HTTP/3 & QUIC Tuning: Essential Parameters [Cheat Sheet]

As we transition to an HTTP/3 dominant web, the underlying QUIC protocol introduces a new paradigm of network tuning. Unlike TCP, which is handled by the kernel, QUIC is implemented in user-space, giving developers unprecedented control over congestion algorithms, flow control, and packet acknowledgment strategies. For high-latency links—such as satellite connections or trans-continental fiber—default settings often result in throughput collapse.

The High-Latency Rule

On links with RTT > 100ms, BBR (Bottleneck Bandwidth and Round-trip propagation time) is mandatory. Standard CUBIC congestion control treats packet loss as a congestion signal, which is fatal on lossy, high-latency wireless links where QUIC is intended to shine.

Live Search: QUIC Parameters

Use the filter below to find specific parameters for your configuration.

Parameter	Recommended Value	Impact
maxidletimeout	60000 (ms)	Prevents premature disconnection on flaky links.
initialmaxdata	10485760 (10MB)	Initial connection-level flow control limit.
initialmaxstreamdatabidi_remote	5242880 (5MB)	Avoids stalling the first large response.
activeconnectionid_limit	8	Enables better connection migration during IP handovers.

Congestion Control & BBR Tuning

For high-latency environments, the choice of congestion controller is the single most significant factor. BBRv2 and BBRv3 significantly outperform CUBIC by measuring the actual bottleneck bandwidth rather than reacting to packet loss.

# Linux Kernel optimization for QUIC/UDP throughput
sysctl -w net.core.rmemmax=2500000
sysctl -w net.core.wmemmax=2500000
sysctl -w net.ipv4.tcpcongestioncontrol=bbr

Flow Control & Window Limits

QUIC uses multi-layered flow control. If your Bandwidth-Delay Product (BDP) is high, you must increase these limits to prevent the sender from stopping while waiting for ACKs.

Connection Flow Control: The total amount of data that can be sent on the entire connection before an update is received.
Stream Flow Control: Limits per individual request/response stream.

When analyzing logs for performance bottlenecks, ensure sensitive connection IDs or IP addresses are anonymized. You can use our Data Masking Tool to sanitize qlog or pcap data before sharing with external vendors.

Implementation Examples

Nginx (quic-go / OpenSSL 3.0+)

http {
    quicbpf on;
    http3 on;
    http3maxfieldsize 16k;
    
    server {
        listen 443 quic reuseport;
        listen 443 ssl;
        
        # QUIC specific tuning
        quicretry on;
        quicgso on;
    }
}

Go (quic-go)

conf := &quic.Config{
    MaxIdleTimeout:  time.Minute,
    InitialStreamReceiveWindow:     5  1024  1024, // 5MB
    InitialConnectionReceiveWindow: 10  1024  1024, // 10MB
    EnableDatagrams: true,
}