GitHub1
SECTION 6.0 // CONCURRENCY SATURATION DRAG RACE

Framework Concurrency Duel

Head-to-head architectural saturation comparison. Compare any two repositories side-by-side to discover which codebase chokes first under concurrent load.

SIMULTANEOUS LOAD INJECTION HAMMERSimulating concurrent HTTP clients against both architectures
350 CONCURRENT CLIENTS
50 conns500 (Moderate)1,500 (Heavy)2,500 (Spike)4,000 (Saturation)
[HEADROOM CHAMPION]
msh-protocol/msh holds +29.4x more concurrent load than tkngate/tkngate before socket exhaustion.
CONTENDER A

tkngate/tkngate

Go Zero-Trust Gateway / Proxy
GRADED
SATURATION CEILING: 85 CONCURRENT[COLLAPSED: 429/504]
EST. P99 LATENCY:4174 ms
BOTTLENECK SYMBOL:proxy.NewProxy (http.DefaultTransport)
PRIMARY WEAKNESS:http.DefaultTransport has MaxIdleConnsPerHost=2, causing TCP socket churn & TLS renegotiation under proxy load
VS
CONTENDER B

msh-protocol/msh

Go Swarm Telemetry & Fleet Mesh
GRADEA
SATURATION CEILING: 2500 CONCURRENT[NOMINAL]
EST. P99 LATENCY:27 ms
BOTTLENECK SYMBOL:fleet.Server.handleRegister
PRIMARY WEAKNESS:Synchronous socket write loop across peer connections exhausts epoll buffer under 2,500 concurrent peers