Skip to content

Latest commit

 

History

History
434 lines (318 loc) · 23.7 KB

srt-live-transmit.md

File metadata and controls

434 lines (318 loc) · 23.7 KB
Raw

srt-live-transmit

The srt-live-transmit tool is a universal data transport tool with a purpose to transport data between SRT and other medium. At the same time it is just a sample application to show some of the powerful features of SRT. We encourage you to use SRT library itself integrated into your products.

Introduction

The srt-live-transmit can be both used as a universal SRT-to-something-else flipper, as well as a testing tool for SRT.

The general usage is the following:

srt-live-transmit <input-uri> <output-uri> [options]

The following medium types are handled by srt-live-transmit:

  • SRT - use SRT for reading or writing, in listener, caller or rendezvous mode, with possibly additional parameters
  • UDP - read or write the given UDP address (also multicast)
  • Local file - read or store the stream into the file
  • Process's pipeline - use the process's stdin and stdout standard streams

Any medium can be used with any direction, although some of them may have special direction-dependent cases.

Mind that the URI has a standard syntax:

scheme://HOST:PORT/PATH?PARAM1=VALUE1&PARAM2=VALUE2&...

The first parameter is introduced with a ? and all following can be appended with an & character.

If you specify only the path (no :// specified), then the scheme defaults to file. The path can be also specified as relative this way. Note also that empty host (scheme://:PORT) defaults to 0.0.0.0, and an empty port (when there's no :PORT part) defaults to port number 0.

Special options for particular medium may be specified in PARAM items. All options are medium-specific, although there may happen some options common for multiple media types.

Note also that the HOST part is always tried to be resolved as a name, if its form is not directly the IPv4 address.

Example for Smoke Testing

First we need to start up the srt-live-transmit app, listening for unicast UDP TS input on port 1234 and making SRT available on port 4201. Note, these are randomly chosen ports. We also open the app in verbose mode for debugging:

srt-live-transmit udp://:1234 srt://:4201 -v

Now we need to generate a UDP stream. ffmpeg can be used to generate bars and tone as follows, doing a simple unicast push to our listening srt-live-transmit application:

ffmpeg -f lavfi -re -i smptebars=duration=300:size=1280x720:rate=30 -f lavfi -re -i sine=frequency=1000:duration=60:sample_rate=44100 -pix_fmt yuv420p -c:v libx264 -b:v 1000k -g 30 -keyint_min 120 -profile:v baseline -preset veryfast -f mpegts "udp://127.0.0.1:1234?pkt_size=1316"

You should see the stream connect in srt-live-transmit.

Now you can test in VLC (make sure you're using the latest version!) - just go to file -> open network stream and enter srt://127.0.0.1:4201 and you should see bars and tone right away.

If you're having trouble, make sure this works, then add complexity one step at a time (multicast, push vs listen, etc.).

URI Syntax

Transmission mediums are specified as the standard URI format:

SCHEME://HOST:PORT?PARAM1=VALUE1&PARAM2=VALUE2&...

The applications supports the following schemes:

  • file - for file or standard input and output
  • udp - UDP output (unicast and multicast)
  • srt - SRT connection

Note that this application doesn't support file as a medium, but this can be handled by other applications from this project.

Medium: FILE (including standard process pipes)

NB! File mode, except file://con, is not supported in the srt-file-transmit tool!

The general syntax is: file:///global/path/to/the/file. No parameters in the URL are extracted. There's one (non-standard!) special case, though:

file://con

That is, con is used as a HOST part of the URI. If you use this URI for <input-uri>, then the data will be read from the standard input. If <output-uri>, the data will be send to the standard output. Be careful with options being specified together with having standard output as output URI - some of them are not allowed as the extra output controlled by options might interfere with the data output.

Medium: UDP

UDP can only be used in listening mode for input, and in calling mode for output. Multicast Streaming is also possible, without any special declaration. Just use an IP address from the multicast range. The specification and meaning of the fields in the URI depend on the mode.

The PORT part is always mandatory and it designates either the port number for the target host or the port number to be bound to read from.

