AISIX (api7) vs Busbar

Operated by the Busbar team; Busbar is also an entrant. Gateway overhead against a local mock is not end-to-end model latency. Methodology and limitations.

Requested cell: OpenAI → OpenAI. Metric: Added latency p99 (µs).

Comparability: Comparable. Ratios require an identical protocol pair, hardware, architecture, metric definition, clean harness commit, recorded mock digest, and metric-specific workload. Missing evidence never matches missing evidence. These are observed gateway measurements against a local mock, not a claim about model latency or statistical significance.

Ratio (Busbar ÷ AISIX (api7)): 0.511. No statistical significance is asserted.

Comparison JSON · All AISIX (api7) evidence · All Busbar evidence

AISIX (api7)

Build ghcr.io/api7/aisix:1.9.0. Measured . Run evidence

Ingress → upstreamMetricValueState / reasonp99 bound (ms)
OpenAI → OpenAIAdded latency p50 (µs)146Valid · n=35,048 gateway / 262,627 direct—
OpenAI → OpenAIAdded latency p99 (µs)174Valid · n=35,048 gateway / 262,627 direct—
OpenAI → OpenAIGateway p99 at concurrency 1 (µs)202Valid · n=35,048—
OpenAI → OpenAIDirect-to-mock p99 at concurrency 1 (µs)28Valid · n=262,627—
OpenAI → OpenAICPU per request (µs)152.6Valid—
OpenAI → OpenAIRequests per CPU-second6,551.3Valid—
OpenAI → OpenAICost window concurrency8Valid—
OpenAI → OpenAICost window core utilisation0.945Valid—
OpenAI → OpenAICost window successful requests148,977Valid—
OpenAI → OpenAICost window throughput (req/s)24,826Valid—
OpenAI → OpenAICost window threads16Valid—
OpenAI → OpenAINon-voluntary context switches per request0Valid—
OpenAI → OpenAICost window major page faults0Valid—
OpenAI → OpenAIAdded time to first event p50 (µs)277Valid · n=100 gateway / 100 direct—
OpenAI → OpenAIAdded time to first event p99 (µs)366Valid · n=100 gateway / 100 direct—
OpenAI → OpenAIAdded inter-frame gap p50 (µs)2Valid · n=63 gateway / 63 direct—
OpenAI → OpenAIAdded inter-frame gap p99 (µs)—Below resolution · n=63 gateway / 63 direct—
OpenAI → OpenAIStreams sustained (frames/s)30,259.2Valid—
OpenAI → OpenAIStreams sustained645Valid—
OpenAI → OpenAIIdle RSS (MiB)34.29Valid—
OpenAI → OpenAISteady-state RSS (MiB)68.63Valid—
OpenAI → OpenAIRecovered RSS (MiB)52.3Valid—
OpenAI → OpenAIPeak RSS (MiB)70.43Valid—
OpenAI → OpenAIPeak RSS high-water mark (MiB)70.29Valid—
OpenAI → OpenAITime to plateau (s)—Not measured—
OpenAI → OpenAIMemory growth (MiB/min)0.376Valid—
OpenAI → OpenAIMemory load duration (s)300Valid—
OpenAI → OpenAIMemory shape under load0Valid—
OpenAI → OpenAIIdle memory shape-1Valid—
OpenAI → OpenAIIdle memory static0Valid—
OpenAI → OpenAIIdle memory growth (MiB/min)-0.853Valid—
OpenAI → OpenAIThroughput (req/s)28,888concurrency 16; n=173,3461 ms
OpenAI → OpenAIThroughput (req/s)31,374concurrency 64; n=188,2545 ms
OpenAI → OpenAIThroughput (req/s)32,092concurrency 128; n=192,58310 ms
OpenAI → OpenAIThroughput (req/s)32,092concurrency 128; n=192,58350 ms
OpenAI → OpenAIThroughput (req/s)32,092concurrency 128; n=192,583100 ms
OpenAI → OpenAIThroughput (req/s)32,092concurrency 128; n=192,583Unbounded

Busbar

Build getbusbar/busbar:1.5.5. Measured . Run evidence

Ingress → upstreamMetricValueState / reasonp99 bound (ms)
OpenAI → OpenAIAdded latency p50 (µs)78Valid · n=59,266 gateway / 266,330 direct—
OpenAI → OpenAIAdded latency p99 (µs)89Valid · n=59,266 gateway / 266,330 direct—
OpenAI → OpenAIGateway p99 at concurrency 1 (µs)117Valid · n=59,266—
OpenAI → OpenAIDirect-to-mock p99 at concurrency 1 (µs)28Valid · n=266,330—
OpenAI → OpenAICPU per request (µs)68.74Valid—
OpenAI → OpenAIRequests per CPU-second14,547.2Valid—
OpenAI → OpenAICost window concurrency8Valid—
OpenAI → OpenAICost window core utilisation0.957Valid—
OpenAI → OpenAICost window successful requests335,167Valid—
OpenAI → OpenAICost window throughput (req/s)55,854Valid—
OpenAI → OpenAICost window threads10Valid—
OpenAI → OpenAINon-voluntary context switches per request0Valid—
OpenAI → OpenAICost window major page faults0Valid—
OpenAI → OpenAIAdded time to first event p50 (µs)118Valid · n=100 gateway / 100 direct—
OpenAI → OpenAIAdded time to first event p99 (µs)254Valid · n=100 gateway / 100 direct—
OpenAI → OpenAIAdded inter-frame gap p50 (µs)—Below resolution · n=63 gateway / 63 direct—
OpenAI → OpenAIAdded inter-frame gap p99 (µs)—Below resolution · n=63 gateway / 63 direct—
OpenAI → OpenAIStreams sustained (frames/s)39,240.2Valid—
OpenAI → OpenAIStreams sustained1,452Valid—
OpenAI → OpenAIIdle RSS (MiB)7.43Valid—
OpenAI → OpenAISteady-state RSS (MiB)21.49Valid—
OpenAI → OpenAIRecovered RSS (MiB)15.7Valid—
OpenAI → OpenAIPeak RSS (MiB)22.35Valid—
OpenAI → OpenAIPeak RSS high-water mark (MiB)22.27Valid—
OpenAI → OpenAITime to plateau (s)—Not measured—
OpenAI → OpenAIMemory growth (MiB/min)0.391Valid—
OpenAI → OpenAIMemory load duration (s)300Valid—
OpenAI → OpenAIMemory shape under load1Valid—
OpenAI → OpenAIIdle memory shape0Valid—
OpenAI → OpenAIIdle memory static0Valid—
OpenAI → OpenAIIdle memory growth (MiB/min)0.004Valid—
OpenAI → OpenAIThroughput (req/s)62,231concurrency 32; n=373,4171 ms
OpenAI → OpenAIThroughput (req/s)62,587concurrency 128; n=375,5785 ms
OpenAI → OpenAIThroughput (req/s)62,587concurrency 128; n=375,57810 ms
OpenAI → OpenAIThroughput (req/s)62,587concurrency 128; n=375,57850 ms
OpenAI → OpenAIThroughput (req/s)62,587concurrency 128; n=375,578100 ms
OpenAI → OpenAIThroughput (req/s)62,587concurrency 128; n=375,578Unbounded