Middle-Mile Resiliency and Delivering Live Streams at Scale
As the last mile of streaming becomes increasingly predictable, according to CacheFly CTO and Founder Matt Levine, the focus of streaming professionals working to enhance reliability in live event delivery shifts to the middle mile. Levine and YouTube Head of OTT Live Engineering Sean McCarthy explore what workflow and architecture elements constitute the middle mile and what it takes to navigate middle-mile variability most effectively origin-to-edge and designing workflows that scale in this conversation with SVTA Subject Matter Expert Bhavesh Upadhyaya at Streaming Media Connect 2026.
Middle-Mile Variance
Upadhyaya says that while certain aspects of the end-to-end workflow are solved, there are problems that crop up with the ingest process. He asks Levine where he sees the problems originate and where his CacheFly customers are looking for solutions.
Levine replies, “I don’t want to overstate the ease and simplicity of delivery at scale, but I find myself saying more and more that the concept of which CDN can deliver an object better [is] … maybe not the most recurring problem these days. And so what we find ourselves working on most, especially when it comes to live events, is the reliability and the resiliency … around actually that middle mile/first mile and making sure that that stream can get into RAM at all the edges.” He notes that while the end user experience is relatively predictable, the middle mile has a lot of variance. “And so the good days and bad days for us that we see for live events tend to show up a lot more on that workflow from the second it hits the glass on the camera to when it makes it to a CDN edge.”
This is when CacheFly and its customers focus on resiliency, “whether that is more aggressive connectivity in terms of active backup, be that getting cross connects in for events and then backing it up over transit.” There are CacheFly customers “we’re testing, doing parallel fetching from, and first success wins when it comes to fetching stuff from their origin,” Levine shares. The company is “starting to treat the concept of a cache miss as a first-class citizen, and what [that would] look like in the same way that an object from RAM in cache is kind of treated like a first-class citizen today.”
Struggles of Scale
Upadhyaya invites McCarthy to give his perspective, saying he’s “been working on something like this as an industry standard as well too, in terms of looking at ingest and onboarding of streams, etc.”
McCarthy confirms, “It’s definitely relevant, I’d say, to … optimize the furthest-edge cache footprint and delivering bits from the RAM, like [Levine] was saying, down the wire as quickly as possible.” He sees it as “a connectivity to your live origin optimization.” This is something YouTube is aware of, McCarthy continues, but YouTube owns and operates its own delivery network, so “it’s architected slightly different than if you were to connect your origin to several CDN vendors. We don’t necessarily have that problem, but we absolutely have the adverse problem, which is getting content into our live origin. So it’s an acquisition challenge.”
McCarthy explains that historically, YouTube has approached this by meeting its top-tier premium broadcast customers “where they are, to build fiber connectivity from their data center to ours, or to co-lo[cate] in their data center in order to cross-connect and get those bits on our network as soon as possible and have dedicated fiber—something reliable, resilient, what have you, and fast, but that doesn’t scale with a large number of content partners.”
McCarthy’s take as he describes it is, “If we as an industry have done so much work to create internet-native formats, be it SRT or … Media over QUIC, where there is a level of resiliency, it’s easier to operate than a multicast UDP environment, how can we leverage the strength of the software network protocols and actually get adoption such that we can better scale our ingestion platform?” He adds that performance is a key part of it. “So we’re still having to work content partner by content partner [to] prove out the performance [and] prove out the technology to migrate them off of these legacy point-to-point fiber systems.”
Related Articles
With all of the inherent difficulties of delivering low-latency live streams at scale, and the growing interest in providing sports viewers with state-of-the-art multiview experiences, what additional technical challenges does multiview delivery create in streaming's fraught middle mile, and how do top-tier global broadcasters like Globo meet those challenges? Globo Head of Streaming and CDN Platform Marcos Petry discusses how Globo maintains and tunes streaming latency for multiview sports streams in this conversation with streaming consultant Bhavesh Upadhyaya at Streaming Media Connect in December.
30 Jan 2026
When pre-existing live broadcast operations add streaming for the same events or content, there's a temptation to "bolt on" streaming to the existing workflow and treat it as just another output or destination, but Warner Bros. Discovery distinguished video platform engineer Neal Roberts insists that doing so means sacrificing the streaming end-user experience in this conversation with Alchemy Creations founder and principal Andy Beach at Streaming Media Connect 2025. He says that also means duplicating operator requirements and goes on to discuss comms between streaming and broadcast teams and other best practices for optimizing experiences for all viewers regardless of platform.
09 Jan 2026
Guaranteeing a satisfying end user experience, whether you're delivering content live or VOD, requires resiliency, ensuring that the stream doesn't break down regardless of the scale, bursts, or other fluctuations in delivery demands. And the challenges are different for live and VOD, with live proving significantly more challenging in most instances. TAG Video Systems' Michael Demb, DAZN's Bob Hannent, and the CDN Alliance's Mark de Jong discuss the key challenges and how to address them in this clip from Streaming Media Connect 2023.
08 Jan 2024
Taking a multi-CDN approach would seem to be a no-brainer for delivering large-scale streams to global audiences and maximizing uptime in the face of bursts, unexpected regional demand, and other impediments to a smoothly delivered high-stakes stream. But DAZN's Bob Hannent says it's not always so, and a multi-CDN approach can actually introduce inefficiencies, in this discussion with CDN Alliance Chairman Mark de Jong at Streaming Media Connect 2023.
03 Jan 2024