For sending to unicast:

udp://TARGET:PORT?parameters...

  • The HOST part (here: TARGET) is mandatory and designates the target host

  • The iptos parameter designates the Type-Of-Service (TOS) field for outgoing packets via IP_TOS socket option.

  • The ttl parameter will set time-to-live value for outgoing packets via IP_TTL socket options.

For receiving from unicast:

udp://LOCALADDR:PORT?parameters...
  • The HOST part (here: LOCALADDR) designates the local interface to bind. It's optional (can be empty) and defaults to 0.0.0.0 (INADDR_ANY).

For multicast the scheme is:

udp://GROUPADDR:PORT?parameters...
  • The HOST part (here: GROUPADDR) is mandatory always and designates the target multicast group. The @ character is handled in this case, but it's not necessary, as the IGMP addresses are recognized by their mask.

For sending to a multicast group:

  • The iptos parameter designates the Type-Of-Service (TOS) field for outgoing packets via IP_TOS socket option.

  • The ttl parameter will set time-to-live value for outgoing packets via IP_MULTICAST_TTL socket options.

  • The adapter parameter can be used to specify the adapter to be set through IP_MULTICAST_IF option to override the default device used for sending

For receiving from a multicast group:

  • The adapter parameter can be used to specify the adapter through which the given multicast group can be reached (it's used to bind the socket)

  • The source parameter enforces the use of IP_ADD_SOURCE_MEMBERSHIP instead of IP_ADD_MEMBERSHIP and the value is set to imr_sourceaddr field.

Explanations for the symbols and terms used above can be found in POSIX manual pages, like ip(7) and on Microsoft docs pages under IPPROTO_IP.

Medium: SRT

Most important about SRT is that it can be either input or output and in both these cases it can work in listener, caller and rendezvous mode. SRT also handles several parameters special way, in addition to standard SRT options that can be set through the parameters.

SRT can be connected using one of three connection modes:

  • caller: the "agent" (this application) sends the connection request to the peer, which must be listener, and this way it initiates the connection.

  • listener: the "agent" waits to be contacted by any peer caller. Note that a listener can accept multiple callers, but srt-live-transmit does not use this ability; after the first connection, it no longer accepts new connections.

  • rendezvous: A one-to-one only connection where both parties are equivalent and both connect to one another simultaneously. Whoever happened to start first (or succeeded to punch through the firewall) is meant to have initiated the connection.

This mode can be specified explicitly using the mode parameter. When it's not specified, then it is "deduced" the following way:

  • srt://:1234 - the port is specified (1234), but host is empty. This assumes listener mode.
  • srt://remote.host.com:1234 - both host and port are specified. This assumes caller mode.

When the mode parameter is specified explicitly, then the interpretation of the host part is the following:

  • For caller, it's always the destination host address. If this is empty, it is resolved to 0.0.0.0, which usually should mean connecting to the local host

  • For listener, it defines the IP address of the local device on which the socket should listen, e.g.:

srt://10.10.10.100:5001?mode=listener

An alternative method to specify this IP address is the adapter parameter:

srt://:5001?adapter=10.10.10.100

The rendezvous mode is not deduced and it has to be specified explicitly. Note also special cases of the host and port parts specified in the URI:

  • CALLER: the host and port parts are mandatory and specify the remote host and port to be contacted.
    • The port parameter can be used to enforce the local outgoing port (not to be confused with remote port!).
    • The adapter parameter is not used.
  • LISTENER: the port part is mandatory and it specifies the local listening port.
    • The adapter parameter can be used to specify the adapter.
    • The host part, if specified, can be also used to set the adapter - although in this case mode=listener must be set explicitly.
    • The port parameter is not used.
  • RENDEZVOUS: the host and port parts are mandatory.
    • The host part specifies the remote host to contact.
    • The port part specifies both local and remote port. Note that the local port is this way both listening port and outgoing port.
    • The adapter parameter can be used to specify the adapter.
    • The port parameter can be used to specify the local port to bind to.

IMPORTANT information about IPv6.

