[Release] MistServer 3.3 stable release
Improved (live) metadata track support
Lower live streaming latency
We reworked the internals of our live streaming core code, lowering the minimum end-to-end latency between input and output from at least 1 frame on each track, to at least 1 frame on any track. Especially considering metadata tracks tend to be pretty sparse, this can make a massive difference in terms of latency - making it possible to achieve sub-second latency with less effort than ever before.
It is now possible to create short VoD clips on-demand from both VoD and live streams, in any/all supported output formats (including MP4, MKV, WebM, TS, FLV, etcetera).
Chromecast and video pop-out support
The built-in meta-player now supports playing back to and remote controlling Chromecast devices, as long as the player was loaded over HTTPS. It is now also possible to use the video pop-out feature some browsers offer.
HLS DVR input now full-featured
It is now possible to use an ongoing HLS-format recording as source for a stream that then will support both live playback and seeking into the full DVR window of the recording. This also works with remote HLS streams, both VoD and live. For now, the HLS streams must use TS segments (fMP4/CMAF segment support will come in a future version).
There is a set of new triggers and log messages that track why a process shut down / stopped, which greatly eases debugging and allows for more reliable tracking if an operation succeeded or failed.
[News] Online documentation now available
Traditionally MistServer had a PDF available as a manual, which contained great amounts of information. Perhaps a little too great as the PDF itself became bigger and bigger, and perhaps slightly too unwieldy.
We've decided to convert the manual to an online environment, allowing easier small edits and improving general usability.
We've reached a point where the majority of the documentation is now available and updated. So we're happy to start redirecting towards it. You can find it here. We've now started adapting some of the blog posts into How-to guides as well. Of course a proper search method will have to follow as well.
We will eventually put the documentation online in Github as well so users can easily suggest edits or request pages. Until we've reached that point feel free to email comments, suggestions or requests to email@example.com.
No worries for the PDF lovers out there, we will update the PDF at our usual irregular schedule.
[Blog] Deepdive into using RIST
What is the goal of this article
RIST (Reliable Internet Stream Transport) is an error correction protocol that ensures accurate transport of media streams across the public Internet. It is still a fairly recent standard, but now mature, and in daily use by media streamers even at the broadcast network level.
This article will explain the mechanics of RIST and give you some best practices/tips to get started. There is a widely available FOSS (Free and Open Source Software) implementation of RIST, called libRIST, and we shall refer to it here.
Do note that this will be written from a usage standpoint to keep things as easy accesible as possible, and in particular, to help the MistServer user implement this error correction protocol. Purely technical reasons as to why technique “X” would make something better than “Y” is a deeper dive than intended for this article.
What is the RIST standard?
A big benefit to RIST is that several companies, organized through the Video Services Foundation, maintain the spec. Some of the most noted experts in packet recovery contributed their knowledge and helped design the standard. One of the design goals is interoperability. A RIST implementation by “Company 1” is meant to be interoperable with another implemenation by “Company 2.”
When would you use RIST?
The biggest reason to use RIST is to move stream data over a lossy packet network; you know, like the Internet! Lab testing shows accurate playback across very extreme conditions, such as 50% packet loss. What you call a “bad hair” day over your Internet connection is probably a 1% loss. In fact, even the best corporate connections drop at least a few packets every hour, so an error correction protocol for important streams can benefit every user.
Right now, most RIST streams run between RIST servers, for transport from a source encoder to a media server which then distributes the stream in some other format such as HLS/DASH. In fact, end-user players such as VideoLAN’s vlc can already directly play a RIST stream! So if you’ve decided the time to familiarize yourself with RIST is now, you may be making exactly the right move!
Quality and reliability
RIST has the first obvious use case to deal with "bad network conditions". So when your connection from source to server or even between servers is expected to be weak you'll most likely be interested in RIST. Simply put, RIST saveguards the stream quality that goes through the connection and provides UDP transport that extra reliability layer that it's otherwise missing.
Simply putting an extra layer on top of the bits for delivery can adds latency, so surely RIST is added delay? Yes. It’s a trade-off.
Let’s explain how RIST works. The server receives a packet of media from the source, adds a sequence number and/or time stamp, and sends it off to the client. The server places a copy of the packet in a buffer. The client receives the packet… and the next packet, and the next, and so forth. But the client doesn’t immediately release each packet. The packets go into a buffer.
