Qwen Bailian Trace A

To-C chat-style interactive services. Source: qwen_traceA_blksz_16.jsonl. Timestamps are trace-relative seconds; no absolute wall-clock times or request durations are included.

Requests

43,058

Duration

2.000 h

Mean RPS

5.981

Peak 1s RPS

18.00

Input p50/p99

1,046.00 / 14,364.59

Input min/max

56 / 89,286

Output p50/p99

375.50 / 1,641.43

Sessions approx.

34,046

Hash reuse

59.8%

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	2,331.34	1,046.00	6,436.00	8,808.15	14,364.59	23,497.35
Output tokens	430.47	375.50	808.00	1,032.00	1,641.43	3,095.72
Total tokens	2,761.81	1,422.00	7,049.00	9,459.30	15,204.16	25,002.01
Turn	2.177	1.000	4.000	6.000	11.00	23.00
Session requests	1.265	1.000	2.000	2.000	2.000	2.000
Hash blocks/request	146.17	66.00	403.00	551.00	898.43	1,468.77

Arrival RPS by Window

Window	Mean	p50	p90	p95	p99	Max
1s	5.980	6.000	10.000	11.000	13.000	18.000
10s	5.980	6.100	7.900	8.400	9.081	9.800
60s	5.980	6.108	7.600	7.902	8.068	8.133
5min	5.980	5.970	7.542	7.642	7.739	7.767
10min	5.980	6.062	7.435	7.566	7.678	7.707

Request Types

Trace A is mostly text. Text-only load is 4.409 RPS versus 5.981 RPS overall; search/image/file account for the remaining arrival load.

Type	Requests	Fraction	Mean RPS
text	31,744	73.72%	4.409
search	8,187	19.01%	1.137
image	1,617	3.76%	0.225
file	1,510	3.51%	0.210

Hash Blocks

Metric	Value
Total hash blocks	6,293,976
Unique hash blocks	2,530,337
Global hash reuse ratio	59.80%
Requests with any previously seen hash	99.95%