This application can also use an address specified as IPv6 with the following restrictions:

  1. The IPv6 address in the URI is specified in square brackets: e.g. srt://[::1]:5000.

  2. In listener mode, if you leave the host empty, the socket is bound to INADDR_ANY for IPv4 only. If you want to make it listen on IPv6, you need to specify the host as ::. NOTE: Don't use square brackets syntax in the adapter specification, as in this case only the host is expected.

  3. If you want to listen for connections from both IPv4 and IPv6, mind the ipv6only option. The default value for this option is system default (see system manual for IPV6_V6ONLY socket option); if unsure, you might want to enforce ipv6only=0 in order to be able to accept both IPv4 and IPv6 connections in the same listener.

  4. In rendezvous mode you may only interconnect both parties using IPv4, or both using IPv6. Unlike listener mode, if you want to leave the socket default-bound (you don't specify adapter), the socket will be bound with the same IP version as the target address. If you do specify adapter, then both this address and the target address must be of the same family.

Examples:

  • srt://:5000 defines listener mode with IPv4.

  • srt://[::]:5000 defines caller mode (!) with IPv6.

  • srt://[::]:5000?mode=listener defines listener mode with IPv6. If the default value for IPV6_V6ONLY system socket option is 0, it will accept also IPv4 connections.

  • srt://192.168.0.5:5000?mode=rendezvous will make a rendezvous connection with local address INADDR_ANY (IPv4) and port 5000 to a destination with port 5000.

  • srt://[::1]:5000?mode=rendezvous&port=4000 will make a rendezvous connection with local address inaddr6_any (IPv6) and port 4000 to a destination with port 5000.

  • srt://[::1]:5000?adapter=127.0.0.1&mode=rendezvous - this URI is invalid (different IP versions for binding and target address)

Some parameters handled for SRT medium are specific, all others are socket options. The following parameters are handled special way by srt-live-transmit:

  • mode: enforce caller, listener or rendezvous mode
  • port: enforce the outgoing port (the port number that will be set in the UDP packet as a source port when sent from this host). This can be used only in caller mode.
  • blocking: sets the SRTO_RCVSYN for input medium or SRTO_SNDSYN for output medium
  • timeout: sets SRTO_RCVTIMEO for input medium or SRTO_SNDTIMEO for output medium
  • adapter: sets the adapter for listening in listener or rendezvous mode

All other parameters are SRT socket options. The following have the following value types:

