Trace: comm_summary and comm_over_time

pipit/tests/trace.py

@@ -32,6 +32,42 @@ def test_comm_matrix(data_dir, ping_pong_otf2_trace):
     assert count_comm_matrix[0][1] == count_comm_matrix[1][0] == 8
 
 
+def test_comm_over_time(data_dir, ping_pong_otf2_trace):
+    ping_pong = Trace.from_otf2(str(ping_pong_otf2_trace))
+
+    hist, edges = ping_pong.comm_over_time(output="size", message_type="send", bins=5)
+
+    assert len(edges) == 6
+    assert all(hist[0:3] == 0)
+    assert hist[4] == 4177920 * 2
+
+    hist, edges = ping_pong.comm_over_time(
+        output="count", message_type="receive", bins=5
+    )
+
+    assert len(edges) == 6
+    assert all(hist[0:3] == 0)
+    assert hist[4] == 8 * 2
+
+
+def test_comm_by_process(data_dir, ping_pong_otf2_trace):
+    ping_pong = Trace.from_otf2(str(ping_pong_otf2_trace))
+
+    sizes = ping_pong.comm_by_process()
+
+    assert sizes.loc[0]["Sent"] == 4177920
+    assert sizes.loc[0]["Received"] == 4177920
+    assert sizes.loc[1]["Sent"] == 4177920
+    assert sizes.loc[1]["Received"] == 4177920
+
+    counts = ping_pong.comm_by_process(output="count")
+
+    assert counts.loc[0]["Sent"] == 8
+    assert counts.loc[0]["Received"] == 8
+    assert counts.loc[1]["Sent"] == 8
+    assert counts.loc[1]["Received"] == 8
+
+
 def test_match_events(data_dir, ping_pong_otf2_trace):
     trace = Trace.from_otf2(str(ping_pong_otf2_trace))
     trace._match_events()

pipit/trace.py

@@ -454,6 +454,63 @@ def message_histogram(self, bins=20, **kwargs):
 
         return np.histogram(sizes, bins=bins, **kwargs)
 
+    def comm_over_time(self, output="size", message_type="send", bins=50, **kwargs):
+        """Returns histogram of communication volume over time.
+
+        Args:
+            output (str, optional). Whether to calculate communication by "count" or
+            "size". Defaults to "size".
+
+            message_type (str, optional): Whether to compute for sends or
+            receives. Defaults to "send".
+
+            bins (int, optional): Number of bins in the histogram. Defaults to
+            50.
+
+        Returns:
+            hist: Volume in size or number of messages in each time interval
+            edges: Edges of time intervals
+        """
+        # Filter by send or receive events
+        events = self.events[
+            self.events["Name"].isin(
+                ["MpiSend", "MpiIsend"]
+                if message_type == "send"
+                else ["MpiRecv", "MpiIrecv"]
+            )
+        ]
+
+        # Get timestamps and sizes
+        timestamps = events["Timestamp (ns)"]
+        sizes = events["Attributes"].apply(lambda x: x["msg_length"])
+
+        return np.histogram(
+            timestamps,
+            bins=bins,
+            weights=sizes.tolist() if output == "size" else None,
+            range=[
+                self.events["Timestamp (ns)"].min(),
+                self.events["Timestamp (ns)"].max(),
+            ],
+            **kwargs
+        )
+
+    def comm_by_process(self, output="size"):
+        """Returns total communication volume in size or number of messages per
+           process.
+
+        Returns:
+            pd.DataFrame: DataFrame containing total communication volume or
+            number of messags sent and received by each process.
+        """
+        comm_matrix = self.comm_matrix(output=output)
+
+        # Get total sent and received for each process
+        sent = comm_matrix.sum(axis=1)
+        received = comm_matrix.sum(axis=0)
+
+        return pd.DataFrame({"Sent": sent, "Received": received}).rename_axis("Process")
+
     def flat_profile(
         self, metrics="time.exc", groupby_column="Name", per_process=False
     ):