[Misc] Update disaggregation benchmark scripts and test logs (#11456)

Signed-off-by: Jiaxin Shan <seedjeffwan@gmail.com>

benchmarks/disagg_benchmarks/disagg_overhead_benchmark.sh

@@ -10,7 +10,8 @@ set -ex
 
 kill_gpu_processes() {
   # kill all processes on GPU.
-  pkill -f pt_main_thread
+  pgrep pt_main_thread | xargs -r kill -9
+  pgrep python3 | xargs -r kill -9
   sleep 10
 
   # remove vllm config file
@@ -54,7 +55,7 @@ benchmark() {
 
   CUDA_VISIBLE_DEVICES=0 python3 \
     -m vllm.entrypoints.openai.api_server \
-    --model meta-llama/Meta-Llama-3.1-8B-Instruct \
+    --model $model \
     --port 8100 \
     --max-model-len 10000 \
     --gpu-memory-utilization 0.6 \
@@ -64,7 +65,7 @@ benchmark() {
 
   CUDA_VISIBLE_DEVICES=1 python3 \
     -m vllm.entrypoints.openai.api_server \
-    --model meta-llama/Meta-Llama-3.1-8B-Instruct \
+    --model $model \
     --port 8200 \
     --max-model-len 10000 \
     --gpu-memory-utilization 0.6 \
@@ -87,7 +88,7 @@ benchmark() {
           --port 8100 \
           --save-result \
           --result-dir $results_folder \
-          --result-filename disagg_prefill_2xtp4.json \
+          --result-filename disagg_prefill_tp1.json \
           --request-rate "inf"
 
 
@@ -105,7 +106,7 @@ benchmark() {
           --port 8200 \
           --save-result \
           --result-dir $results_folder \
-          --result-filename disagg_prefill_2xtp4.json \
+          --result-filename disagg_prefill_tp1_overhead.json \
           --request-rate "$qps"
   kill_gpu_processes
 
@@ -118,7 +119,7 @@ main() {
   (which jq) || (apt-get -y install jq)
   (which socat) || (apt-get -y install socat)
 
-  pip install quart httpx
+  pip install quart httpx datasets
 
   cd "$(dirname "$0")"
 

benchmarks/disagg_benchmarks/disagg_performance_benchmark.sh

@@ -1,13 +1,12 @@
 #!/bin/bash
 
-# Requirement: 8x H100 GPUs.
+# Requirement: 2x GPUs.
 
 
-# Model: neuralmagic/Meta-Llama-3-70B-Instruct-FP8-KV 
-# Query: 2048 input tokens, 11 output tokens, QPS 4, 500 requests
-# Resource: 8x H100
+# Model: meta-llama/Meta-Llama-3.1-8B-Instruct
+# Query: 1024 input tokens, 6 output tokens, QPS 2/4/6/8, 100 requests
+# Resource: 2x GPU
 # Approaches:
-# 1. Chunked prefill: 1 vllm instance with tp=8
 # 2. Chunked prefill: 2 vllm instance with tp=4, equivalent to 1 tp=4 instance with QPS 4
 # 3. Disaggregated prefill: 1 prefilling instance and 1 decoding instance
 # Prefilling instance: max_output_token=1
@@ -114,7 +113,6 @@ benchmark() {
           --request-rate "$qps"
 
   sleep 2
-
 }
 
 
@@ -123,8 +121,9 @@ main() {
   (which wget && which curl) || (apt-get update && apt-get install -y wget curl)
   (which jq) || (apt-get -y install jq)
   (which socat) || (apt-get -y install socat)
+  (which lsof) || (apt-get -y install lsof)
 
-  pip install quart httpx matplotlib aiohttp
+  pip install quart httpx matplotlib aiohttp datasets
 
   cd "$(dirname "$0")"