Tom van Dijk

So far, I have found that parallel performance seriously degraded with commit 2d5d83 in Sylvan. This is a rewrite of the CMake files. What I have discovered so far is...

I made a small script to run a few performance tests: ``` #!/bin/bash FILES="anderson.4.bdd bakery.4.bdd schedule_world.2.bdd lifts.6.bdd" for f in $FILES; do echo $f examples/bddmc -s bfs -w 1 ../models/$f...

I have now pushed Sylvan v1.7.0 on my github repository, I have not yet checked if this fixes the issue in Storm, in part because I cannot build Storm right...

@tquatmann I am not sure if the higher CMake version is still needed in Sylvan v1.7.0, as I have pruned the CMake script a bit. But I have not yet...

I still notice a clear problem in how Storm uses Sylvan, which is related to the changes in Lace. **In the old way**, initializing Lace would create `n_workers-1` threads as...

> thanks for all the input! What are the consequences of running everything inside Lace? A few minor inconveniences: - you can't disable Lace workers if you want to do...

So if Storm is used from Python bindings, then I recommend that whenever you're going to perform many BDD operations in a Storm method, wrap that in the Lace `TASK`....

There is by the way no clean way of having a "single-threaded" Sylvan with the current design. For example, the `bdd_refs_push` methods use thread-local stacks and the garbage collector needs...

By the way, I discovered CMake 3.14 has very nice support for dependencies now using FetchContent, so I played a bit with it and now I can just include Lace...

Somewhat related to this discussion is the design of Lace. Triggered by some comments of Jaco, I investigated whether I could actually significantly update the design of Lace and what...