Skip to main content

Skimming the cost of cloud DVR


As the use of cloud DVR grows – for personal recording, video-on-demand and especially time-shifted TV – operators face two inter-linked challenges: rising costs for storage space and pressure on origin servers to process an increasing number of content requests at peak viewing times. These challenges are exacerbated in geographies where local copyright laws require unique copies of content for every user. Nokia Bell Labs has developed a new just-in-time transcoding (JITX) technique called skim storage to significantly reduce storage costs and maintain the quality of experience for end-users.

Rising popularity of cloud DVR

Cloud DVR usage reported by our customers, and confirmed by recent findings from Nielsen, show that more than 90% of TV programs are watched in the week after they have been aired. After this 7-day window, there is a high probability that these programs are never watched. While the value of the content diminishes (Figure 1), these “long-tail” programs must still be stored. More importantly, multiple versions are also required for different adaptive bitrate (ABR) profiles.

With ABR, the content is encoded at various bitrates. Each version of the content is then broken into chunks of a few seconds each. On playback, different ABR profiles are streamed depending to the device – e.g. a 4” SD smartphone compared to 50” UHD TV – and depending on the quality of the user’s connection, which may change during streaming, especially if the user is on a mobile network. In this way, an optimal experience can be guaranteed for each user at all times.

However, multiple copies of each piece of content represent a huge cost in terms of storage, footprint and power consumption.

Figure 1: Content value decreases over time

Using JITX to save cloud DVR storage space

To save on storage costs for this long-tail content, just-in-time transcoding and packaging (JITX) is often used. With JITX, only a master copy of each piece of content is needed – the highest definition profile. As and when a user makes a request to view that content, JITX creates and streams the required ABR profile from the master copy, switching between profiles depending on the user’s connection. In this way, service providers do not need to store any intermediate quality ABR profiles. This reduces maintenance costs and creates more storage space, which increases the overall value for service providers. Additional space can be allocated to more valuable, recent content allowing, for example, a provider to monetize a 7-day catch-up feature for a premium channel.

JITX represents only 3-7% of the total CAPEX costs of a cloud DVR solution while storage capacity accounts for around 80%. In a private copy scenario, JITX typically reduces storage capacity requirements by approximately 50%, making a compelling business case for service providers.

However, JITX demands increased processing power in the origin server to generate the required ABR profiles on demand, and change between them on-the-fly. This can lead to an increase in latency, which results in buffering and negatively impacts the user experience.

This problem is especially apparent during peak viewing hours, as on-demand and time-shifted consumption follow a similar pattern to live TV. Most cloud DVR playback (34%) occurs between 7pm and 11pm, the same period as peak live TV viewing (Figure 2). This pattern has remained consistent over the years and means that origin servers are under extreme pressure during peak hours to process pause and rewind requests on live content as well as transcode catch-up and VOD content.

Figure 2: Viewing throughout the day

Overcoming JITX latency and processing challenges

To maximize costs savings in cloud DVR and ensure quality of service for users, Nokia has developed a new JITX technique called skim storage, based on a patented innovation from Nokia Bell Labs. Like JITX, skim storage creates significant storage savings but without needing extra processing power at the time of retrieval.

Skim storage begins by saving each piece of content in the highest profile as a master copy. However, skim storage goes further than JITX as it also saves a small amount of control stream data for every ABR profile that is required.

When a piece of content is requested, this information is used to simplify the execution of some of the heaviest tasks of the encoding process, reducing the complexity by a factor of 5 compared to standard JITX techniques (Figure 3). The encoding and packaging process becomes faster and simpler, requiring less processing power and reducing latency without affecting the quality of the content. The extra stored information generates a very limited increase in storage requirements.

Skim storage is built on top of Intel® Media SDK, a cross-platform API that ensures fast video playback, encoding, processing and media format conversion.

Figure 3: Nokia skim storage

Skim storage is particularly effective for long-tail content. According to Nielson, users watch an average of 180 minutes of DVR each week, yet only delete content that is more than 3 months’ old. A user’s long-tail content accounts for around 200 GB of storage in every personal DVR library. Similarly, long-tail content accounts for 75% of titles in an average VOD library. By applying skim storage to both cloud DVR and VOD storage, providers can make considerable savings in storage capacity (Figure 3) while maintaining QoS.

Figure 4: Space savings using skim storage

The business case for skim storage makes a lot of sense. Let’s take as an example a private copy use case with 200,000 cloud DVR subscribers recording 180 minutes per week, and an average lifetime of 3 months for any recorded content. ABR content has a total bitrate of 15 Mb/s and the highest quality profile is HD at 7 Mb/s. All content is “skimmed” after day 7.

At the egress we have assumed a maximum concurrency of 50% (100,000 users), DVR playout of 15% (15,000 users) and 3% long tail playout (450 users). In this scenario, skim storage provides a total cost saving of nearly 46%.

We unveiled our skim storage technique together with Intel at the 2017 NAB Show, demonstrating an increased transcoding density of 1500 percent compared to an x264 “medium” preset. With the continuing popularity of cloud DVR services, skim storage will enable providers to make more time-shifted and on-demand content available to users while maintaining quality of service and lowering capital and operating expenditure.

Álvaro Mattos Velasco

About Álvaro Mattos Velasco

Álvaro Mattos is part of the product managers team leading the Cloud DVR program of Nokia. Graduated in computer science and with a MBA, with more than 10 years in the software industry in a variety of sectors, he now manages the multicast adaptive bitrate and just in time transcoding solutions provided by Nokia.

Roland Mestric

About Roland Mestric

Roland Mestric is head of marketing for IP/optical networks automation at Nokia. In this role, Roland is responsible for defining and executing global marketing programs for the management and control of transport networks. Roland has 20 years of experience in multimedia and next-generation networks in pre-sales and solutions marketing roles with Nokia and Alcatel-Lucent.

Tweet me at @RolandMestric

Article tags