Back to Blog

How Common Media Client Data (CMCD) Transforms Streaming Monitoring & Analysis

Photograph of an engineer on a computer working with data

If you are operating a streaming platform and aren’t aware of the CTA-5004 specification of Common Media Client Data (CMCD), you should do so immediately, and even make a plan for implementation. CMCD is a critical bridge between player and CDN data, allowing for better operational analysis and identification of issues with video delivery much quicker. 

However, as indicated in Touchstream’s upcoming industry research, The Challenges of Streaming Operations, you wouldn’t be alone in not only being unaware of CMCD but in having no real plan for getting it installed and operational. This article will thus show the necessity for CMCD within your monitoring workflow and demonstrate how combined with other monitoring techniques and technologies, it can improve viewer QoE.

How can CMCD help improve monitoring?

There are numerous challenges within streaming video operations, many of which were revealed in Touchstream’s recent ‘Challenges of Streaming Operations’ survey and fall into four primary buckets:

  • Data
  • Time
  • People
  • Automation

At the heart of all these challenges is a common thread: being able to resolve technical issues quickly to avoid negative impacts on operational KPIs. The relationship between two of the challenges is particularly telling as it relates to monitoring. The first, “connecting data together to identify the root cause of issues”, shows that streaming operators face multiple data sets that must be connected together to be meaningful, such as those from the player and the CDN. The second, “identifying last mile issues between CDN and ISP”, directly relates to the first. In fact, 60% of respondents who indicated that connecting data together was a “difficult challenge” also indicated that “identifying last mile issues” was problematic. That’s exactly where CMCD comes into play.

CMCD provides a linkage between CDN and player. When the player data shows rebuffer events, it can be tied to CDN logs, which may also show that there was a drop in the connection with the ISP through a break in requests for additional segments of the video. That data could be used to demonstrate that the player couldn’t connect to the ISP. It could also be used to prove that the service issue, which the viewer might complain to the streaming operator about, is neither CDN- nor provider-related. 

However, implementing CMCD is a double-edged sword. Yes, it can provide the linkage between player and CDN datasets, which improves telemetry and issue resolution, but it is also more data. One of the other challenges that streaming operators identified in the survey was finding problems before they affect viewer QoE, something that is only exacerbated by more data. Even if the data sets are connected, solving one of those challenges, there is more data to sift through, which is sure to increase the time to find a problem.

The solution? Implement CMCD with automation

Two of the other challenges identified by operators, “continuing to do simple tasks manually which could be automated” and “having enough people to manage operational issues” show that there is more than CMCD needed for streaming monitoring. Yet more data, even more of the right data, doesn’t solve the problems of manually sifting through it with people that the operator doesn’t have. 

Given the challenges that operators expressed in the survey, machine learning (ML) can provide the following streaming monitoring benefits:

  • Sampling data
  • Making connections
  • Visualising patterns

Sampling data

If one of the challenges is the time it takes to find problems, data volume can be a huge contributor to that. However, ML can help by only taking a sampling of data. The choice of which records to include and which to exclude can happen intelligently using ML. For example, let’s say that instead of capturing every record, you only capture every 10th until there is an issue (which can be identified by using ML to examine specific values against thresholds). When there’s an issue, the system captures every record until the issue has passed.

Making connections

Identifying individual issues takes time. Moreover, every person focused on trying to make data connections to uncover those issues is not available for other tasks. This is both a problem of too much data and not enough people. Machine learning, though, can thankfully help. Even if ML could accomplish 70% of the work, that’s 70% of time that could be reallocated. Remember that ML doesn’t take the place of human ingenuity, creativity, or intelligence. Rather, operations engineers could use that 70% saved time to analyse the results of ML, leading to a quicker understanding of the meaning of the connections resulting in lower MTTD.

Visualising patterns

At the heart of identifying issues quicker is effective visualisation. While ML can be used to sample data and make connections, it can also be employed to generate those visualisations. However, it’s not just graphs and charts. It’s also about visual indicators. For example, as the amount of data increases from sampling to every record (because a threshold for a data value has been exceeded), operational engineers could be alerted with a red/yellow/green indication in the dashboard.

ML is not about intelligence, it’s about automation

With CMCD added, the challenges of time, data, and people all get worse for operational engineers. ML provides a new layer of OTT monitoring automation to reduce the impact of the increased data. Not only do streaming operators get the added value of player-to-CDN connections afforded by CMCD, but with ML, they don’t have to shoulder the impact of all that extra data because many of the previously manual tasks carried out by engineers (connecting data sets together, identifying patterns, visualising issues) is automated through ML.

What’s more, through this ML automation, streaming operators can have much more visibility into the KPIs that drive their business. For example, as identified through the survey, streaming operators are still focused on the following KPIs (ranked in order of importance):

  • Buffer ratio
  • CDN errors
  • Overall latency

By combining ML automation with CMCD data, the ability to detect and resolve issues related to those KPIs is profoundly improved.

👉 You may like: Sustainable Streaming: A Greener Way to Video Quality Analysis

How to get from your current monitoring state to one with CMCD and ML

Just implementing CMCD into your monitoring and analytics stack isn’t enough. As described, doing so just increases the volume of data, which exacerbates all the challenges uncovered in our strategy. Carefully considered implementation is crucial. Anything that impacts analytics needs to go through a monitoring harness with a proper business logic layer.

CMCD is just another data source. As such, plugging it into the monitoring harness is vital, especially if the standard needs to be implemented across different players, and different player versions. Without doing that, ongoing management is more difficult, as the harness provides the mechanism by which to do that.

Using ML against all the data in your analytics stack, not just CMCD, is difficult if it’s not centralised. ML is integrated into the heart of the monitoring harness business logic layer. Rather than having different systems responsible for applying ML to different data sets, a single set of business rules can be applied to all data sets through the harness. This allows data sampling, connectivity, and visualisation across all the sets with a single set of rules.

Streaming operations is challenging, but it doesn’t have to stay that way

The Challenges of Streaming Operations research report by Touchstream and the Streaming Video Technology Alliance (SVTA) reveals exactly what is so challenging within streaming operators. It comes down to three things: data (not having enough people and getting access to the right sources), people (not having enough people), and time (taking too long to find and resolve errors).

CMCD can solve many of the data-related problems by providing linkage between players and CDNs, enabling operations engineers to overcome some of the challenges of last-mile visibility. Yet, by itself, it only adds to the challenges of monitoring. When combined with ML, it can solve more of them. Manual tasks can be automated, connections between data sets can be surfaced, and visualisations can be created in real-time, allowing operations engineers to leverage the skills that ML doesn’t possess: intelligence, creativity, and ingenuity.

However, none of this should be done in a vacuum. Most streaming operators face managing their players in a very fragmented environment. That means multiple CMCD implementations across a variety of devices. Integrating CMCD into a monitoring harness, though, can centralise the management of that deployment. Plus, combined with the harness’s business logic layer, rules for sampling and visualisations can be applied across all the data sets plugged into the harness, including CMCD.

Want to understand more about the challenges of streaming operations? Sign up here to receive your copy of the survey report the moment it is published.