Add cost distribution analysis

Author: networmix

docs/examples/basic.md

@@ -121,9 +121,71 @@ print(f"Equal-balanced flow: {max_flow_shortest_balanced}")
 
 Note that `EQUAL_BALANCED` flow placement is only applicable when calculating MaxFlow on shortest paths.
 
+## Cost Distribution Analysis
+
+The cost distribution feature analyzes how flow is distributed across paths of different costs for latency span analysis and network performance characterization.
+
+```python
+# Get flow analysis with cost distribution
+result = network.max_flow_with_summary(
+    source_path="A",
+    sink_path="C",
+    mode="combine"
+)
+
+# Extract flow value and summary
+(src_label, sink_label), (flow_value, summary) = next(iter(result.items()))
+
+print(f"Total flow: {flow_value}")
+print(f"Cost distribution: {summary.cost_distribution}")
+
+# Example output:
+# Total flow: 6.0
+# Cost distribution: {2.0: 3.0, 4.0: 3.0}
+#
+# This means:
+# - 3.0 units of flow use paths with total cost 2.0 (A→B→C path)
+# - 3.0 units of flow use paths with total cost 4.0 (A→D→C path)
+```
+
+### Latency Span Analysis
+
+When link costs represent latency (e.g., distance-based), the cost distribution provides insight into traffic latency characteristics:
+
+```python
+def analyze_latency_span(cost_distribution):
+    """Analyze latency characteristics from cost distribution."""
+    if not cost_distribution:
+        return "No flow paths available"
+
+    total_flow = sum(cost_distribution.values())
+    weighted_avg_latency = sum(cost * flow for cost, flow in cost_distribution.items()) / total_flow
+
+    min_latency = min(cost_distribution.keys())
+    max_latency = max(cost_distribution.keys())
+    latency_span = max_latency - min_latency
+
+    print(f"Latency Analysis:")
+    print(f"  Average latency: {weighted_avg_latency:.2f}")
+    print(f"  Latency range: {min_latency:.1f} - {max_latency:.1f}")
+    print(f"  Latency span: {latency_span:.1f}")
+    print(f"  Flow distribution:")
+    for cost, flow in sorted(cost_distribution.items()):
+        percentage = (flow / total_flow) * 100
+        print(f"    {percentage:.1f}% of traffic uses paths with latency {cost:.1f}")
+
+# Example usage
+analyze_latency_span(summary.cost_distribution)
+```
+
+This analysis helps identify:
+- **Traffic concentration**: How much traffic uses low vs. high latency paths
+- **Latency span**: The range of latencies experienced by traffic
+- **Performance bottlenecks**: When high-latency paths carry traffic due to capacity constraints
+
 ## Advanced Analysis: Sensitivity Analysis
 
-For deeper network analysis, you can use the low-level graph algorithms to perform sensitivity analysis and identify bottleneck edges:
+For network analysis, you can use the low-level graph algorithms to run sensitivity analysis and identify bottleneck edges:
 
 ```python
 from ngraph.lib.algorithms.max_flow import calc_max_flow, saturated_edges, run_sensitivity

docs/reference/api-full.md

@@ -10,7 +10,7 @@ For a curated, example-driven API guide, see **[api.md](api.md)**.
 > - **[CLI Reference](cli.md)** - Command-line interface
 > - **[DSL Reference](dsl.md)** - YAML syntax guide
 
-**Generated from source code on:** July 29, 2025 at 16:59 UTC
+**Generated from source code on:** July 29, 2025 at 17:52 UTC
 
 **Modules auto-discovered:** 53
 
@@ -2205,6 +2205,9 @@ Attributes:
     residual_cap: Remaining capacity on each edge after flow placement.
     reachable: Set of nodes reachable from source in residual graph.
     min_cut: List of saturated edges that form the minimum cut.
+    cost_distribution: Distribution of flow volume over path costs.
+        Maps each cost value to the total volume of flow placed on
+        paths with that cost during sequential augmentation.
 
 **Attributes:**
 
@@ -2213,6 +2216,7 @@ Attributes:
 - `residual_cap` (Dict[Edge, float])
 - `reachable` (Set[str])
 - `min_cut` (List[Edge])
+- `cost_distribution` (Dict[Cost, float])
 
 ---
 

docs/reference/api.md

@@ -142,6 +142,18 @@ max_flow = network.max_flow(
     shortest_path=True,         # Use only shortest paths
     flow_placement=FlowPlacement.PROPORTIONAL  # UCMP load balancing
 )
+
+# Detailed flow analysis with cost distribution
+result = network.max_flow_with_summary(
+    source_path="datacenter.*",
+    sink_path="edge.*",
+    mode="combine"
+)
+(src_label, sink_label), (flow_value, summary) = next(iter(result.items()))
+
+# Cost distribution shows flow volume per path cost (useful for latency analysis)
+print(f"Cost distribution: {summary.cost_distribution}")
+# Example: {2.0: 150.0, 4.0: 75.0} means 150 units on cost-2 paths, 75 on cost-4 paths
 ```
 
