diff --git a/garlic/exp/nbody/nblocks.nix b/garlic/exp/nbody/nblocks.nix
new file mode 100644
index 0000000..ce4a025
--- /dev/null
+++ b/garlic/exp/nbody/nblocks.nix
@@ -0,0 +1,110 @@
+{
+  stdenv
+, stdexp
+, bsc
+, targetMachine
+, stages
+, garlicTools
+
+# Options for the experiment
+, enableJemalloc ? false
+, enableCTF ? false
+# Number of cases tested
+, steps ? 7
+# nbody iterations
+, timesteps ? 10
+# nbody total number of particles
+, particles ? null
+, gitBranch ? "garlic/tampi+send+oss+task"
+, loops ? 10
+, nblocks0 ? null
+}:
+
+with stdenv.lib;
+with garlicTools;
+
+let
+
+  defaultOpt = var: def: if (var != null) then var else def;
+
+  machineConfig = targetMachine.config;
+  inherit (machineConfig) hw;
+
+  # Initial variable configuration
+  varConf = with bsc; {
+    # Create a list with values 2^n with n from 0 to (steps - 1) inclusive
+    i = expRange 2 0 (steps - 1);
+  };
+
+  # Generate the complete configuration for each unit
+  genConf = var: fix (self: var // targetMachine.config // {
+    expName = "nbody-nblocks";
+    unitName = "${self.expName}${toString self.nblocks}";
+
+    inherit (machineConfig) hw;
+
+    # nbody options
+    particles = defaultOpt particles (4096 * self.hw.cpusPerSocket);
+    nblocks0 = defaultOpt nblocks0 (self.hw.cpusPerSocket / 2);
+    # The number of blocks is then computed from the multiplier "i" and
+    # the initial number of blocks "nblocks0"
+    nblocks = self.i * self.nblocks0;
+
+    totalTasks = self.ntasksPerNode * self.nodes;
+    particlesPerTask = self.particles / self.totalTasks;
+    blocksize = self.particlesPerTask / self.nblocks;
+    cc = bsc.icc;
+    mpi = bsc.impi;
+    cflags = "-g";
+    inherit timesteps gitBranch enableJemalloc enableCTF loops;
+
+    # Resources
+    qos = "debug";
+    cpusPerTask = self.hw.cpusPerSocket;
+    ntasksPerNode = self.hw.socketsPerNode;
+    nodes = 1;
+    jobName = self.unitName;
+  });
+
+  # Compute the array of configurations
+  configs = stdexp.buildConfigs {
+    inherit varConf genConf;
+  };
+
+  perf = {nextStage, conf, ...}: with conf; stages.perf {
+    inherit nextStage;
+    perfOptions = "record --call-graph dwarf -o \\$\\$.perf";
+  };
+
+  ctf = {nextStage, conf, ...}: with conf; stages.exec {
+    inherit nextStage;
+    env = optionalString (conf.enableCTF) ''
+      export NANOS6_CONFIG_OVERRIDE="version.instrument=ctf,\
+        instrument.ctf.conversor.enabled=false"
+    '';
+  };
+
+  exec = {nextStage, conf, ...}: with conf; stages.exec {
+    inherit nextStage;
+    argv = [ "-t" timesteps "-p" particles ];
+  };
+
+  program = {nextStage, conf, ...}: with conf;
+    let
+      /* These changes are propagated to all dependencies. For example,
+      when changing nanos6+jemalloc, we will get tampi built with
+      nanos6+jemalloc as well. */
+      customPkgs = bsc.extend (self: super: {
+        mpi = conf.mpi;
+        nanos6 = self.nanos6.override { inherit enableJemalloc; };
+      });
+    in
+      customPkgs.apps.nbody.override ({
+        inherit cc blocksize mpi gitBranch cflags;
+      });
+
+  pipeline = stdexp.stdPipeline ++ [ ctf exec program ];
+
+in
+ 
+  stdexp.genExperiment { inherit configs pipeline; }
diff --git a/garlic/exp/nbody/tampi.nix b/garlic/exp/nbody/tampi.nix
deleted file mode 100644
index 09516be..0000000
--- a/garlic/exp/nbody/tampi.nix
+++ /dev/null
@@ -1,101 +0,0 @@
-{
-  stdenv
-, stdexp
-, bsc
-, targetMachine
-, stages
-, garlicTools
-, enableJemalloc ? false
-, particles ? null
-}:
-
-with stdenv.lib;
-with garlicTools;
-
-let
-
-  machineConfig = targetMachine.config;
-  inherit (machineConfig) hw;
-
-  # Number of cases tested
-  steps = 7;
-
-  # First value for nblocks: we want to begin by using 1/2 blocks/cpu so we set
-  # the first number of blocks to cpusPerSocket / 2
-  nblocks0 = hw.cpusPerSocket / 2;
-
-  # Initial variable configuration
-  varConf = with bsc; {
-    # Create a list with values 2^n with n from 0 to (steps - 1) inclusive
-    i = expRange 2 0 (steps - 1);
-  };
-
-  # Set here the particles, so we don't have an infinite recursion in the
-  # genConf attrset.
-  _particles = if (particles != null)
-    then particles
-    else 4096 * hw.cpusPerSocket;
-
-  # Generate the complete configuration for each unit
-  genConf = with bsc; c: targetMachine.config // rec {
-    expName = "nbody-nblocks";
-    unitName = "${expName}${toString nblocks}";
-
-    inherit (machineConfig) hw;
-    # nbody options
-    particles = _particles;
-    timesteps = 10;
-    nblocks = c.i * nblocks0;
-    totalTasks = ntasksPerNode * nodes;
-    particlesPerTask = particles / totalTasks;
-    blocksize = particlesPerTask / nblocks;
-    cc = icc;
-    mpi = impi;
-    gitBranch = "garlic/tampi+send+oss+task";
-    cflags = "-g";
-    inherit enableJemalloc;
-    
-    # Repeat the execution of each unit 10 times
-    loops = 10;
-
-    # Resources
-    qos = "debug";
-    cpusPerTask = hw.cpusPerSocket;
-    ntasksPerNode = hw.socketsPerNode;
-    nodes = 1;
-
-    jobName = unitName;
-  };
-
-  # Compute the array of configurations
-  configs = stdexp.buildConfigs {
-    inherit varConf genConf;
-  };
-
-  perf = {nextStage, conf, ...}: with conf; stages.perf {
-    inherit nextStage;
-    perfOptions = "record --call-graph dwarf -o \\$\\$.perf";
-  };
-
-  exec = {nextStage, conf, ...}: with conf; stages.exec {
-    inherit nextStage;
-    argv = [ "-t" timesteps "-p" particles ];
-  };
-
-  program = {nextStage, conf, ...}: with conf;
-  let
-    customPkgs = stdexp.replaceMpi conf.mpi;
-  in
-    customPkgs.apps.nbody.override ({
-      inherit cc blocksize mpi gitBranch cflags;
-    } // optionalAttrs enableJemalloc {
-      mcxx = bsc.mcxx.override {
-        nanos6 = bsc.nanos6Jemalloc;
-      };
-    });
-
-  pipeline = stdexp.stdPipeline ++ [ exec program ];
-
-in
- 
-  stdexp.genExperiment { inherit configs pipeline; }
diff --git a/overlay.nix b/overlay.nix
index cbcabf5..77e5ab8 100644
--- a/overlay.nix
+++ b/overlay.nix
@@ -372,14 +372,30 @@ let
       # Experiments
       exp = {
         nbody = rec {
-          tampi = callPackage ./garlic/exp/nbody/tampi.nix { };
+          baseline = callPackage ./garlic/exp/nbody/nblocks.nix { };
 