By examining the sequence numbers of the packets in the buffer, the client can send a re-request to the server if it finds a packet is missing or corrupted. The server then resends that packet. Once the client knows it has all the packets, it can then release the video stream to wherever it’s supposed to go.
As you can see, this means that you’ll want a buffer of at least three times the duration that it takes a packet to travel from server to client, or vice versa, plus a small amount to time for handling, or to wait and see if the original packet will arrive out of order, which sometimes happens on the Internet. If it’s a good RIST implementation, then the software will intermittently measure the “ping” time between server and client, and adjust the buffer time to be as short as possible, while also being efficient. Note also that RIST allows for time stamps (not all media formats include them). This allows for the client at the destination side to emit the stream at exactly the same speed that it was received at the source side by the server. The end result should be no (or very, very few) dropped packets, and no jitter. You may have read about humorous implentations of IP networks using carrier pigeons to transport packets; RIST would theoretically carry video over pigeon-Internet, though it might require a lot of tired pigeons!
The bottom line for a media provider is that at the cost of a latency of, for example, half-a-second to a few seconds, your streams will be flawlessly transported.
RIST does allow you to make a secure AES-encrypted tunnel from point-to-point distribution, or even point-to-many-point distribution, using either pre-shared key, rotating-keys and an optional authorization protocol based upon existing protocols, and currently submitted as an Internet RFC (Request for Comment).
What does RIST do differently compared to SRT?
The current leader in the field of error correction transport is Haivision SRT (Secure Reliable Transport). In fact, SRT and RIST provide similar functions. Some of the companies and engineers who designed RIST also sell implementations of SRT. But you could point to some differences in approach:
SRT tends to focus on professional installations. Engineers at a given location expect a known or constant latency between source and destination, and that is usually the starting point in the configuration of the link. Note that while there is a free implementation of SRT, large customers generally use non-free versions.
RIST, depending upon the implementation, can dynamically set and modify the buffer in accordance with changes in network conditions. This sometimes means that from start of receiving the stream to viewing it may take longer. The buffer’s size is being determined and the buffer filled. It also, generally, means a lesser overhead (in CPU processing and average bandwidth per connection).
Note also that while there are paid versions available, a free version, libRIST, is already extensively used by at least one of the big three U.S. broadcast networks. libRIST also provides a very flexible license so that developers can easily incorporate it into their own, free or non-free projects.
RIST and MistServer
There’s one very big reason to choose RIST when using MistServer: RIST is able to provide a buffer window, which is incompatible with most other applications. MistServer, however, can handle this just fine. We’ll obviously have to document this for you elsewhere! The benefit is that when you use RIST with MistServer, your latency/quality balance can actually adapt to your current connection. Most connections only do this at the start if at all.
What are the RIST profiles
Profiles are specified “flavors” of RIST. Each successive profile implements the features of the previous and adds additional features beyond the previous.
Simple (Released in October 2018)
The first and most common implementation of RIST. The core of the Simple profile is to just get the error corrected transport going and optimize for quality/latency. Note that RIST Simple profile require two ports, with the first an even number. Encryption is supported externally (DTLS). Stream types must be RTP or UDP.
Main (Released in March 2020)
The second profile specification released. It adds support for encryption with AES, has an optional authorization process, and supports point-to-multipoint services (or vice versa). It also supports multi-path routing, allowing for load balanced (for redundancy) or split (for speed) paths using two Internet service providers on the source side. It provides for GRE tunneling, thus supporting multi-plexing of streams and compatibility for any type of stream format. It also makes it easier for different sides to “initiate” the connection vs. the start-of-stream. This makes it very useful when one side or the other is behind a firewall.
Advanced (Released in October 2021)
This profile supports additional tunneling capabilities (effectively providing VPN-like services) and realtime LZO compression (which is surprisingly effective for very high density streams, or for WebRTC)!
When do you choose which profile?
In testing start with Simple profile and work your way up. In production, if you’re not too concerned with security, then Simple will do. It’s more likely, however, you’ll use Main profile.
Usage quickstart guide
If you start with libRIST, you may wish to first try a libRIST to libRIST connection. There is a step-by-step guide with command line examples and explanations.
The rest of this article’s quickstart will focus upon MistServer and libRIST. MistServer (3.0 and higher) incorporates libRIST code for their RIST implementation.