 **Key Options:**
@@ -150,6 +162,11 @@ max_flow = network.max_flow(
 - `shortest_path`: `True` (shortest only) or `False` (all available paths)
 - `flow_placement`: `FlowPlacement.PROPORTIONAL` (UCMP) or `FlowPlacement.EQUAL_BALANCED` (ECMP)
 
+**Advanced Features:**
+
+- **Cost Distribution**: `FlowSummary.cost_distribution` provides flow volume breakdown by path cost for latency span analysis and performance characterization
+- **Analytics**: Edge flows, residual capacities, min-cut analysis, and reachability information
+
 **Integration:** Available on both Network and NetworkView objects. Foundation for FailureManager Monte Carlo analysis.
 
 ### NetworkView

ngraph/lib/algorithms/max_flow.py

@@ -1,11 +1,11 @@
 """Maximum flow algorithms and network flow computations."""
 
-from typing import Literal, Union, overload
+from typing import Dict, Literal, Union, overload
 
 from ngraph.lib.algorithms.base import EdgeSelect, FlowPlacement
 from ngraph.lib.algorithms.flow_init import init_flow_graph
 from ngraph.lib.algorithms.place_flow import place_flow_on_graph
-from ngraph.lib.algorithms.spf import spf
+from ngraph.lib.algorithms.spf import Cost, spf
 from ngraph.lib.algorithms.types import FlowSummary
 from ngraph.lib.graph import NodeID, StrictMultiDiGraph
 
@@ -208,6 +208,7 @@ def calc_max_flow(
                 capacity_attr,
                 flow_attr,
                 tolerance,
+                {},  # Empty cost distribution for self-loop case
             )
         else:
             return 0.0
@@ -220,8 +221,11 @@ def calc_max_flow(
         reset_flow_graph,
     )
 
+    # Initialize cost distribution tracking
+    cost_distribution: Dict[Cost, float] = {}
+
     # First path-finding iteration.
-    _, pred = spf(
+    costs, pred = spf(
         flow_graph, src_node, edge_select=EdgeSelect.ALL_MIN_COST_WITH_CAP_REMAINING
     )
     flow_meta = place_flow_on_graph(
@@ -236,6 +240,13 @@ def calc_max_flow(
     )
     max_flow = flow_meta.placed_flow
 
+    # Track cost distribution for first iteration
+    if dst_node in costs and flow_meta.placed_flow > 0:
+        path_cost = costs[dst_node]
+        cost_distribution[path_cost] = (
+            cost_distribution.get(path_cost, 0.0) + flow_meta.placed_flow
+        )
+
     # If only one path (single augmentation) is desired, return early.
     if shortest_path:
         return _build_return_value(
@@ -247,11 +258,12 @@ def calc_max_flow(
             capacity_attr,
             flow_attr,
             tolerance,
+            cost_distribution,
         )
 
     # Otherwise, repeatedly find augmenting paths until no new flow can be placed.
     while True:
-        _, pred = spf(
+        costs, pred = spf(
             flow_graph, src_node, edge_select=EdgeSelect.ALL_MIN_COST_WITH_CAP_REMAINING
         )
         if dst_node not in pred:
@@ -274,6 +286,13 @@ def calc_max_flow(
 
         max_flow += flow_meta.placed_flow
 
+        # Track cost distribution for this iteration
+        if dst_node in costs and flow_meta.placed_flow > 0:
+            path_cost = costs[dst_node]
+            cost_distribution[path_cost] = (
+                cost_distribution.get(path_cost, 0.0) + flow_meta.placed_flow
+            )
+
     return _build_return_value(
         max_flow,
         flow_graph,
@@ -283,6 +302,7 @@ def calc_max_flow(
         capacity_attr,
         flow_attr,
         tolerance,
+        cost_distribution,
     )
 