           # Experiment variants
-          baseline = tampi;
-          small = baseline.override { particles = 12 * 4096; };
+          small = baseline.override {
+            particles = 12 * 4096;
+          };
           # TODO: Update freeCpu using a non-standard pipeline
           #freeCpu = baseline.override { freeCpu = true; };
           jemalloc = baseline.override { enableJemalloc = true; };
+
+          # Some experiments with traces
+          trace = {
+            # Only one unit repeated 30 times
+            baseline = small.override {
+              enableCTF = true;
+              loops = 30;
+              steps = 1;
+            };
+
+            # Same but with jemalloc enabled
+            jemalloc = trace.baseline.override {
+              enableJemalloc = true;
+            };
+          };
         };
 
         saiph = {
@@ -450,6 +466,7 @@ let
           small = merge [ small ];
           jemalloc = merge [ baseline jemalloc ];
           #freeCpu  = merge [ baseline freeCpu ];
+          ctf = merge [ ctf ];
         };
 
         hpcg = with exp.hpcg; {
@@ -490,6 +507,10 @@ let
           #  script = ./garlic/fig/nbody/freeCpu.R;
           #  dataset = ds.nbody.freeCpu;
           #};
+          ctf = pp.rPlot {
+            script = ./garlic/fig/nbody/baseline.R;
+            dataset = ds.nbody.ctf;
+          };
         };
 
         hpcg = {