  • bool. Possible values: yes/no, on/off, true/false, 1/0.
  • bytes positive integer [1; INT32_MAX].
  • ms - positive integer value of milliseconds.
URI param Values SRT Option Description
congestion {live, file} SRTO_CONGESTION Type of congestion control.
conntimeo ms SRTO_CONNTIMEO Connection timeout.
drifttracer bool SRTO_DRIFTTRACER Enable drift tracer.
enforcedencryption bool SRTO_ENFORCEDENCRYPTION Reject connection if parties set different passphrase.
fc bytes SRTO_FC Flow control window size.
groupconnect {0, 1} SRTO_GROUPCONNECT Accept group connections.
groupstabtimeo ms SRTO_GROUPSTABTIMEO Group stability timeout.
inputbw bytes SRTO_INPUTBW Input bandwidth.
iptos 0..255 SRTO_IPTOS IP socket type of service
ipttl 1..255 SRTO_IPTTL Defines IP socket "time to live" option.
ipv6only -1..1 SRTO_IPV6ONLY Allow only IPv6.
kmpreannounce 0.. SRTO_KMPREANNOUNCE Duration of Stream Encryption key switchover (in packets).
kmrefreshrate 0.. SRTO_KMREFRESHRATE Stream encryption key refresh rate (in packets).
latency 0.. SRTO_LATENCY Defines the maximum accepted transmission latency.
linger 0.. SRTO_LINGER Link linger value
lossmaxttl 0.. SRTO_LOSSMAXTTL Packet reorder tolerance.
maxbw 0.. SRTO_MAXBW Bandwidth limit in bytes
mininputbw 0.. SRTO_MININPUTBW Minimum allowed estimate of SRTO_INPUTBW
messageapi bool SRTO_MESSAGEAPI Enable SRT message mode.
minversion maj.min.rev SRTO_MINVERSION Minimum SRT library version of a peer.
mss 76.. SRTO_MSS MTU size
nakreport bool SRTO_NAKREPORT Enables/disables periodic NAK reports
oheadbw 5..100 SRTO_OHEADBW limits bandwidth overhead, percents
packetfilter string SRTO_PACKETFILTER Set up the packet filter.
passphrase string SRTO_PASSPHRASE Password for the encrypted transmission.
payloadsize 0.. SRTO_PAYLOADSIZE Maximum payload size.
pbkeylen {16, 24, 32} SRTO_PBKEYLEN Crypto key length in bytes.
peeridletimeo ms SRTO_PEERIDLETIMEO Peer idle timeout.
peerlatency ms SRTO_PEERLATENCY Minimum receiver latency to be requested by sender.
rcvbuf bytes SRTO_RCVBUF Receiver buffer size
rcvlatency ms SRTO_RCVLATENCY Receiver-side latency.
retransmitalgo {0, 1} SRTO_RETRANSMITALGO Packet retransmission algorithm to use.
sndbuf bytes SRTO_SNDBUF Sender buffer size.
snddropdelay ms SRTO_SNDDROPDELAY Sender's delay before dropping packets.
streamid string SRTO_STREAMID Stream ID (settable in caller mode only, visible on the listener peer).
tlpktdrop bool SRTO_TLPKTDROP Drop too late packets.
transtype {live, file} SRTO_TRANSTYPE Transmission type
tsbpdmode bool SRTO_TSBPDMODE Timestamp-based packet delivery mode.

The list of socket options can also be found in SRT header file srt.h (SRT_SOCKOPT enum type). Please note that the set of available options may be version dependent. All options are available under the lowercase name of the option without the SRTO_ prefix. For example, SRTO_PASSPHRASE can be set using a passphrase parameter. The mapping table srt_options can be found in common/socketoptions.hpp file.

Important thing about the options (which holds true also for options for TCP and UDP, even though it's not described anywhere explicitly) is that there are two categories of options:

  • PRE options: these options must be set to the socket prior to connecting and they cannot be altered after the connection is made. A PRE option set to a listening socket will be also derived by the socket returned by srt_accept().
  • POST options: these options can be set to a socket at any time. The option set to a listening socket will not be derived by an accepted socket.

You don't have to worry about that actually - the application is aware of this and it sets these options at appropriate time.

Note also that blocking option has no practical use for users. Normally the non-blocking mode is used only when you have an event-driven application that needs a common signal bar for multiple event sources, or you prefer fibers to threads, when working with multiple SRT sockets in one application. The srt-live-transmit application isn't defined this way. This makes that the practical result of non-blocking mode here is that it uses polling on exactly one socket with infinite timeout. Every reading and writing operation will then return always without blocking, but when they report the "again" situation the application will stall on srt_epoll_wait() call. This option then exists for the testing purposes, as well as educational, to serve as an example of how your application should use the non-blocking mode.

Command-Line Options

The following options are available in the application. Note that some may affect specifically only selected type of medium.

Options usually have values and they are set using colon: for example, -t:60. Alternatively you can also separate them by a space, but this space must be part of the parameter and not extracted by a shell (using " " quotes or backslash).

