fixing plotting script for time series

Signed-off-by: Le Xu <[email protected]>

Author: happyandslow

benchmarks/autoscaling/plot-everything.py

@@ -28,7 +28,7 @@ def parse_experiment_output(lines):
             # required_fields = ['status_code', 'start_time', 'end_time', 'latency', 'throughput', 
             #                  'prompt_tokens', 'output_tokens', 'total_tokens', 'input', 'output']
             required_fields = ['status', 'start_time', 'end_time', 'latency', 'throughput', 
-                             'prompt_tokens', 'output_tokens', 'total_tokens', 'input', 'output']
+                             'prompt_tokens', 'output_tokens', 'total_tokens', 'input', 'output', 'ttft', 'tpot']
             if any(field not in data for field in required_fields):
                 missingfields = [field not in data for field in required_fields]
                 print(missingfields)
@@ -178,11 +178,11 @@ def analyze_performance(df):
 
 def plot_combined_visualization(experiment_home_dir, workload_type):
     # Create figure
-    fig = plt.figure(figsize=(12, 12))
+    fig = plt.figure(figsize=(12, 15))
 
     # Create GridSpec for time series plots
-    gs_ts = GridSpec(2, 2, figure=fig)
-    gs_ts.update(top=0.95, bottom=0.45, hspace=0.3, wspace=0.25)
+    gs_ts = GridSpec(4, 2, figure=fig)
+    gs_ts.update(top=0.95, bottom=0.4, hspace=0.5, wspace=0.3)
 
     # Create subplots for time series
     ax_cdf = fig.add_subplot(gs_ts[0, 0])
@@ -196,6 +196,11 @@ def plot_combined_visualization(experiment_home_dir, workload_type):
 
     ax_cost = fig.add_subplot(gs_ts[1, 1])
 
+    ax_ttft = fig.add_subplot(gs_ts[2, 0])
+    ax_tpot = fig.add_subplot(gs_ts[2, 1])
+
+    ax_throughput = fig.add_subplot(gs_ts[3, 0])
+    
     # Create 1x4 grid for bar plots
     gs_bars = GridSpec(1, 4, figure=fig)
     gs_bars.update(top=0.35, bottom=0.1, left=0.1, right=0.9, wspace=0.4)
@@ -308,7 +313,11 @@ def plot_combined_visualization(experiment_home_dir, workload_type):
         # ax_latency.scatter(df['end_time'], df['latency'], label=f'{autoscaler}', color=color, marker='.', s=5, alpha=0.3)
         ax_rps.plot(df['start_time'], df['rps'], label=label_name, color=color, linewidth=0.5, alpha=0.7)
         print(f"Max time: {label_name}, {df['start_time'].max()}")
-        
+        ax_ttft.plot(df['start_time'], df['ttft'], label=label_name, color=color, linewidth=0.5, alpha=0.7)
+        ax_tpot.plot(df['start_time'], df['tpot'], label=label_name, color=color, linewidth=0.5, alpha=0.7)
+        ax_throughput.plot(df['start_time'], df['throughput'], label=label_name, color=color, linewidth=0.5, alpha=0.7)
+
+
 
         # 3. Running pods over time
         ax_pods.plot(pod_count_df['asyncio_time'], pod_count_df['Running'],
@@ -345,6 +354,21 @@ def plot_combined_visualization(experiment_home_dir, workload_type):
     ax_rps.grid(True)
     ax_rps.legend()
 
+    ax_ttft.set_title("TTFT Over Time")
+    ax_ttft.set_xlabel("Time (s)")
+    ax_ttft.set_ylabel("TTFT (s)")
+    ax_ttft.legend()
+
+    ax_tpot.set_title("TPOT Over Time")
+    ax_tpot.set_xlabel("Time (s)")
+    ax_tpot.set_ylabel("TPOT (s)")
+    ax_tpot.legend()
+
+    ax_throughput.set_title("Throughput Over Time")
+    ax_throughput.set_xlabel("Time (s)")
+    ax_throughput.set_ylabel("Tokens/sec")
+    ax_throughput.legend()
+    
     ax_pods.set_title('Running Pods', fontsize=12)
     ax_pods.set_xlabel('Time (seconds)')
     ax_pods.set_ylabel('# Running Pods')