Hi, and thank you for your work on VF3P.
I wanted to report several issues I encountered while using VF3P on larger datasets and more complex patterns:
-
Crashes on Large Graphs and Patterns
When running VF3P on large graphs (e.g., H01 dataset) with moderately large query patterns, I observed frequent crashes—especially when using more than 8 threads. The problem becomes more pronounced with increasing pattern size.
-
Multithreading Scalability Issue
I found that using more threads (e.g., 8 or 16) leads to higher memory usage and even slower performance than 4 threads. In some cases, memory usage explodes to tens of gigabytes, followed by segmentation faults.
-
Multiple Label Support
I could not find a clear way to load multiple labels for nodes or edges. Even when I tried to encode multiple labels as a single label, it caused crashes once the label sizes increased slightly.
-
Crash After Dataset Load
Even when the large graph appears to load successfully, starting the subgraph matching phase with large patterns and multithreading (again, 8+ threads) often causes an immediate crash.
Hi, and thank you for your work on VF3P.
I wanted to report several issues I encountered while using VF3P on larger datasets and more complex patterns:
Crashes on Large Graphs and Patterns
When running VF3P on large graphs (e.g., H01 dataset) with moderately large query patterns, I observed frequent crashes—especially when using more than 8 threads. The problem becomes more pronounced with increasing pattern size.
Multithreading Scalability Issue
I found that using more threads (e.g., 8 or 16) leads to higher memory usage and even slower performance than 4 threads. In some cases, memory usage explodes to tens of gigabytes, followed by segmentation faults.
Multiple Label Support
I could not find a clear way to load multiple labels for nodes or edges. Even when I tried to encode multiple labels as a single label, it caused crashes once the label sizes increased slightly.
Crash After Dataset Load
Even when the large graph appears to load successfully, starting the subgraph matching phase with large patterns and multithreading (again, 8+ threads) often causes an immediate crash.