News Comcast's L4S low-latency tech promises up to a 78% latency reduction, testing begins in select cities

This is a live issue for me, since latency affects interactive stuff I do when working from home.

I wish they'd say more about device compatibility. I only upgraded my modem 2 years ago, but I'd be willing to consider another upgrade if the gains were substantial.
 
This is a live issue for me, since latency affects interactive stuff I do when working from home.

I wish they'd say more about device compatibility. I only upgraded my modem 2 years ago, but I'd be willing to consider another upgrade if the gains were substantial.
My guess is this will be implemented on their switch hardware. No change in the first hop from your modem. There is no alternate path to change to.
 
  • Like
Reactions: bit_user
QoS packet marking has been available since the ip protocols were defined 30 years ago. This is nothing new and they still have not solved the major problems. First everyone will just mark all their data to be the highest and fastest class. Next all ISP must agree on what the marking mean and which traffic gets priority. And third this violates the concept of net neutrality people are all up in arms about. So if some game company pays extra their traffic is prioritized over other game companies.

This project is all marketing speak. Comcast can do nothing by themselves unless the game companies and server location have direct connection to the comcast network with no other ISP in between. It still doesn't solve the issue of people like me who when the document the packet markings will mark all his traffic high priority.

Almost all ISP have abilities to detect overloaded connections and move some traffic to other less loaded connections. This though tends to increase the latency for the people whose traffic takes a different path.

I agree the whole bottom of the article is completely idiotic written by someone who has no idea how network latency really affects games. The refresh rates of monitors and video cards is massively faster than any network connection.
 
Was this article written by AI? That last paragraph is unhinged lol.

The editors made an undesired change to it- I don't know why- that first sentence originally ended with "though", which greatly improves the flow.

But also, no. I god damn hate generative AI and everyone with anything to do with it. You can check my history of published articles if you're unsure. My advice regarding uncapped framerate is simple matter-of-fact but also rooted in my own competitive gaming experience. Seemed worth noting in the article about latency reduction, but, you know, whatever. Just call any stylistic choice you don't understand AI, huh?

Weak. Will not survive the Winter.
 
I agree the whole bottom of the article is completely idiotic written by someone who has no idea how network latency really affects games. The refresh rates of monitors and video cards is massively faster than any network connection.

also- input lag is input lag, whether enforced by networking or local hardware, and I promise I have more competitive gaming experience than you do.
 
QoS packet marking has been available since the ip protocols were defined 30 years ago. This is nothing new and they still have not solved the major problems. First everyone will just mark all their data to be the highest and fastest class. Next all ISP must agree on what the marking mean and which traffic gets priority.
That's not how I read it. I think they're going to throttle connections at/near the edge, in order to eliminate congestion in the core. This not only reduces packet loss, but also keeps queues mostly empty and thus we see a reduction in latency. It doesn't need to rely on packet tagging or traffic classification. The only real assumption built into it would be that the most latency-sensitive connections won't also be the highest bandwidth ones, because those might indeed see a slight worsening of latency.

As you point out, they can only control their own network. However, a lot of big content providers are using content delivery networks (CDNs) which have co-located nodes inside the networks of big ISPs like Comcast. If most traffic is staying inside Comcast's network and they can virtually eliminate congestion there, then you'll likely experience an overall improvement when connecting out over backbones they don't own. Sure, if you're going transoceanic, all bets are off.

And third this violates the concept of net neutrality people are all up in arms about. So if some game company pays extra their traffic is prioritized over other game companies.
That battle was fought and lost (in the US), from what I recall.

I agree the whole bottom of the article is completely idiotic written by someone who has no idea how network latency really affects games. The refresh rates of monitors and video cards is massively faster than any network connection.
Let's say you game at 165 Hz (no framegen, because it doesn't help with latency). That's still 6 ms. It potentially stacks with the rest of the latencies in the system. It's the additive nature of latency that gets some people get so nuts about trying to maximize framerates and any other bits of latency they have control over.

I think it's a smart move by Comcast, who can't really compete with fiber on raw bitrates. However, if they can achieve meaningful latency reductions, that's another way they can differentiate themselves and try to compete, maybe even offering a higher price tier for it.
 
Last edited:
This doesn't sound like something new, it reads like all they've done is to turn on ECN.

https://en.wikipedia.org/wiki/Explicit_Congestion_Notification
That seems to be part of it. Somewhere else, I found a link to a RFC that references L4S right in the title. I just read the abstract, which does indeed mention ECN, but it seems there's a little more to it.


As for my question about modem compatibility, I get the impression that it will initially be limited to Comcast-supplied modems. However, it sounds like 3rd party modems should eventually provide support for it. Since the model I bought has already been superseded, I doubt they'll backport support to it and presume I'll have to buy a new one, if/when it's available in my area.
 
This is indeed about marking packets. That is needed, because there are now two network queues. By marking them, you can avoid getting stuck in a queue that is subject to bufferbloat (to a degree, thanks to AQM) and packet loss.

L4S allows you to use the entire bandwidth and still keep very low queueing delay. L4S must be supported by the endpoints and by the network (the bottlenecks - the modem and CMTS in this context).
 
Yeah, lol.

and @nightbird321

Sorry, I still need to come back to this— what exactly makes yall so certain a robot would write such a specific sidenote about reducing/stabilizing local input lag in an article about Comcast supposedly reducing web latency for gaming and real-time video? Wouldn't making no such mention be the MORE robotic thing to do???

Seriously, this is getting OLD. I've never touched Gen AI for my work in my life and I NEVER will, but I keep getting these slanderous accusations despite well over a decade of experience. What is the deal??? Are y'all actually that narrow-minded and stupid?