Vendor Content.

Designing For Streaming At Scale – Combining Hydrology & Plumbing (aka Computing & Networking)

Delivery of anything – a product, a service, a project, or a pizza – is not a trivial task. Logistics experts around the world spend their whole careers figuring out how to deliver on-time and with the right levels of quality and efficiency to their customers. This day-in, day-out challenge is full of problems to solve, all with the aim of delivering what customers want.

Sergio Carulli - Chief Innovation Officer, MainStreaming.

Sergio Carulli - Chief Innovation Officer, MainStreaming.

Effective logistics in any industry depends on two key points. First, the capacity must be in place to meet all demand on-time – there must be enough workers / drivers / machines / vehicles / fuel / etc. Second, the capacity must be used intelligently to achieve maximum productivity and best value for money for customers – there needs to be a dynamic way to understand and manage changes in demand and supply conditions (including capacity availability). Imagine a fleet of delivery vehicles trying to ensure every customer receives their parcel during a very specific window of time on a specific day, overcoming road traffic, weather, roadworks, accidents, vehicle breakdowns, driver sickness, and any other number of obstacles. Delivery Capacity plus Intelligent Delivery Management are the two critical ingredients to achieve highly effective and efficient delivery.

Before moving to video streaming, let’s talk first about the concepts involved in the delivery of video segments to consumers who have pressed play. Streaming is the name we use for delivering video over the internet, and the water connotations this name carries are helpful for us to discuss video streaming delivery concepts. In water management there are two distinct disciplines – plumbing, which is defined as a system that conveys fluids for a wide range of applications (e.g., cleaning, irrigating, drinking, etc.), and hydrology, which is the study of the movement, distribution, and management of water. When we consider how we consume water as a basic need in our lives, both plumbing and hydrology are fundamental.

In video streaming, these two concepts are fully applicable. Plumbing refers to the content delivery networks that are implemented to transport video segments from encoders to CDN edges. As a networking concept, plumbing also refers to the internet in general as well as ISP’s broadband and mobile networks. Hydrology refers to the study of streaming video consumption and associated performance planning and management, which places an important emphasis on capacity planning and management of networks and computing resources.

When we develop and deliver streaming video platforms to achieve both quality and efficiency of video delivery, we need to think about the plumbing and the hydrology at the same time. One without the other will not deliver the optimal result we need.

First Principles Of Video Streaming Delivery

Video streaming is a pull system. In other words, we only deliver content that is requested. Anything else is wasteful, but there is always an efficiency balance to strike between viewer experience (e.g., fast start-up time, low latency, no rebuffering) and delivery efficiency (e.g., content in the right place at the right time to deliver the desired viewer experience). The reason that streaming is a pull system is because we are delivering a form of customized viewing experience, including VOD, catch-up TV, and FAST. All these use cases, that represent over 80% of total viewing, are unique to the viewer in some way. When we also factor in targeted advertising, we increase the individual customization of the streaming experience.

Perfect delivery exists when supply chain bottlenecks do not exist. As we do not live in a perfect world, delivery platforms routinely experience bottlenecks and disruptions as indicated in Figure 1. These disruptions need to be managed, prevented, and circumnavigated to ensure the delivery gets through to the final recipient.

Figure 1 – examples of bottlenecks with capacity and video management resolutions:<br />Scenario 1: Create direct connection to ISPs to avoid IXP bottlenecks.<br />Scenario 2: Reroute stream to find uncongested route to the viewer.<br />Scenario 3: Create Edge capacity inside the ISP network (as close as possible to the access network) to avoid ISP Core Network bottlenecks.

Figure 1 – examples of bottlenecks with capacity and video management resolutions:
Scenario 1: Create direct connection to ISPs to avoid IXP bottlenecks.
Scenario 2: Reroute stream to find uncongested route to the viewer.
Scenario 3: Create Edge capacity inside the ISP network (as close as possible to the access network) to avoid ISP Core Network bottlenecks.


Hydrology then plumbing, or plumbing then hydrology. Which comes first? Is it a “catch-22” situation? Actually, the answer is simple. Hydrology is first.

The design-work to enable optimized video streaming delivery is a hydrology discipline. This design work must be done first, and then the plumbing must be implemented according to the design. If we do the plumbing first, the “water” can easily end up in the wrong place, or there can be unforeseen bottlenecks. Then the customers will be upset with performance, the pipes could burst, and we would have a long list of problems to manage.

