Media Services live streaming best practices guide

Customers often ask how they can reduce the latency of their live stream. This article describes best practices for achieving low-latency live streams with in addition to live event encoding.

Note

Before you continue reading this article, read the Low Latency HLS (LL-HLS) article to understand low latency with live event encoding. Then, come back to this guide to understand what else may affect streaming latency.

There are many factors that determine the end-to-end latency of a stream besides how the media is encoded. Here are some that you should consider:

  1. Delays on the contribution encoder side. When customers use an encoding software such as OBS Studio, Wirecast, or others to send an RTMP live stream to Media Services. Settings on this software affect the end-to-end latency of a live stream.

  2. Delays in the live streaming pipeline within Azure Media Services

  3. CDN performance

  4. Buffering algorithms of the video player and network conditions on the client side

  5. Timing of provisioning

Contribution encoder

You are in control of the settings of the source encoder settings before the RTMP stream reaches Media Services. Here are some recommendations for the settings that would give you the lowest possible latency:

  1. Pick the physical region closest to your contribution encoder for your Media Services account. This will ensure that you have a great network connection to the Media Services account.

  2. Use a consistent fragment size. We recommend a GOP size of 2 seconds. The default on some encoders, such as OBS, is 8 seconds. Make sure that you change this setting.

  3. Use the GPU encoder if your encoding software allows you to do that. This would allow you to offload CPU work to the GPU.

  4. Use an encoding profile that is optimized for low-latency. For example, with OBS Studio, if you use the Nvidia H.264 encoder, you may see the “zero latency” preset.

  5. Send content that is no higher in resolution than what you plan to stream. For example, if you're using 720p standard encoding live events, send a stream that is already at 720p.

  6. Keep your framerate at 30fps or lower unless using pass-through live events. While we support 60 fps input for live events, our encoding live event output is still not above 30 fps. For Low Latency HLS, the fixed frame rate is recommended, and the max frame duration should not exceed 0.5 seconds for the best experience.

Configuration of the Azure Media Services live event

Here are some configurations that will help you reduce the latency in our pipeline:

  1. Use the low latency stream options for live events. For the Standard encoding (up to 720p) and Premium encoding (up to 1080p) stream options, unless you need a DVR window longer than 6 hours or smooth streaming output, use the Low latency stream latency setting.

  2. We recommend that you choose CMAF output for both HLS and DASH playback. This allows you to share the same fragments for both formats. It increases your cache hit ratio when CDN is used. For example:

    Type Format URL example
    HLS CMAF format=m3u8-cmaf https://amsv3account-usw22.streaming.media.azure.net/21b17732-0112-4d76-b526-763dcd843449/ignite.ism/manifest(format=m3u8-cmaf)
    MPEG-DASH CMAF format=mpd-time-cmaf https://amsv3account-usw22.streaming.media.azure.net/21b17732-0112-4d76-b526-763dcd843449/ignite.ism/manifest(format=mpd-time-cmaf)
  3. If you must choose TS output, use an HLS packing ratio of 1. This allows us to pack only one fragment into one HLS segment. You won't get the full benefits of LL-HLS in native Apple players.

Player optimizations

When choosing and configuring a video player, make sure you use settings that are optimized for lower latency.

Media Services supports different streaming protocol outputs – DASH, HLS with TS output and HLS with CMAF fragments. When using the LowLatencyV2 stream option, be sure to find a player that supports Low Latency HLS (LL-HLS). Depending on the player’s implementation, buffering decisions impact the latency a viewer observes. Poor network conditions or default algorithms that favor quality and stability of playback could cause players to decide to buffer more content upfront to prevent interruptions during playback. These buffers, before and during the playback sessions, would add to the end-to-end latency.

When Azure Media Player is used, the Low Latency Heuristics profile optimizes the player to have the lowest possible latency on the player side. This player only supports DASH unless it is being used on Safari on Apple devices.

CDN choice and optimizations

Streaming endpoints are the origin servers that deliver the live and VOD streaming content to the CDN or to the customer directly. It is best to use a Content Delivery Network (CDN) with a shielded origin to ensure that traffic for your media content is delivered efficiently.

We recommend using Azure CDN which is provided by Verizon (Standard or Premium). We've optimized the integration experience so that a customer could configure this CDN with a single select in the Azure portal. Be sure to turn on Origin Shield and streaming optimizations for your CDN endpoint whenever you start your streaming endpoint.

Our customers also have good experiences bringing their own CDN. Ensure that measures are taken on the CDN to shield the origin from excessive traffic.

You can also improve performance by configuring rules for the CDN profile. See How to enable CDN optimizations.

Streaming endpoint scaling

Note

A standard streaming endpoint/origins is a shared resource that allows customers with low traffic volumes to stream content at a lower cost. You would not use a standard streaming endpoint to scale streaming units if you expect large traffic volumes or you plan to use a CDN.

A premium streaming endpoint/origin offers more flexibility and isolation for customers to scale by adding or removing dedicated streaming units. A streaming unit is a compute resource allocated to a streaming endpoint. Each streaming unit can stream approximately 200 Mbps of traffic.

While you can concurrently stream many live events at once using the same streaming endpoint, the maximum default streaming units needed for one streaming endpoint is 10. You can open a support ticket to request more than the default 10.