Note that RIST implementation in Mistserver up to version 3.2 requires compiling your own version of MistServer. RIST is fully included in 3.2 and every release after.
All you need to do to use rist is provide MistServer a source or push target starting with
rist://. libRIST’s quick start and documentation explain the URLs thoroughly! A
rist:// URL can incorporate multiple sources or targets as well as multiple streams, each with different parameters! Start small, and work your way up!
Another thing to keep in mind is that the stream buffer within MistServer also determines how much should be sent to the other side when connection is made. So keeping that low or high depending on your goal may help as well.
Understanding the RIST defaults
The default RIST values for MistServer are compatible with libRIST and will work well for most networks and other application targets by default. We always recommend changing settings to fit your needs, but the defaults are a great starting point.
The defaults are made to work with networks fitting within:
- Buffer sizes from 50 ms to 30 seconds
- Networks with round trip times from 0ms to 5000ms
- Bitrates from 0 to 1 Gbps
- Packet size should be kept under the path's MTU (typically 1500). The library does not support packet fragmentation.
- Bi-directional networks (not one-way satellite systems, though Advanced profile and other updates to the RIST specification are in progress).
When used with MistServer, RIST creates the connection between two points. An input must connect with an output. In order to match the connection an address and port is needed.
You will sometimes see a
rist:// URL in which an
@ character precedes an IP address/port (as in
@192.168.1.1:12345; IPv6 is also supported by the way). This applies to an IP on your host, and signifies that your host is to be in listening mode: it waits for the other side to contact it to establish the connection. Without an
@ character, it means, just start sending to (or receiving from) at that address/port.
MistServer using a RIST stream as input
The MistServer input will listen at an address/port (or on all interfaces through
0.0.0.0 if not set). This assumes the source sends to the specified input address. Note that multicast is supported, with the device name being a parameter.
@: This signifies that the MistServer host will wait for an incoming RIST connection.
address: Optional for input side, if given it will listen on that specific address, if unset it will listen on all addresses.
port: UDP port to use
parameters: Additional options you can set, see below for all available parameters.
MistServer sending a RIST stream to a location
The MistServer will reach out to the given address and start sending a stream towards it using RIST transport. It assumes the other side listens for it.
address: Must be set for output side, the address to connect towards.
port: UDP port to use
parameters: Additional options you can set, see below for all available parameters
The most common mistake is mismatched parameters. If, for example, the sender encrypts at AES 128, and the receiver at AES 252, transport will not be possible.
Simple Protocol Port Assignments
As mentioned previously, Simple Protocol uses two ports, and the first one must be an even number. Currently MistServer will enforce you to use even ports just in case. So an error message that you should pick an even port will come up regardless of profile!
Routing and Firewalls
Where possible use ping and tracert to verify each side and reach the other. You can also search the web for articles regarding UDP pings – they can be done using nmap or netcat, and can help you verify that a path to the precise port you wish to use is open.
Parameters aren’t working
Note that libRIST has a verbosity setting which is extremely useful in debugging what might be the cause.
How do You “Tune” Your Connection for:
Latency and Efficiency
You can trust the latest versions of libRIST and MistServer to negotiate and auto-configure the best latency values for efficiency and reliability. But we encourage you, especially if you’re doing this for the first time, to understand manual setup as well.
Also, when you expect major transport problems, you may wish to “tune” your settings for a specific connection. Or if you’re MistServer installation is not quite up-to date, you may also wish to manually configure.
You’ll find that MistServer works best with a dynamic buffer size for its RIST connection. The
buffer-min and buffer-max. Parameters set the absolute smallest and largest amounts in milliseconds that your buffer shall use. The example below sets very small and large values.
When RIST starts, it UDP-pings the other side to measure the round trip time. RIST then sets the initial buffer to six times the RTT whenever you manually set the buffer-min and buffer-max. Thereafter, by monitoring pings between the server and client, RIST can shrink or enlarge the buffer to the min or max, based upon network conditions. The “viewer” of the stream sees a constant playback because RIST has previously enlarged the buffer at the first sign of network problems.
