Understanding the roles of latency and buffers in SRT

[MalikAhmed2]

Hi,
I have a question about how SRT latency interacts with the sender and receiver buffers.

According to The SRT Protocol:
https://haivision.github.io/srt-rfc/draft-sharabayko-srt.html

"On the sender, latency is the time that SRT holds a packet to give it a chance to be delivered successfully while maintaining the rate of the sender at the receiver. If an acknowledgment (ACK) is missing or late for more than the configured latency, the packet is dropped from the sender buffer. A packet can be retransmitted as long as it remains in the buffer for the duration of the latency window. On the receiver, packets are delivered to an application from a buffer after the latency interval has passed."

Given this description, what is the purpose of configuring the sender and receiver buffer sizes? If a packet is dropped once the configured latency has expired, how can larger sender or receiver buffers provide any additional benefit?

[cl-ment]

The latency setting determines how long a packet remains eligible for retransmission.

For example, if the latency is 120 ms:

A packet transmitted at t = 0 ms can be retransmitted until t = 120 ms.
After that, SRT assumes it is too late to meet the playout deadline and removes the packet from the retransmission buffer.

The sender buffer has one primary purpose: storing packets that may need to be retransmitted (ARQ). It should be sized to hold all data that may require retransmission based on the configured latency and the stream bitrate.

The receiver buffer serves a different purpose. Incoming packets are:

• Stored
• Reordered
• Repaired through retransmissions
• Held until their scheduled delivery time

Simply increasing the sender or receiver buffer size does not extend the retransmission window. If the latency remains 120 ms, packets older than 120 ms are no longer eligible for retransmission, regardless of how much buffer memory is available.

However, the buffers must still be large enough to hold all packets that can exist during the configured latency interval. Otherwise, packets may be dropped because the buffer becomes full before the latency timer expires.

As a rule of thumb:

buffer_size ≥ (latency + RTT) × bitrate / 8

where:

• Latency and RTT are expressed in seconds.
• Bitrate is expressed in bits per second.
• Dividing by 8 converts bits to bytes.

[MalikAhmed2]

Thanks for the clarification.

I have a couple more questions:

• Should the sender and receiver buffers be the same size, or is it recommended to configure them differently?

• What is the difference between RCVLATENCY and PEERLATENCY?

• At what point does the receiver deliver packets to the decoder? Is each packet released after its configured latency has elapsed, or does the receiver wait until the entire receive buffer is filled before passing packets to the decoder?