of pings were <5ms, but one was 40ms, you'd define that link as a 40ms experience?

For gaming, your experience is absolutely degraded by spikes in ping. A consistent 50ms latency is far better than a 5ms with the occasional >50ms latency.

So yes, one packet a day at 50ms would be considered unplayable to you, even if 99.9999999999999% of the packets were <5ms, because it is the "occasional" 50ms latency.

That's what you're arguing when looking at just the max value.

Makes sense.

We aren't talking once a day though. We are talking like once a minute or two, at least by the data you have provided.

An important aspect that I think is overlooked: even if the latency is so subtle that it's overlooked, subconsciously you may have a feeling that something is off, your perception of the product will be worse, you will perhaps become frustrated without even knowing why.

A person I know was diagnosed with tinnitus. And it (the diagnosis) made him happy. For over a year he was frustrated, nervous, but he didn't know why. Turns out he was hearing noise all that time, but just didn't realize it. Sounds (pun not intended) ridiculous, but that's how the mind works.

You're dancing around the issue at play though. An application that hitches or glitches for a second every few minutes will quickly become irritating. It doesn't matter if it works 99% of the time. It has to work 100% of the time or its strictly inferior to console or PC.

> 64 packets transmitted, 64 received, 0% packet loss, time 156ms rtt min/avg/max/mdev = 2.293/6.699/63.116/7.364 ms

So if a video frame consists of 64 packets, the whole frame will be delayed by 63.1ms.

Yes there are ways around this (FEC), but getting 1ms 99%ile latency over wifi IRL is nontrivial.

Btw fat-client PCs have much smaller latency spikes, but even those spikes are noticable and make games feel jittery. That's why we compare GPUs based on "1% lows" instead of average FPS. Also see frame pacing.

> Yes there are ways around this (FEC), but getting 1ms 99%ile latency over wifi IRL is nontrivial.

Streaming video is extremely nontrivial to start with. Use FEC when necessary.

Also I just threw a thousand 1400B packets over my wifi and 99% was something like 5ms, max 10ms. 5ms is close enough for me, and I could probably hit 2ms with light effort.

Naw, streaming video is trivial if you have a 1 second latency budget so you can use LL-DASH and CMAF. We hit that target with <100 hours of dev work on a co-watching/cloud-streaming project with only 2 PoPs in the US. WebRTC FEC is a relatively modern and wasn't on the table in 2020.

Btw you almost certainly tested on an unloaded connection so you're not seeing any HOL blocking like a real video stream would. Try playing a 20mbps YouTube video in the background while you ping. Oh and turn on a microwave oven when you're 3 walls (a room and a hallway) away from the AP for extra hilarity on the order of 80ms

I meant game control video. You don't get a second of buffer for the player.

And you don't need to use webrtc, or you can add your own FEC on top for your client software.

And yes a 4k YouTube video makes it worse but "don't do that" is a valid response and realtime video won't be so bursty. I can get a very nice ping while more stable bandwidth hogs are running. Microwave is also "don't do that".

A ping result from my WiFi 6 laptop while watching YouTube on the laptop playing through Bluetooth headphones, messing around in an RDP session, a few SSH consoles actively refreshing, chatting on Teams, music streaming over WiFi on the audio receiver, while cooking lunch in my microwave:

  Ping statistics for router.home.lan:
      Packets: Sent = 793, Received = 793, Lost = 0 (0% loss),
  Approximate round trip times in milli-seconds:
      Minimum = 1ms, Maximum = 5ms, Average = 1ms

You probably live in the literal woods.

Over here in the urbanized world it's a whole different world of pain.

I live in one of the top ten largest cities in the US. 5GHz doesn't go that far through walls.

I live walking distance to a train to take me to an airport non stop around the world.

Try again.

There's no 100 ms. Current generation made a huge leap forward. Controller and whole input stack add 2-8 ms. TVs with 120Hz, VRR and low latency mode add 6-10ms in 4k HDR (no idea about DolbyVision HDR).

People's wifi ISN'T half decent though. They use 10 year old hardware stored in a closet to service their two story house then connect all the devices furthest from the router to the 5 ghz network and get 10Mbps intermittent service partying like its 2002 instead of 2022.

By far the bigger issue with wifi is periodic bursts of noise and wifi scanning introducing stutter. Wifi latency is low enough to be a complete non-issue, the only place you'd care is with VR (and even then it's probably fine, but you're much much more sensitive to that stutter, so it's worth doing something special).

30-50ms is ISDN territory. You either have absolutely lousy Wi-Fi (like a soft AP running on ancient hardware) or you’re confusing latency with jitter driven by signal interference and retries (which is also lousy Wi-Fi).

> Using WiFi will probably add ~30-50ms of latency to your setup

Uhm... What?

I'm assuming the worst here. I've seen people rocking decade-old Netgear routers that struggle to work with their media center, much less an entire household of devices.

That's a pretty catastrophically high estimate though, you're right.

Hot damn son your WiFi must be poorly configured. I got sites w WiFi aps about the 4th hop after a ptp WiFi link. Adds about 2ms latency the furthest you can get from the main router. Bout half a ms per hop.

30-50ms per hop is absurd.

Reply on news.ycombinator.com