Cliente
Saving Time with Stream Quality Notifications
You asked and we answered.
Introducing individualized stream quality notifications for your Microsoft Teams data.
We are launching an exciting new feature as part of our Collaboration Observability for Teams platform. Are you tired of monitoring dashboards just to see if there is a problem? We can do that for you and let you know if something needs your attention.
With our Stream Quality notifications for Collaboration Observability data, we utilize a sophisticated anomaly detection algorithm called PEWMA (defined below) to learn what is normal behavior for a location in your network. When a location displays behavior that significantly deviates from that of the norm, we will notify you. This will allow you to investigate and rectify the issue in a timely manner.
What Behaviors are We Monitoring?
When your company hosts a Live Microsoft Teams event, there are three streams that we are interested in: Audio, Video, and Video Based Screen Sharing (VBSS). Microsoft will classify these streams as either ‘Good’ or ‘Poor.’ What we do is take that data and turn it into a Good Stream Quality Rate. In other words, in each location, how often was your stream classified as ‘Good.’
For example, suppose that your location ‘The Moon’ had 100 audio streams on a given day and 85 of them were classified as ‘Good,’ keeping the numbers simple for easy math. Then we would say that on that day ‘The Moon’ had a Good Audio Stream Quality Rate of 85%.
What is PEWMA?
PEWMA stands for Probabilistic Exponentially Weighted Moving Average. Let’s break that down.
First, we are dealing with timeseries data, gathering data about stream quality over time. Timeseries data is inherently NOISY. With the natural fluctuations of timeseries data it can be difficult to tell what an anomaly is and what is just noise. That’s where the moving average comes in.
A moving average smooths out the data. We set a window and calculate the average. Then when you get a new data point, you move the window over, dropping the first data point and picking up the last. And again, you calculate the average. This smoothing effect makes it easier to see the trend of the data through the noise.
Now, an Exponentially Weighted Moving Average, or EWMA, simply gives more weight to recent data. Older data is weighted exponentially less. This makes the algorithm more responsive to recent changes.
To take this one step further, we include an extra factor using a Probabilistic Exponentially Weighted Moving Average, or PEWMA, which makes the algorithm more robust to outliers. Let’s say that ‘The Moon’ location had 5 days where the Good Audio Quality Stream Rate was 85%, 83%, 88%, 15% and 87%. As we can see, that 15% stands out and will likely be flagged as an anomaly to be alerted on (see next section).
Given that it was an anomaly, we don’t want it to impact the exponentially weighted moving average going forward. This is where the probabilistic part comes in – the probability of seeing 15% is low given what we’ve seen so far, and we can down-weight this data point in our moving average calculation going forward.
What Would Trigger a Notification?
As described above, we calculate the probabilistic exponentially weighted moving average over time at a given location. This essentially tells us what is ‘normal’ for that location. Each new datapoint is compared to the latest PEWMA value, and if it falls below this value in a statistically significant way, an alert is triggered. That alert then makes its way over to the Customer Portal as a notification for you with some information to help you get started troubleshooting.