diff --git a/test/perf/generalized_assigment.jl b/test/perf/generalized_assigment.jl
new file mode 100644
index 000000000..c4f3f6362
--- /dev/null
+++ b/test/perf/generalized_assigment.jl
@@ -0,0 +1,95 @@
+using Profile, PProf
+using Coluna
+
+using GLPK, ColunaDemos, JuMP, BlockDecomposition
+
+function gap_toy_instance()
+    data = ColunaDemos.GeneralizedAssignment.data("play2.txt")
+
+    coluna = JuMP.optimizer_with_attributes(
+        Coluna.Optimizer,
+        "params" => Coluna.Params(solver = Coluna.Algorithm.BranchCutAndPriceAlgorithm(
+            branchingtreefile = "playgap.dot"
+        )),
+        "default_optimizer" => GLPK.Optimizer
+    )
+
+    model, x, dec = ColunaDemos.GeneralizedAssignment.model(data, coluna)
+    BlockDecomposition.objectiveprimalbound!(model, 100)
+    BlockDecomposition.objectivedualbound!(model, 0)
+
+    JuMP.optimize!(model)
+end
+
+
+function gap_strong_branching()
+    println("\e[45m gap strong branching \e[00m")
+    data = ColunaDemos.GeneralizedAssignment.data("mediumgapcuts3.txt")
+
+    coluna = JuMP.optimizer_with_attributes(
+        Coluna.Optimizer,
+        "params" => Coluna.Params(
+            solver = Coluna.Algorithm.BranchCutAndPriceAlgorithm(
+                maxnumnodes = 300,
+                colgen_stabilization = 1.0,
+                colgen_cleanup_threshold = 150,
+                stbranch_phases_num_candidates = [10, 3, 1],
+                stbranch_intrmphase_stages = [(userstage=1, solverid=1, maxiters=2)]
+            )
+        ),
+        "default_optimizer" => GLPK.Optimizer
+    )
+
+    model, x, dec = ColunaDemos.GeneralizedAssignment.model(data, coluna)
+
+    # we increase the branching priority of variables which assign jobs to the first two machines
+    for machine in 1:2
+        for job in data.jobs
+            BlockDecomposition.branchingpriority!(x[machine,job], 2)
+        end
+    end  
+
+    BlockDecomposition.objectiveprimalbound!(model, 2000.0)
+    BlockDecomposition.objectivedualbound!(model, 0.0)
+
+    JuMP.optimize!(model)
+end
+
+function gap_big_instance()
+    data = ColunaDemos.GeneralizedAssignment.data("mediumgapcuts1.txt")
+
+    coluna = JuMP.optimizer_with_attributes(
+        Coluna.Optimizer,
+        "params" => Coluna.Params(solver = Coluna.Algorithm.BranchCutAndPriceAlgorithm(
+            maxnumnodes=5
+        )
+        ),
+        "default_optimizer" => GLPK.Optimizer
+    )
+
+    model, x, dec = ColunaDemos.GeneralizedAssignment.model(data, coluna)
+    #BlockDecomposition.objectiveprimalbound!(model, 100)
+    #BlockDecomposition.objectivedualbound!(model, 0)
+
+    JuMP.optimize!(model)
+end
+
+
+#gap_toy_instance()
+#gap_strong_branching()
+gap_big_instance()
+
+Profile.clear()
+Profile.init(; n = 10^6, delay = 0.005)
+#@profile gap_toy_instance()
+#@profile gap_strong_branching()
+@profile gap_big_instance()
+pprof()
+readline()
+
+# Collect an allocation profile
+#Profile.Allocs.@profile gap_big_instance()
+
+# Export pprof allocation profile and open interactive profiling web interface.
+#PProf.Allocs.pprof()
+#readline()
\ No newline at end of file
diff --git a/test/perf/glossary.md b/test/perf/glossary.md
new file mode 100644
index 000000000..e5f7c45e3
--- /dev/null
+++ b/test/perf/glossary.md
@@ -0,0 +1,35 @@
+## Generate Julia code profile using ```PProf.jl```
+
+https://github.com/JuliaPerf/PProf.jl
+
+
+
+## Interpret code profiling
+
+useful links:
+https://docs.julialang.org/en/v1/devdocs/init/
+https://docs.julialang.org/en/v1/devdocs/ast/
+https://docs.julialang.org/en/v1/devdocs/functions/
+
+
+some Julia build-in methods:
+
+- ```jl_repl_entrypoint```: loads the contents of ```argv[ ]```
+- ```eval``` calls ```jl_toplevel_eval_in``` which calls ```jl_toplevel_eval_flex```
+- ```jl_toplevel_eval_flex``` implements a simple heuristic to decide whether to compile a given code thunk or run it by interpreter.
+- ```jl_interpret_toplevel_thunk``` is called by ```jl_toplevel_eval_flex``` when deciding to run the code by interpreter. It then calls ```eval_body```
+- ```jl_apply_generic```: perfoms the dispatch process.
+- ```kwcall```: "keyword argument sorter" or "keyword sorter", manages the keyword argurments when calling a method, e.g. consider method (see ```/devdocs/functions.md``` from Julia repo to get an example and more details)
+
+
+how to read pprof output ?
+
+- graph: (taken from https://github.com/google/pprof/issues/493)
+    * Dotted/dashed lines indicated that that intervening nodes have been removed. Nodes are removed to keep graphs small enough for visualization.
+    * The wider an arrow the more of the metric being measured is used along that path.
+    *  Nodes and edges are colored according to their total value (total being the amount of a metric used by a node and all of its children); large positive values are red; large negative values are green.
+
+- flat/cum:
+    
+    * flat: the value of the location itself.
+    * cum: the value of the location plus all its descendants.