diff --git a/benchmarks/object_storage_server.py b/benchmarks/object_storage_server.py
new file mode 100644
index 000000000..efbff4199
--- /dev/null
+++ b/benchmarks/object_storage_server.py
@@ -0,0 +1,165 @@
+import os
+import time
+from typing import Callable
+
+import matplotlib.patches as mpatches
+import matplotlib.pyplot as plt
+import numpy as np
+from matplotlib.colors import Normalize
+
+from scaler.client.client import Client
+from scaler.cluster.combo import SchedulerClusterCombo
+
+
+def plot_delta_with_errors(delta, errors, n_objects, object_size, out_path):
+    """
+    delta: 2D numpy array (shape: len(n_objects) x len(object_size))
+    errors: boolean 2D array same shape, True where a run failed
+    n_objects: 1D array of row labels
+    object_size: 1D array of column labels (in bytes)
+    out_path: filename to save the PNG
+    """
+    h, w = delta.shape
+    rgba = np.ones((h, w, 4), dtype=float)  # default white
+
+    # Masks
+    pos_mask = (delta > 0) & (~errors)
+    neg_mask = (delta < 0) & (~errors)
+    zero_mask = (delta == 0) & (~errors)
+    err_mask = errors
+
+    # --- Color computation ---
+    if np.any(pos_mask):
+        max_pos = float(np.nanmax(delta[pos_mask])) or 1.0
+        norm_pos = Normalize(vmin=0.0, vmax=max_pos, clip=True)
+        rgba[pos_mask] = plt.get_cmap("Reds")(norm_pos(delta[pos_mask]))
+
+    if np.any(neg_mask):
+        min_neg = float(np.nanmin(delta[neg_mask])) or -1.0
+        norm_neg = Normalize(vmin=min_neg, vmax=0.0, clip=True)
+        rgba[neg_mask] = plt.get_cmap("Greens_r")(norm_neg(delta[neg_mask]))
+
+    rgba[zero_mask] = (1.0, 1.0, 1.0, 1.0)
+    rgba[err_mask] = (0.6, 0.6, 0.6, 1.0)
+
+    # --- Plot setup ---
+    fig, ax = plt.subplots(figsize=(10, 6))
+    ax.imshow(rgba, origin="lower", extent=[0, w, 0, h], interpolation="nearest")
+
+    # Grid lines
+    for x in range(w + 1):
+        ax.axvline(x, color="black", linewidth=0.5)
+    for y in range(h + 1):
+        ax.axhline(y, color="black", linewidth=0.5)
+
+    # Ticks
+    ax.set_xticks(np.arange(w) + 0.5)
+    ax.set_xticklabels(
+        [f"{s//1024} KB" if s < 1024**2 else f"{s//1024**2} MB" for s in object_size], rotation=45, ha="right"
+    )
+    ax.set_yticks(np.arange(h) + 0.5)
+    ax.set_yticklabels(n_objects)
+    ax.set_xlabel("Object Size")
+    ax.set_ylabel("Number of Objects")
+    ax.set_title("Object Storage Server Performance, ZMQ+TCP vs YMQ backends")
+
+    # --- Write cell values ---
+    for i in range(h):
+        for j in range(w):
+            if errors[i, j]:
+                text = "ERR"
+                color = "black"
+            else:
+                val = delta[i, j]
+                if np.isnan(val):
+                    text, color = "", "black"
+                else:
+                    text = f"{val:.2f}"
+                    color = "black" if abs(val) < 0.5 else "white"  # contrast rule
+            ax.text(j + 0.5, i + 0.5, text, ha="center", va="center", fontsize=8, color=color, fontweight="bold")
+
+    # --- Legend ---
+    red_patch = mpatches.Patch(color=plt.get_cmap("Reds")(0.6), label="Δ > 0 (ymq is slower)")
+    green_patch = mpatches.Patch(color=plt.get_cmap("Greens")(0.6), label="Δ < 0 (ymq is faster)")
+    err_patch = mpatches.Patch(color=(0.6, 0.6, 0.6), label="Error (gray)")
+    ax.legend(
+        handles=[red_patch, green_patch, err_patch],
+        bbox_to_anchor=(1.02, 1),
+        loc="upper left",
+        title="Delta (seconds)",
+        title_fontsize=11,
+        fontsize=9,
+    )
+
+    plt.tight_layout()
+    plt.savefig(out_path, dpi=200)
+    plt.close(fig)
+
+
+def timed_execution(test: Callable[[Client], None], port: int, *args, **kwargs) -> float:
+    address = f"tcp://127.0.0.1:{port}"
+    combo = SchedulerClusterCombo(1, address)
+    client = Client(address)
+
+    # warm up
+    client.submit(lambda: 0).result()
+    start = time.perf_counter()
+
+    test(client, *args, **kwargs)
+
+    # ensure all objects have been sent
+    client.submit(lambda: 0).result()
+    assert client._object_buffer._pending_objects.__len__() == 0
+
+    elapsed = time.perf_counter() - start
+
+    client.disconnect()
+    combo.shutdown()
+
+    return elapsed
+
+
+def send_objects(client: Client, n_objects: int, object_size: int) -> None:
+    payload = b"." * object_size
+    for _ in range(n_objects):
+        _ = client.send_object(payload)
+
+
+def main():
+    n_objects = np.array([100, 1000, 10_000, 20_000, 30_000, 40_000])
+    object_size = np.array([1024, 10 * 1024, 20 * 1024])
+
+    ymq_times = np.zeros((len(n_objects), len(object_size)))
+    zmq_times = np.zeros((len(n_objects), len(object_size)))
+    errors = np.zeros_like(ymq_times, dtype=bool)
+
+    port = 2345
+
+    for i, n in enumerate(n_objects):
+        for j, sz in enumerate(object_size):
+            print(f"Executing case: {n} objects of {sz} bytes")
+
+            try:
+                os.environ["SCALER_NETWORK_BACKEND"] = "ymq"
+                ymq_times[i, j] = timed_execution(send_objects, port, n, sz)
+                port += 1
+
+                os.environ["SCALER_NETWORK_BACKEND"] = "tcp_zmq"
+                zmq_times[i, j] = timed_execution(send_objects, port, n, sz)
+                port += 1
+            except KeyboardInterrupt:
+                port += 1
+                ymq_times[i, j] = np.nan
+                zmq_times[i, j] = np.nan
+                errors[i, j] = True
+
+    delta = ymq_times / zmq_times
+
+    print(delta)
+
+    plot_delta_with_errors(delta, errors, n_objects, object_size, out_path="plot.png")
+    print("Saved plot.png")
+
+
+if __name__ == "__main__":
+    main()