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... :-)

EnasYorl

https://broadbandbreakfast.com/comcast-introduces-new-ultra-low-latency-internet-connectivity/

videomatic3

videomatic3
another new image.

kmax

videomatic3

You are light years beyond the version they have released on our small 400K town.

SBT13

videomatic3

I am still on CGM4981COM_7.6p9s2_PROD_sey.

videomatic3

videomatic3
someone already posted with this version… it made it to my modem now.

EnerJi

Is there a specific plan required to enjoy the benefits of the low latency rollout?

I was browsing new Xfinity Internet plans today and noticed that the 600 Mbps plan lists “Low lag when streaming” and the 1100 and 1300 Mbps plans list “Ultra low lag.”

Is this mostly just marketing, or is there a technical capability difference between these plans? The 400 Mbps plan doesn’t say anything about latency and wondering if I would lose anything if I took the 400 plan (I think 400/100 speeds are likely fine at my household and saturating either direction would be very rare.)

videomatic3

EnerJi while comcast and others would like to make a huge deal about it, with our increasing speeds the purpose of lld the way its being done fades away. “lld” is being rolled out with enhanced speed areas to my knowledge.

as of now its just a special no delay queue for a minority of apps that support it.

MacLeech

videomatic3 as of now its just a special no delay queue for a minority of apps that support it.

It will always be a special no delay queue for apps that support it. It’s not meant for normal bulk traffic, that would defeat it’s operation.

If everyone is special, then no one is special, they’re just everyone.

SBT13

You can also mark packets with a DSCP of 45 to get some benefit from LLD/L4S.

cypherstream

SBT13 Yes we are using group policy to tag voip apps at work.

fmd

A small update from Jason: https://youtu.be/c3BOUmDE1lU?t=3375

cypherstream

fmd Good content.

Now only if they could get the latency down in this area, 29-40ms at idle is just bad for wireline broadband IMO.

My ISP is half that and its also DOCSIS 3.1.

videomatic3

Hokage

Have we heard on when they are actually going to deploy this?

MacLeech

Hokage LLD deployment has already started. The majority of apps (including common tools like ping and tracert) won’t see any difference as they’re not designed to use it. A handful of apps will get subtly better.

SpottedCat

MacLeech Is there any technical reason why DOCSIS can’t match PON fiber’s low latency? When you think about it, they’re both shared mediums that schedule transmit slots for each device. Why does DOCSIS have to be so much worse, and why is the only solution they’ve come up with so far half-baked like this?

MacLeech

SpottedCat DOCSIS has a lot of techniques to fight noise effects, such as interleaving, which add latency. DOCSIS also has a grant request/receive cycle that slows down scheduled transmit slots, they’re not all prescheduled slots for each device. Not sure how much MSOs have spent on reducing interleaving or shortening grant request timing, but that changes unloaded latency from 20ms to single digits. Comcast studied latency (among other things) through it’s RIPE Atlas test units years ago.

Sure they can get unloaded latency into single digits messing with interleaving and request grant cycle, but it doesn’t mean much if the loaded latency jumps it up a couple hundred ms.

The AQM and LLD they’ve been working on stabilizes the wide jitter swings traffic loading causes. From the 200-300ms to 10-20ms or lower. They’re going after the bigger latency issues first, as it affects more of their users, whether they realize it or not. LLD is the DOCSIS part of the wider scope L4S the communication industry in general is working on.

https://account.cablelabs.com/server/alfresco/0affb960-25f4-44f9-b747-74ad8555fd06

fmd

To add: Comcast said that they will test LLD’s proactive grant service. Things to consider:

1) Overhead. If a modem doesn’t use the proactive grant, the bandwidth is wasted. The operator configures how often the CMTS must send these grants. Testing various configs, combined with how well the CMTS can predict the traffic…
2) FDX caveats (dynamic spectrum and tier based bonding)? FDX makes a lot of upstream capacity available, but how well can it be used with PGS?
3) Will it be free, or only by request (less PGS users = less overhead)?

It will be interesting to hear from Jason on this.

cypherstream

fmd so my ping to the cmts is min 7.6 to max 11.5, average 8.9ms.

I never considered that “bad” considering the medium (Docsis) and all the various RF magic going on.

Interested to see what PGS does to that. For comparison… parts of the system that have been upgraded to PON are seeing between 1-2ms to the gateway.

satyr

fmd

Is there a reason why Comcast did not reduce the MAP interval to 1 ms throughout the trials? I get that PGS has bandwidth tradeoffs, which is why they want to test it independently. But, wouldn’t LLD without PGS have decreased idle latency to roughly ~ 4-5 ms and have no jitter? Seems like a no-brainer to test if they’re already deploying LLD/L4S.

fmd

cypherstream Interested to see what PGS does to that.

I agree. 🙂

Adding: PGS doesn’t eliminate request-grant. If the proactive grants are not sufficient, the modem will send requests. This will add to the jitter. Another thing to test.

fmd

There was talk about the predictive DOCSIS scheduler in the old DSLR thread, which was used to test proactive grants in one of the Expo papers. It uses any unused bandwidth for these grants. That is different from PGS, which guarantees a certain amount of grants when there is LL traffic (forced overhead), combined with unused bandwidth.

cypherstream

fmd as pipes get larger and larger (wider OFDM / OFDMA channels) and plants get cleaner (more successful usage of higher order modulations) - there should be more bandwidth to go around to spare on PGS. Especially when you get into FDX territory and smaller node sizes, specially crafted interference groups.

« Previous Page Next Page »