 
@@ -295,6 +315,7 @@ def _build_return_value(
     capacity_attr: str,
     flow_attr: str,
     tolerance: float,
+    cost_distribution: Dict[Cost, float],
 ) -> Union[float, tuple]:
     """Build the appropriate return value based on the requested flags."""
     if not (return_summary or return_graph):
@@ -303,7 +324,13 @@ def _build_return_value(
     summary = None
     if return_summary:
         summary = _build_flow_summary(
-            max_flow, flow_graph, src_node, capacity_attr, flow_attr, tolerance
+            max_flow,
+            flow_graph,
+            src_node,
+            capacity_attr,
+            flow_attr,
+            tolerance,
+            cost_distribution,
         )
 
     ret: list = [max_flow]
@@ -322,6 +349,7 @@ def _build_flow_summary(
     capacity_attr: str,
     flow_attr: str,
     tolerance: float,
+    cost_distribution: Dict[Cost, float],
 ) -> FlowSummary:
     """Build a FlowSummary from the flow graph state."""
     edge_flow = {}
@@ -364,6 +392,7 @@ def _build_flow_summary(
         residual_cap=residual_cap,
         reachable=reachable,
         min_cut=min_cut,
+        cost_distribution=cost_distribution,
     )
 
 

ngraph/lib/algorithms/types.py

@@ -5,6 +5,8 @@
 from dataclasses import dataclass
 from typing import Dict, List, Set, Tuple
 
+from ngraph.lib.algorithms.base import Cost
+
 # Edge identifier tuple: (source_node, destination_node, edge_key)
 Edge = Tuple[str, str, str]
 
@@ -23,10 +25,14 @@ class FlowSummary:
         residual_cap: Remaining capacity on each edge after flow placement.
         reachable: Set of nodes reachable from source in residual graph.
         min_cut: List of saturated edges that form the minimum cut.
+        cost_distribution: Distribution of flow volume over path costs.
+            Maps each cost value to the total volume of flow placed on
+            paths with that cost during sequential augmentation.
     """
 
     total_flow: float
     edge_flow: Dict[Edge, float]
     residual_cap: Dict[Edge, float]
     reachable: Set[str]
     min_cut: List[Edge]
+    cost_distribution: Dict[Cost, float]

ngraph/network.py

@@ -463,6 +463,7 @@ def _compute_flow_with_summary_single_group(
                 residual_cap={},
                 reachable=set(),
                 min_cut=[],
+                cost_distribution={},
             )
             return 0.0, empty_summary
 
@@ -584,6 +585,7 @@ def _compute_flow_detailed_single_group(
                 residual_cap={},
                 reachable=set(),
                 min_cut=[],
+                cost_distribution={},
             )
             return 0.0, empty_summary, base_graph
 
@@ -1202,6 +1204,7 @@ def _max_flow_with_summary_internal(
                     residual_cap={},
                     reachable=set(),
                     min_cut=[],
+                    cost_distribution={},
                 )
                 return {(combined_src_label, combined_snk_label): (0.0, empty_summary)}
 
@@ -1215,6 +1218,7 @@ def _max_flow_with_summary_internal(
                     residual_cap={},
                     reachable=set(),
                     min_cut=[],
+                    cost_distribution={},
                 )
                 return {(combined_src_label, combined_snk_label): (0.0, empty_summary)}
             else:
@@ -1240,6 +1244,7 @@ def _max_flow_with_summary_internal(
                                 residual_cap={},
                                 reachable=set(),
                                 min_cut=[],
+                                cost_distribution={},
                             )
                             flow_val, summary = 0.0, empty_summary
                         else:
@@ -1255,6 +1260,7 @@ def _max_flow_with_summary_internal(
                             residual_cap={},
                             reachable=set(),
                             min_cut=[],
+                            cost_distribution={},
                         )
                         flow_val, summary = 0.0, empty_summary
                     results[(src_label, snk_label)] = (flow_val, summary)
@@ -1439,6 +1445,7 @@ def _max_flow_detailed_internal(
                     residual_cap={},
                     reachable=set(),
                     min_cut=[],
+                    cost_distribution={},
                 )
                 return {
                     (combined_src_label, combined_snk_label): (
@@ -1459,6 +1466,7 @@ def _max_flow_detailed_internal(
                     residual_cap={},
                     reachable=set(),
                     min_cut=[],
+                    cost_distribution={},
                 )
                 return {
                     (combined_src_label, combined_snk_label): (
@@ -1501,6 +1509,7 @@ def _max_flow_detailed_internal(
                                 residual_cap={},
                                 reachable=set(),
                                 min_cut=[],
+                                cost_distribution={},
                             )
                             flow_val, summary, flow_graph = (
                                 0.0,
@@ -1521,6 +1530,7 @@ def _max_flow_detailed_internal(
                             residual_cap={},
                             reachable=set(),
                             min_cut=[],
+                            cost_distribution={},
                         )
                         flow_val, summary, flow_graph = 0.0, empty_summary, base_graph
                     results[(src_label, snk_label)] = (flow_val, summary, flow_graph)