Determine the premium streaming units needed

There are two steps to determine the number of streaming endpoints and streaming units needed:

  1. Determine the total egress needed.

  2. Divide the total egress by 200, which is the maximum Mbps each streaming unit can stream.

Determine the total egress needed

Determine the total egress needed by using the following formula.

Total egress needed = average bandwidth x number of concurrent viewers x percent handled by the streaming endpoint.

Let’s take a look at each of the multipliers in turn:

Average bandwidth. What is the average bitrate you plan to stream? In other words, if you're going to have multiple bitrates available what bit rate is the average of all the bitrates you're planning for? You can estimate this using one of the following methods:

For a live event that includes encoding:

  • If you don’t know what your average bandwidth is going to be, you could use our top bitrates as an estimate. Our top bitrate is 5.5Mbps for 1080p encoded live events, therefore, your average bitrate is going to be somewhere around 3.5Mbps.

  • Look at the encoding preset used for encoding the live event, for example, the AdaptiveStreaming(H.264) preset. See this output example.

For a live event that is simply using pass-through and not encoding:

  • Check the encoding bitrate ladder used by your local encoder.

Number of concurrent viewers. How many concurrent viewers are expected? This could be hard to estimate, but do your best based on your customer data. Are you streaming a conference to a global audience? Are you planning to live stream to sell a set of products to your customers?

Percent of traffic handled by the streaming endpoint. This can also be expressed as “the percent of traffic NOT handled by the CDN” since that is the number that actually goes into the formula. So, with that in mind, what is the CDN offload you expect? If the CDN is expected to handle 90% of the live traffic, then only 10% of the traffic would be expected on the streaming endpoint. The number used in the formula is .10 which is the percentage of traffic expected on the streaming endpoint.

Determine the number of premium streaming units needed

Premium streaming units needed = Average bandwidth x # of viewers x Percentage of traffic not handled by the CDN / 200 Mbps

Example

You've recently released a new product and want to present it to your established customers. You want low latency because you don’t want to frustrate your already busy audience, so you'll use premium streaming endpoints and a CDN.

You have approximately 100,000 customers, but they probably aren’t all going to watch your live event. You guess that in the best case, only 1% of them will attend, which brings your expected concurrent viewers to 1,000.

Number of concurrent users = 1,000

You've decided that you're going to use Media Services to encode your live stream and won't be using pass-through. You don’t know what the average bandwidth is going to be, but you do know that you'll deliver in 1080p (top bitrate of 5.5 Mbps), so your average bandwidth is estimated to be 3.5 Mbps for your calculations.

Average bandwidth = 3.5

Since your audience is dispersed worldwide, you expect that the CDN will handle most (90%) of the live traffic. Therefore, the premium streaming endpoints will only handle 10% of the traffic.

Percent handled by the streaming endpoint = 10% = 0.1

Using the formula provided above:

Total egress needed = average bandwidth x number of concurrent viewers x percent handled by the streaming endpoint.

total egress needed = 3.5 x 1,000 x 0.1

total egress needed = 350 Mbps

Dividing the total egress by 200 you determine that you need 1.75 premium streaming units.

premium streaming units needed = total egress needed/200Mpbs

premium streaming units needed = 1.75

We'll round up this number to 2, giving us 2 units needed.

Use the portal to estimate your needs

The Azure portal can help you simplify the calculations. On the streaming page, you can use the calculator provided to see the estimated audience reach when you change the average bandwidth, CDN hit ratio and number of streaming units.

  1. From the media services account page, select Steaming endpoints from the menu.

  2. Add a new streaming endpoint by selecting Add streaming endpoint.

  3. Give the streaming endpoint a name.

  4. Select Premium streaming endpoint for the streaming endpoint type.

  5. Since you're just getting an estimate at this point, don’t start the streaming endpoint after creation. Select No.

  6. Select Standard Verizon or Premium Verizon for your CDN pricing tier. The profile name will change accordingly. Leave the name as it is for this exercise.

  7. For the CDN profile, select Create New.

  8. Select Create. Once the endpoint has been deployed, the streaming endpoints screen will appear.

  9. Select the streaming endpoint you just created. The streaming endpoint screen will appear with audience reach estimates.

  10. The default setting for the streaming endpoint with 1 streaming unit shows that it's estimated to stream to 571 concurrent viewers at 3.5 Mbps using 90% of the CDN and 10% of the streaming endpoint.

  11. Change the percentage of the Egress source from 90% from CDN cache to 0%. The calculator will estimate that you'll be able to stream to 57 concurrent viewers at 3.5 Mbps at 200 Mbps without a CDN.

  12. Now change the Egress source back to 90%.

  13. Then, change the streaming units to 2. The calculator will estimate that you'll be able to stream to 1143 concurrent viewers at 3.5 Mbps with 4000Mpbs with the CDN handling 90% of the traffic.

  14. Select Save.

  15. You can start the streaming endpoint and try sending traffic to it. The metrics at the bottom of the screen will track actual traffic.

Timing

You may want to provision streaming units 1 hour ahead of the expected peak usage to ensure streaming units are ready.

Get help and support

You can contact Media Services with questions or follow our updates by one of the following methods: