Comcast Reduced ‘Working Latency’ By 90% with AQM. Is This the Future?
Comcast fully deployed bufferbloat fixes across their entire network over the past year, demonstrating 90% improvements in working latency and jitter — which is described in this article by by Comcast Vice President of Technology Policy & Standards. (The article’s Cumulative Distribution Function chart is to die for…) But: did anybody notice? Did any other ISPs adopt AQM tech? How many of y’all out there are running smart queue management (sch_cake in linux) nowadays?
But wait — it gets even more interesting…
The Comcast official anticipates even less latency with the newest Wi-Fi 6E standard. (And for home users, the article links to a page recommending “a router whose manufacturer understands the principles of bufferbloat, and has updated the firmware to use one of the Smart Queue Management algorithms such as cake, fq_codel, PIE.”)
But then the Comcast VP looks to the future, and where all of this is leading:
Currently under discussion at the IETF in the Transport Area Working Group is a proposal for Low Latency, Low Loss Scalable Throughput. This potential approach to achieve very low latency may result in working latencies of roughly one millisecond (though perhaps 1-5 milliseconds initially). As the IETF sorts out the best technical path forward through experimentation and consensus-building (including debate of alternatives), in a few years we may see the beginning of a shift to sub-5 millisecond working latency. This seems likely to not only improve the quality of experience of existing applications but also create a network foundation on which entirely new classes of applications will be built.
While we can certainly think of usable augmented and virtual reality (AR and VR), these are applications we know about today. But what happens when the time to access resources on the Internet is the same, or close to the time to access local compute or storage resources? What if the core assumption that developers make about networks — that there is an unpredictable and variable delay — goes away? This is a central assumption embedded into the design of more or less all existing applications. So, if that assumption changes, then we can potentially rethink the design of many applications and all sorts of new applications will become possible. That is a big deal and exciting to think about the possibilities!
In a few years, when most people have 1 Gbps, 10 Gbps, or eventually 100 Gbps connections in their home, it is perhaps easy to imagine that connection speed is not the only key factor in your performance. We’re perhaps entering an era where consistently low working latency will become the next big thing that differentiates various Internet access services and application services/platforms. Beyond that, factors likely exceptionally high uptime, proactive/adaptive security, dynamic privacy protection, and other new things will likely also play a role. But keep an eye on working latency — there’s a lot of exciting things happening!
Read more of this story at Slashdot.