Low Latency DOCSIS

fmd · 2025-01-18T22:48:16+00:00

Comcast has now begun the mass rollout of LLD. With a little hope, Jason Livingood will join us here to keep us updated. The thread is ready... :-)

fmd

Great news. It's also great that Comcast has chosen not to make it a paid service. That was once a worry. This is the way.

telcodad

jlivingood

Jason - You forgot to also mention this item of yours: 🙂

What is Latency?
....We spoke with Jason Livingood, VP, Technology Policy, Product and Standards, who has spearheaded Comcast’s industry-leading work on the technology responsible for delivering this ultra-low-lag experience about what it means for customers and the future of their connected experiences....

https://corporate.comcast.com/stories/latency-speed-explainer

cypherstream

fmd Yes DSCP46 (Expedited Forwarding) is what I use on our voice vlans in our network at work. We have plenty of bandwidth, so this is just being kinda overcautious.

CommunityMember

cypherstream

Yes DSCP46 (Expedited Forwarding) is what I use on our voice vlans in our network at work. We have plenty of bandwidth, so this is just being kinda overcautious.

Due to various (real world) network artifacts in most enterprise (and even consumer) networks, even with "plenty of bandwidth" you can encounter packet loss(s) if you don't prioritize your VoIP traffic (or whatever traffic you don't want to lose), and have the devices respond accordingly.

Man-of-Leisure

jlivingood

Are any retail 4.0 modems expected?

The recent LightReading article said:

A number of retail DOCSIS gateway suppliers are in the lab now iterating their code to work with FDX, he said.

But I'm not aware of any announced.

jlivingood

Man-of-Leisure

True on 4.0 (in the backlog) but at least the next-gen FDX ones (https://assets.xfinity.com/assets/dotcom/projects/cix-4997_compatible-devices/2024.09.18%20Full%20List%20of%20Compatible%20Devices.pdf).

Man-of-Leisure

jlivingood

I'm not seeing any FDX modems in that list there. "Next-gen" in Comcast terms is mid-split, not FDX.

If "retail suppliers are in the lab working on supporting FDX" does that mean retail 4.0 modems are coming?

So far, the only 4.0 modems announced are only direct sale to ISPs, not retail for customers.

fmd

Btw, videomatic3 must be happy about how the title of the press release is worded. 🙂

"Ultra-low latency" is reserved for the release of PGS. 🙂

videomatic3

fmd you noitced how i called it a no delay queue i assume? because i seriously disagree with this being called ultra low latency

fmd

videomatic3 I remember you nagging about it on the old site. That is was only about queueing delay. Comcast did phrase it "ultra low latency" previously 🙂

CommunityMember

videomatic3

you noitced how i called it a no delay queue i assume? because i seriously disagree with this being called ultra low latency

Have you ever worked closely with marketing at a large organization (I have)? It is about the spin, and not the details. Marketing by companies does not want to state actual falsehoods (they get called out on those), but using words that make their position look better is fair game (and they will take that to the limit, as most people want to hear what they want to hear). There is the old joke (about IBM at the time): "Q: How can you tell if marketing is lying to you? A: Their lips are moving."). If you want actual facts, read the specifications, or talk to the actual engineers (if they are allowed to talk with you).

videomatic3

fmd yep. because one of these days im really hoping for fiber like latency. and the more im learning about this... this is not our answer. so of course its disappointing.
and the chances of my direct bury att land of getting fiber is next to 0%

now of course on cablelabs they advertise it like "getting docsis to <1ms latency" and this image i keep posting over and over gave me a spark of hope. only to be shreded.

fmd

videomatic3 Low queueing delay is still important. If this is not addressed, you will still have large latency spikes and packet loss when other parts of the delay chain get addressed.

When you have two queues: You can let one of them get hammered and filled up, resulting in increased queueing delay and packet loss. The 2nd queue awaits your packets that are sensitive to latency and loss.

adlr

videomatic3

Meanwhile, folks on fiber are asking how to get L4S, heh. https://forums.sonic.net/viewtopic.php?t=18190

lesmikesell

fmd

fmd When you have two queues: You can let one of them get hammered and filled up, resulting in increased queueing delay and packet loss. The 2nd queue awaits your packets that are sensitive to latency and loss.

But the effectiveness of that depends on applications correctly marking packets that are important. Why is any application going to think its packets are unimportant? And how are you going to measure it? Is 'piing' an important packet that should have priority or just one that anyone can send any number of for any reason?

fmd

lesmikesell It's not about being "important". Fx if there is congestion, you won't get access to an unfair share of the shared capacity by marking your packets. If you (your application) have packets that are sensitive to latency and loss, just mark them. Game packets, DNS packets, etc.

L4S allows the application to use high bandwidth without creating high queueing delay. It allows you to pull your entire bandwidth through the low latency queue.

EnerJi

I know the focus is on working latency, but will I notice significant changes in a simple traceroute or ping to (or just past) my VCMTS? I'm not currently in one of the first low latency markets, and pinging the second hop in a traceroute just now gives me the below results. When/if it gets rolled out in my area, I'm curious if I will see a significant improvement on my ~15ms average latency?

58 packets transmitted, 58 packets received, 0.0% packet loss
round-trip min/avg/max/stddev = 8.731/15.567/23.799/4.081 ms

jlivingood

EnerJi You need to generate load when running the ping or trace. It may be better to run https://speed.cloudflare.com/ and then look at the latency under load - also try https://www.waveform.com/tools/bufferbloat

fmd

jlivingood Are there other DOCSIS tweaks? Fx is there a change in the amount of contention request opportunities made available to the modems, in times where there is no congestion vs congestion?

cypherstream

Here’s the internet draft on this topic submitted by our very own VIP here Jason Livingood.

Good read on the topic for those tech mined and curious:
https://www.ietf.org/archive/id/draft-livingood-low-latency-deployment-07.html

« Previous Page Next Page »