Qwen Bailian Trace B

To-B API-driven task automation. Source: qwen_traceB_blksz_16.jsonl. Timestamps are trace-relative seconds; no absolute wall-clock times or request durations are included.

Requests

172,800

Duration

2.000 h

Mean RPS

24.000

Peak 1s RPS

52.00

Input p50/p99

574.00 / 6,294.01

Input min/max

19 / 66,446

Output p50/p99

39.00 / 1,006.02

Sessions approx.

172,800

Hash reuse

61.5%

Concurrency note: this trace gives request arrival timestamps and token lengths, but not service durations. The RPS plots describe arrival load; actual in-flight concurrency depends on the serving system latency during replay.

Arrival Rate

Token Lengths

Structure

Statistics

Metric	Mean	p50	p90	p95	p99	p99.9
Input tokens	914.52	574.00	1,720.00	2,491.00	6,294.01	14,201.81
Output tokens	91.63	39.00	180.00	328.00	1,006.02	2,153.61
Total tokens	1,006.15	667.00	1,897.00	2,700.00	6,894.00	14,750.87
Turn	1.000	1.000	1.000	1.000	1.000	1.000
Session requests	1.000	1.000	1.000	1.000	1.000	1.000
Hash blocks/request	57.62	36.00	108.00	156.00	394.00	888.20

Arrival RPS by Window

Window	Mean	p50	p90	p95	p99	Max
1s	24.000	24.000	33.000	36.000	41.000	52.000
10s	24.000	23.700	30.200	31.600	33.881	37.000
60s	24.000	23.583	29.557	31.128	32.278	32.683
5min	24.000	23.642	27.573	28.748	29.997	30.330
10min	24.000	24.456	26.131	26.461	26.723	26.788

Request Types

Trace B aggregate RPS is dominated by api requests. Text-only load is 3.037 RPS, while api requests contribute 20.963 RPS of the 24.000 RPS overall arrival rate.

Type	Requests	Fraction	Mean RPS
api	150,936	87.35%	20.963
text	21,864	12.65%	3.037

Hash Blocks

Metric	Value
Total hash blocks	9,956,137
Unique hash blocks	3,830,390
Global hash reuse ratio	61.53%
Requests with any previously seen hash	99.52%