  • -timeout, -t, -to - Sets the timeout for any activity from any medium (in seconds). Default is 0 for infinite (that is, turn this mechanism off). The mechanism is such that the SIGALRM is set up to be called after the given time and it's reset after every reading succeeded. When the alarm expires due to no reading activity in defined time, it will break the application. Notes:
    • The alarm is set up after the reading loop has started, not when the application has started. That is, a caller will still wait the standard timeout to connect, and a listener may wait infinitely until some peer connects; only after the connection is established is the alarm counting started.
    • The timeout mechanism doesn't work on Windows at all. It behaves as if the timeout was set to -1 and it's not modifiable.
  • -timeout-mode, -tm - Timeout mode used. Default is 0 - timeout will happen after the specified time. Mode 1 cancels the timeout if the connection was established.
  • -st, -srctime, -sourcetime - Enable source time passthrough. Default: disabled. It is recommended to build SRT with monotonic (-DENABLE_MONOTONIC_CLOCK=ON) or C++ 11 steady (-DENABLE_STDCXX_SYNC=ON) clock to use this feature.
  • -buffering - Enable source buffering up to the specified number of packets. Default: 10. Minimum: 1 (no buffering).
  • -chunk, -c - use given size of the buffer. The default size is 1456 bytes, which is the maximum payload size for a single SRT packet.
  • -verbose, -v - Display additional information on the standard output. Note that it's not allowed to be combined with output specified as file://con.
  • -statsout - SRT statistics output: filename. Without this option specified, the statistics will be printed to the standard output.
  • -pf, -statspf - SRT statistics print format. Values: json, csv, default. After a comma, options can be specified (e.g. "json,pretty").
  • -s, -stats, -stats-report-frequency - The frequency of SRT statistics collection, based on the number of packets.
  • -loglevel - lowest logging level for SRT, one of: fatal, error, warning, note, debug (default: error)
  • -logfa, -lfa - selected FAs in SRT to be logged (default: all are enabled). See the list of FAs running -help:logging.
  • -logfile:logs.txt - Output of logs is written to file logs.txt instead of being printed to stderr.
  • -help, -h - Show help.
  • -version - Show version info.

Testing Considerations

Before starting any test with srt-live-transmit please make sure your video source works properly. For example: if you use VLC as a test player, send a UDP stream directly to it before routing it through srt-live-transmit.

For any MPEG-TS UDP based source make sure it has packet sizes of 1316 bytes. When using ffmpeg like in the "Example for Smoke Testing" section above set the pkt_size=1316 parameter in case your input is a continuous data stream like from a file, camera or data-generator.

When leaving the LAN for testing, please keep an eye on statistics and make sure your round-trip-time (RTT) is not drifting. It's recommended to set the latency 3 to 4 times higher than RTT. Especially on wireless links such as WLAN, Line-of-Sight Radio (LOS) and mobile links such as LTE/4G or 5G the RTT can vary a lot.

If you perform tests on the public Internet, consider checking your firewall rules. The SRT listener must be reachable on the chosen UDP port. Same applies to routers using NAT. Please set a port forwarding rule with protocol UDP to the local IP address of the SRT listener.

The initiation of an SRT connection (handshake) is decoupled from the stream direction. The sender of a stream can be an SRT listener or an SRT caller, as long as the receiving end uses the opposite connection mode. Typically you use the SRT listener on the receiving end, since it is easier to configure in terms of firewall/router setup. It also makes sense to leave the Sender in listener mode when trying to connect from various end points with possibly unknown IP addresses.

UDP Performance

Performance issues concerning reading from UDP medium were reported in #933 and #762.

The dedicated research showed that at high and bursty data rates (~60 Mbps) the epoll_wait(udp_socket) is not fast enough to signal about the possibility of reading from a socket. It results in losing data when the input bitrate is very high (above 20 Mbps).

PR #1152 (v1.5.0 and above) adds the possibility of setting the buffer size of the UDP socket in srt-live-transmit. Having a bigger buffer of UDP socket to store incoming data, srt-live-transmit handles higher bitrates.

The following steps have to be performed to use the bigger UDP buffer size.

Increase the system-default max rcv buffer size

$ cat /proc/sys/net/core/rmem_max
212992
$ sudo sysctl -w net.core.rmem_max=26214400
net.core.rmem_max = 26214400
$ cat /proc/sys/net/core/rmem_max
26214400

Specify the size of the UDP socket buffer via the URI

Example URI:

"udp://:4200?rcvbuf=67108864"

Example full URI:

./srt-live-transmit "udp://:4200?rcvbuf=67108864" srt://192.168.0.10:4200 -v