DOCSIS 3.1 Tweak Bringing Ultra Low Latency to Cable Broadband

Cable broadband and TV operators, such as Virgin Media in the UK, that have adopted or are in the process of adopting the new DOCSIS 3.1 standard could also benefit from a new software update. This could enable them to deliver ultra low latency times of around 1ms (millisecond) or less.

Latency is simply a measure of the time (i.e. the delay in milliseconds – 1000ms = 1 second) that it takes for a packet of data to travel from your router to a remote server and then back again (ping). The shorter the delay, the better. All of this is particularly important for fans of fast paced online multiplayer games, where a low ping (lower figures are better) can result in smoother gameplay.

Generally most modern connections already deliver good latency performance and the best tend to be “full fibre” (FTTP) links, as well as future 5G mobile services, where latency times on the network can be down in the low single figures (1-10ms). Admittedly you can get this on FTTC connections too, at least provided the remote server, ISP side network and your local network are all optimally setup (varies from server to server – ISPs can only do such much to improve this).

The new DOCSIS 3.1 standard for cable broadband networks has already been designed to improve upon the older methods via Active Queue Management (AQM). As a result it experiences typical latency performance of around 10ms on the Access Network link, although spikes of 100ms+ are still possible when under heavy load (consumers will rarely see it this bad).

Earlier this year a new specification was added to the D3.1 standard by CableLabs called Low Latency DOCSIS (LLD), which aims to reduce typical latency performance to around 1ms even when under heavy load and the best thing is that the operator can deploy this via a cost-effective software update to existing kit. One step closer to ensuring that hybrid fibre coax networks are able to remain reasonably competitive with full fibre ones.

So how is this possible? Internet traffic consists of various types of data and latency isn’t as important to all of those (e.g. file downloads) as it to others (e.g. online video games). Conversely the types of data where latency tends to be less important are often also the same types that tend to benefit from more bandwidth (broadband speed).

In simple terms, LLD recognises this difference and somewhat pre-emptively optimizes the flow of internet traffic through the network in order to reduce latency. It does this by, among other things, targeting two of the biggest delay areas for latency within the DOCSIS 3.1 network – “Media Acquisition” and “Queuing“.

Queuing

The “Queuing” delay is mainly caused by the current TCP protocol and its variants, which uses a “congestion control” algorithm to adjust to the available bandwidth at the bottleneck link through the network. Typically, this will be the last mile link – the DOCSIS link for cable customers – where the bandwidth available for each application often varies rapidly as the activity of all the devices in the household varies.

With today’s congestion control algorithms, the sender ramps up the sending rate until it’s sending data faster than the bottleneck link can support. Packets then start queuing in a buffer at the entrance to the link, i.e. the CM or CMTS. This queue of packets grows quickly until the device decides to discard some newly arriving packets, which triggers the sender to pause for a bit in order to allow the buffer to drain somewhat before resuming sending.

This process is an inherent feature of the TCP family of Internet transport protocols, and it repeats over and over again until the file transfer completes. In doing so, it causes latency and packet loss for all of the traffic that shares the broadband link.

LLD addresses Queueing Delay by allowing non-queue-building applications to avoid waiting behind the delays caused by the current TCP or its variants. At a high level, the low-latency architecture consists of a dual-queue approach that treats both queues as a single pool of bandwidth.

Media Acquisition

This delay is a result of the shared-medium scheduling currently provided by DOCSIS technology, in which the CMTS arbitrates access to the upstream channel via a request-grant mechanism. LLD cuts Media Acquisition Delay by using a faster request-grant loop and by adding support for a new proactive scheduler that can provide extremely low latency service.

The White Paper (PDF) goes into a lot more detail, although at present we’re not sure if it will be included when Virgin Media unveil their first DOCSIS 3.1 based “gigabit cities” later this year (here). The specification is fairly recent and we imagine that an operator like Virgin might wish to test it a lot before introducing into a live environment.

We have asked their press team about this but as D3.1 hasn’t yet launched (commercially) in the UK then they may be unable to answer. In any case the move to regular DOCSIS 3.1 will still bring plenty of latency improvements and it’s worth noting that, in terms of online video games, the fastest rate at which humans appear to be able to process incoming visual stimuli is about 13ms.