L4S (for Low Latency, Low Loss and Scalable Throughput) is an IETF network protocol and congestion control technology designed to lower network latency [1] by reducing bufferbloat throughout the Internet.
L4S uses novel congestion control mechanisms to reduce queuing in the network. [1] It uses Explicit Congestion Notification (ECN) to transmit information about path latency problems, and allows congested nodes to use the ECN bits to send information back to senders that will allow them to adjust their transmit rate, reducing the need for data buffering within router queues.
L4S has the advantage that it is an incremental technology which can start to provide incremental latency improvements without having to be adopted throughout the entire Internet. [2]
L4S is specified in RFC 9330. It uses the last codepoint of the Internet Protocol header's ECN field that had not previously been assigned to signal that traffic is from an L4S-capable sender. [3] The full set of four ECN codes for packets are thus: [4]
Code | Meaning |
---|---|
00 | Non-ECN |
01 | L4S ECN |
10 | Classic ECN |
11 | Marked as congested |
Routers can thus treat L4S traffic differently from non-L4S traffic, knowing that L4S endpoints will respond to throttle back traffic in a more controlled way than would be possible using classic ECN. This is done by treating L4S traffic differently for both the cases of queuing and marking. [5]
As of January 2025 [update] , Internet service providers had started to roll out L4S in their production networks, with Comcast being an early adopter. [6] Apple have incorporated L4S support in their newer operating systems since 2023. [7] Linux support for L4S, in the form of TCP Prague, is available on an experimental basis, and is expected to be merged into the main Linux kernel tree soon. [5]
As of July 21,2025 [update] T-Mobile announced support for L4S at a network level. [8]