RTT-min and RTT-max are related parameters. Their primary use is to “signal” to the RIST algoritms that there may be either a very good network or very bad network condition ahead, based upon the difference between the min and max. In testing, ping the other side about 100 times (ping -c 100), or at different times of day. Make a note of the minimum and maximum values in milliseconds, and substitute the values in the example below:
This would tell libRIST to calculate the lowest latency and best quality sweet spot between 10ms and 1000ms and keep doing this for as long as the connection is busy. RIST will shrink or enlarge the buffer as network conditions change.
In general, you’ll use buffer-min and max when you’re guessing at network quality, and rtt-min and max when you can actually test the network quality. Another thing you can do is change the MistServer default buffer size from 50000 to a lower amount, 10000 for example. Do note that MistServer will make this buffer higher if it’s needed to output certain protocols like HLS.
Note: MistServer RIST defaults
MistServer, its built in RIST transport defaults to:
You’ll be using the defaults automatically by simply by filling in the bare minimum:
URL Parameter List
Note that if you use libRIST, you may see help text with rist-sender --help, ristsender –help-url, plus the same for ristreceiver. We might also note here that if you have multiple MistServers set up at multiple points-of-presence, libRIST includes a third binary (rist2rist) which acts a “switching point” that can “distribute” from your original source to multiple RIST receivers without having to decode/re-encode the stream.
For every parameter below we’ll have it set with its default as used in MistServer:
Simple, Main and Advanced profile
Default 1000ms. (both min/max), the buffer by which stream data could be delayed between in and out (jitter is compensated for).
Default 100000 Kbps. Sets the maximum bandwidth in Kbps for the connection. Measure your stream bitrate and recommended to use 10% higher for a constant bitrate, 100% higher for variable bitrate.
Default 0 Kbps (no limit). Sets the maximum bandwidth from receiver to sender (communication of re-requests, pings, metrics) in Kbps.
Default 25ms. Sets a secondary buffer in ms to reorder any out-of-order packets. Usually not necessary to change.
Default (not set). The RTT (in ms) determines the time between requests and makes sure the spacing between them fits the network conditions. Setting this value sets both the rtt-min/max to the same value, which is not recommended.
Default 0 (not set). Arbitrary name for stream for display in logging. For MistServer the stream name will be copied in here if possible.
Default 5. Relative share for load balanced connections, distribution paths will be determined by the weight in comparison to other weights. Use 0 for multi-path.
By default equal to buffer in ms. Determines the shortest the buffer is allowed to be.
By default equal to buffer in ms. Determines the longest the buffer is allowed to be.
Default 50ms, the RTT will not search below this value. Will set itself to 3ms if you try to set it lower.
Default 500ms, the RTT will not search above this value. Will overwrite rtt-min.
0 = RTP Timestamp (default); 1 = Arrival Time, 2 = RTP/RTCP Timestamp+NTP.
The RIST specification does not mandate time synchronization. Using this parameter, librist shall attempt to release the packets according to the time stamp indicated by the option specified. When not set, it emits the media packets at a speed equal to that at which the packet was received at the sender side plus the time buffered. Note that the Network Time Protocol option is designed so that you can synchronize playback of multiplexed streams using ntp plus the buffer size as a guide. The allowed values are1, for Arrival Time and 2, for RTP/RTCP Timestamp plus NTP. Note that this different than the RTP-timestamp=# and the RTP-sequence=# URL parameters in that the latter two will not attempt to synchronize the release of the packets to the player.
Default 1968. the port within the GRE (Generic Routing Encapsulation) tunnel. This has nothing to do with the media port(s). Assume the GRE is device /dev/tun11, having an address of 18.104.22.168, and you set the virtual destination port to 10000 and your media is using port 8193/4. The operating system will use 22.214.171.124:10000 for the GRE. As far as your media source and media player are concerned, the media is on ports 8193/4 on their respective interfaces. The media knows nothing of the tunnel.
0 = Simple, 1 = Main, 2 = Advanced
Default main. Rist profile in use.
Disable -1; Error 3, Warning 4, Notice 5, Info 6, Debug 7, simulation/dry-run 100
Default level is 6. Allows you to set the verbosity of the RIST protocol, 100 is used to do a simulation/dry run of the connection. High and Advanced only
High and Advanced only
128 = AES-128, 256 = AES-256,
Specifies the specific encrytion. Specify “128” for AES-128 or “256” for AES-256. Remember that you must also specify the pass phrase. Both sides must match the passphrase (“secret” parameter).
