HTTP/2 Multiplex Streams Calculator
Estimates how many HTTP/2 streams fit per window from MTU and average frame size.
β
HTTP/2 Multiplexing & Stream Calculator
HTTP/2 runs many independent streams over one TCP connection, which clears up the head-of-line blocking that sat at the HTTP layer in HTTP/1.1 (where pipelining mostly fell out of use). Every stream carries one request/response pair as a run of binary DATA and HEADERS frames, tagged with an odd-numbered stream ID. A peer announces how many streams it will take through the SETTINGS_MAX_CONCURRENT_STREAMS parameter. RFC 7540 asks for at least 100, and most servers (NGINX, Apache, h2o) ship with a default of 100.
Headers get compressed too, with HPACK (RFC 7541), a stateful scheme built on Huffman coding plus a dynamic table. On repeated headers like user-agent and cookie it usually shaves off 80β90% of the size. Server push arrived as a way to send resources ahead of a request, but Chrome dropped it in 2022 once it became clear the cache-hit payoff was small. The protocol landed as RFC 7540 (May 2015), and every major browser and CDN supports it.
Applications
Reach for this calculator when you need to size connection pools, set max_concurrent_streams on a reverse proxy, work out frame throughput per MTU, or figure out how many parallel sub-requests one h2 connection can handle before it saturates the TCP window. You'll hit these cases with REST API gateways, gRPC servers (which require HTTP/2), SPA asset delivery, and microservice meshes.
FAQ
Does HTTP/2 fully eliminate head-of-line blocking? No, it only handles it at the HTTP layer. Drop a single TCP segment and every stream on that connection stalls, because TCP hands bytes over in order. That's the reason HTTP/3 switched to QUIC over UDP.
Why is the default 100 streams and not unlimited? Every open stream costs memory for HPACK tables, flow-control counters, and priority state. Capping it at 100 keeps a reasonable amount of concurrency while holding back DoS attempts. Without that limit an attacker could open thousands of streams and burn through server memory (CVE-2023-44487 "Rapid Reset").
Should I still use HTTP/2 in 2026? Yes, wherever HTTP/3 isn't an option. For browser traffic, lean on h3 with an h2 fallback advertised through Alt-Svc. For internal gRPC, h2 over TLS is still the standard.
Related Tools
HTTP/3 QUIC Overhead Calculator
Computes useful HTTP/3 payload over QUIC, subtracting UDP, QUIC and TLS overhead.
MTU Calculator by Tunnel Encapsulation
Subtracts tunnel overhead from a base MTU of 1500 bytes: PPPoE 8, GRE 24, VXLAN 50, IPsec 56. Returns the effective size to set on the link.
MSS from MTU: TCP Maximum Segment Size
Subtracts the IP and TCP headers from a link MTU, 40 bytes on IPv4 and 60 on IPv6, so 1500 leaves 1460 and 1440 bytes of payload per segment.
TCP MSS from MTU Calculator
Estimates TCP MSS from link MTU, subtracting IP and TCP headers.
VXLAN Overhead Calculator
Computes effective payload MTU on VXLAN by subtracting outer Ethernet, IP, UDP and VXLAN headers.
3D Dot Product Calculator
Calculate the dot product (scalar product) of two 3D vectors from their x, y and z coordinates. Useful in physics, computer graphics and vector geometry.
The results provided by this tool are for general informational and educational purposes only and do not constitute professional, financial, medical, legal, tax or accounting advice. Always confirm important decisions with a qualified professional and official sources.