docs: update trace replay tutorial with real mooncake data

docs/benchmark_modes/trace_replay.md

@@ -11,6 +11,8 @@ This tutorial takes you through an example trace replay profile. Trace Replay be
 
 - [MoonCake Traces](#mooncake-traces)
 - [Profiling using a MoonCake Trace](#profiling-using-a-mooncake-trace)
+- [Real Mooncake Trace Example](#real-mooncake-trace-example)
+  - [Download and Benchmark Mooncake Trace](#download-and-benchmark-mooncake-trace)
 
 ## MoonCake Traces
 
@@ -50,7 +52,29 @@ echo \
 aiperf profile \
     -m deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B \
 	--input-file example_trace.jsonl \
-	--custom-dataset-type mooncake_trace
+	--custom-dataset-type mooncake_trace \
+    --fixed-schedule
 ```
 
-The code above will create an example trace file formatted in jsonl. AIPerf will use it to define the dataset and timing to replay the trace.
+The code above will create an example trace file formatted in jsonl. AIPerf will use it to define the dataset and timing to replay the trace.
+
+## Real Mooncake Trace Example
+
+For real-world benchmarking, you can use the actual FAST25 production trace data from the Mooncake research paper. This trace contains realistic request patterns from production workloads.
+
+### Download and Benchmark Mooncake Trace
+
+```bash
+# Download the Mooncake trace data
+curl -o mooncake_trace.jsonl https://raw.githubusercontent.com/kvcache-ai/Mooncake/refs/heads/main/FAST25-release/arxiv-trace/mooncake_trace.jsonl
+
+# Create a subset of the file for quick testing (skip if you want to run the full dataset)
+head -n 5 mooncake_trace.jsonl > mooncake_trace_short.jsonl
+
+# Run a small portion of the trace dataset
+aiperf profile \
+    -m deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B \
+    --input-file mooncake_trace_short.jsonl \
+    --custom-dataset-type mooncake_trace \
+    --fixed-schedule
+```