The encryption passphrase that must match on both sides. Requires an aes-type to be set.
Default 2000 in ms. terminates the RIST connection after inactivity/lack of keepalive response for the limit (in milliseconds) which you set with this parameter.
Default 1000 in ms. Time in milliseconds between pings. As is standard practice for GRE tunnels, the keep alive helps ensure the tunnel remains connected and open should no media be traversing it at a given time.
Default 1ms. sets the key rotation period in milliseconds when aes and a passphrases are specified.
mitigation mode: (0=disabled, 1=normal, 2=aggressive)
libRIST provides built in congestion control, which is important in situations in which a sender drops off the connection, but the receiver still sends re-requests. The three options for this parameter are 0=disabled, 1=normal and 2=aggressive. In general, don’t set the parameter to “aggressive” unless you’ve definitely established that congestion is a problem.
Default 6. sets a minimum number of re-requests for a lost packet before congestion control kicks in. Note that setting this too high can lead to congestion. Regardless of this setting, the size of the buffer and the roundtrip time will render too high a minimum value here irrelevant.
Default 20. sets a maximum number of re-requests for a lost packet.
Default 5 Relative weight for multi-path load balancing. Use 0 for duplicate paths.
this corresponds to the srp-auth credentials defined (globally) on the “other” side, when the “other” side is in listen mode with an srp-auth file holding the corresponding credentials. Note that libRIST includes a password utility. If you’re familiar with Apache’s htpasswd, it works just like that.
this corresponds to the srp-auth credentials defined (globally) on the “other” side, when the “other” side is in listen mode with an srp-auth file holding the corresponding credentials.
Controls how rist payload is muxed/demuxed (-1=auto-detect, 0=rist/raw, 1=vrtsrcport, 2=ipv4)
When using mux-mode=ipv4, this is the string to be used for data filter. It should be written as destination IP:PORT
Advanced Profile only
1 for enable, 0 for disable
enable lz4 levels
Usage: append to end of individual
rtp:// url(s) as
The device name to multicast on (e.g. eth0), if unset uses the system default.
ID number (arbitrary) for multiplex/demultiplexing steam in peer connector.
carry over the timestamp to/from the RTP header into/from rist (0 or 1).
carry over the sequence number to/from the RTP header into/from rist (0 or 1).
override the default RTP PTYPE to this value. RFC 3551 describes the standard types.
[News] Exhibiting at IBC2023
Like always we'll be exhibiting at IBC. We have our own booth 5.B57 in Hall 5 (Content Everywhere). If you're visiting Amsterdam during 15-18 September this year be sure to drop by!
This year Marco van Dijk is giving a presentation about our new creation: Streamcrafter
Date: Saturday 16 September
Stage: 1 (Stand 5A.28)
It'll be the first showcase of it. We'll want to keep the specifics a surprise for the presentation, but what we can share at this point is that Streamcrafter is a in-browser broadcasting studio.
[Release] 3.2.2 bugfix release
Bugfix release 3.2.2 is now available! We've made several minor improvements that we deemed important enough to release, but did not warrant a full stable release. Downloads can be found here and the full changelog can be found here.
- Feature: Support for AV1 codec in MP4 input and output
- Bugfix: RTMP push output sent initial data with wrong timestamps, fixed
- Bugfix: The new UDP packet pacing code accidentally broke WebRTC ingest, fixed
- Bugfix: A rolling restart will no longer shut down a large percentage of segmented-protocol (e.g. HLS) viewers
- Bugfix: Fix local file path DTSH writing (no longer written to a folder called "file:")
- Bugfix: HTTP parser no longer chokes on reading an empty response body in non-chunked transfer modes
- Bugfix: Incoming TS-SRT pushes set their UTC offset correctly now
- Bugfix: Fixed track masking feature in MistProcMKVExec
- Bugfix: If an TS-SRT push comes in and does not immediately send media data, no longer fails to receive the push
- Bugfix: Packet jitter calculator no longer shared between inputs of single-process inputs (e.g. TS-SRT, TS-RIST)
- Bugfix: Fixed output session shutting down while waiting for a
pushdelaysetting to be applied
- Bugfix: SRT and RIST listening sockets no longer kill all connections when there is a viewer for a mid-shutdown stream connected over these protocols
- Bugfix: Removed dead code in Meta-Player RAW WebSocket wrapper