Hydrology in video streaming is about knowing audience consumption patterns. Who wants to watch what, where and when? If we understand the consumption profile, we can plan and manage the necessary resources to transport the video from its origination point to each and every viewer, achieving the required quality and cost metrics in the process.

Streaming at scale is one of the media industry’s biggest technical challenges. Big brands like Viaplay, Apple, Amazon Prime, and Netflix have had significant technical problems streaming live events to large audiences in the last 6-12 months. To continue with the water analogies, streaming at scale means video flows through the delivery pipes like a large wave which places high, and often excessive, pressure on the plumbing. An important characteristic of live streaming is that it is often localized in nature when it happens. A good example is a local sports “derby” – when AC Milan play Inter-Milan, or Manchester United play Manchester City, or the New York Giants play the New York Jets. Guess where most viewers stream from in these examples? Guess where the streaming capacity needs to be to deliver excellent video quality? The bottom line – live streaming to specific geographies can place excessive pressure on delivery capacity, which can result in significant delivery quality issues.

But one might ask, isn’t most of this content delivery over optical fiber connections that deliver data at lightspeed? Yes, it is. So, people can be served video from an Edge location that is hundreds of miles away and still receive video segments in milliseconds? Yes, they can. This is why streaming video over the internet works most of the time, and we can trust the basic technology to scale and support our needs in future. But when a lot of people request a lot of content – whether video on demand or live streaming, or game downloads, or video calls, etc. – then we can easily experience congestion and bottlenecks. And congestion is the enemy of delivery quality – just like a traffic jam will delay your parcel delivery and may really disrupt the overall resource plan of the parcel delivery company. Congestion along a video pathway is what causes quality issues like buffering and long latency. And the longer the video pathway, and the more congested it is, the higher the risk that quality issues will occur.

Hydrology is critical to ensure that video delivery systems can function well under pressure. The design of the system should enable excellent delivery even when congestion is high – whether congestion is caused by unexpectedly high demand (e.g., more people stream simultaneously than expected, or the same people stream at a higher rate than expected) or by unexpectedly low capacity (e.g., a computing or network outage, or general reduction in available resources). Hydrological thinking is how we efficiently plan for unexpected eventualities without over-engineering the plumbing, which would be irresponsible from a sustainability perspective.

Plumbing, on the other hand, works to the very simple principle that delivering services from near to the consumer is better. In video streaming, because it is a pull system, we know where Edge servers should be for best streaming video delivery performance – as close as possible to the consumer.

The technical performance issues we have observed recently from leading streamers tells us two things – either they did not have sufficient capacity (“plumbing”) in the right place to fulfil the demand, and/or they did not have the necessary capacity management tools (“hydrology”) to make maximum use of the capacity they did have available in order to fulfil the demand.

In the end, hydrology and plumbing are both critical to deliver streams to consumers. And while good plumbing in all the right places can result in high-performance video streaming at scale, this is not yet available today in every scenario (e.g., local sports derbies). The ideal solution now is for intelligent hydrology to allow for efficient plumbing to be built out.

Design Principles

MainStreaming approached this hydrology problem from the ground up and therefore built its Video Edge solution based on the following principles:

  • Distribute computing resources and delivery management at the Edge.
  • Process as much content as possible at the Edge.
  • Create ultra-low-latency inter-communication paths between all Edges in the network.
  • Ask for content from the Origin only once.
  • Implement ISP-optimized routing methodologies.
  • Optimize software and hardware for video streaming delivery.
  • Easily scale performance monitoring and routing management with video streaming.
Figure 2 – Video Edge architecture for sustainable streaming at scale.

Figure 2 – Video Edge architecture for sustainable streaming at scale.

The Video Edge solution designed this way results in:

  • Less total hardware in the system.
  • Maximum system resilience through distribution of resources.
  • Deployment of Edges deep inside ISP networks.
  • Lowest possible Origin egress.
  • Best possible QoE for viewers.
  • Lower energy consumption.
  • Cost savings.

In the MainStreaming whitepaper “Broadcast-grade Streaming” we explain in detail how software (i.e., hydrology) and hardware (i.e., plumbing) co-exist and cooperate in a video delivery network, and how DAZN delivers best in class QoE with this approach.

For more information about MainStreaming, please contact us here, or book an IBC meeting here.