diff --git a/rucat_job/src/rdd/dependency.rs b/rucat_job/src/rdd/dependency.rs
new file mode 100644
index 0000000..1c90443
--- /dev/null
+++ b/rucat_job/src/rdd/dependency.rs
@@ -0,0 +1,20 @@
+use super::rdd::RDD;
+
+pub(super) struct Dependency {
+    rdd: RDD,
+    dependency_type: DependencyVariant,
+}
+
+impl Dependency {
+    pub fn get_rdd(&self) -> &RDD {
+        &self.rdd
+    }
+}
+
+/// [Dependency] can be converted to [RDD]
+pub(super) enum DependencyVariant {
+    Narrow,
+    Shuffle,
+    OneToOne,
+    Range,
+}
diff --git a/rucat_job/src/rdd/mod.rs b/rucat_job/src/rdd/mod.rs
index 7dcb093..3d79b8e 100644
--- a/rucat_job/src/rdd/mod.rs
+++ b/rucat_job/src/rdd/mod.rs
@@ -1,3 +1,4 @@
+mod dependency;
 mod deterministic_level;
 mod map_partition_rdd;
 mod partition;
diff --git a/rucat_job/src/rdd/partition.rs b/rucat_job/src/rdd/partition.rs
index 7c0613e..71a89de 100644
--- a/rucat_job/src/rdd/partition.rs
+++ b/rucat_job/src/rdd/partition.rs
@@ -7,7 +7,7 @@ pub(super) enum Partition {
 
 impl Partition {
     /// Get the identifier for a partition in an RDD
-    fn index(&self) -> PartitionIndex {
+    pub fn index(&self) -> PartitionIndex {
         match self {
             Partition::Dummy => 0,
         }
diff --git a/rucat_job/src/rdd/rdd.rs b/rucat_job/src/rdd/rdd.rs
index ad0bd56..32c845d 100644
--- a/rucat_job/src/rdd/rdd.rs
+++ b/rucat_job/src/rdd/rdd.rs
@@ -1,17 +1,15 @@
-use std::task::Context;
-
-use super::{partition::PartitionIndex, storage_level::StorageLevel};
+use super::{dependency::Dependency, partition::PartitionIndex, storage_level::StorageLevel};
 use crate::{rdd::partition::Partition, task_context::TaskContext};
 
 /// Element types of RDD
-enum RDDElem {
+pub(super) enum RDDElem {
     U8(u8),
 }
 
 /// define RDD as (Dependencies, RDDVariant) where
 /// RDDVariant is the enum of all kinds of RDDs.
 /// Unlilke the RDD in Spark, we don't define the context as an argument of RDD as it is `global`
-struct RDD {
+pub(super) struct RDD {
     dependencies: Vec<Dependency>,
     storage_level: Option<StorageLevel>,
     rdd_core: RDDVariant,
@@ -22,8 +20,28 @@ impl RDD {
         &self.dependencies
     }
 
-    fn compute(&self, split: Partition, context: Context) -> impl Iterator<Item = RDDElem> {
-        std::iter::empty()
+    /// the first parent RDD
+    fn get_first_parent(&self) -> Option<&RDD> {
+        self.dependencies.first().map(|d| d.get_rdd())
+    }
+
+    fn compute(&self, split: Partition, context: TaskContext) -> impl Iterator<Item = RDDElem> {
+        match &self.rdd_core {
+            RDDVariant::MapPartitionRDD { f } => f(
+                context,
+                split.index(),
+                Box::new(self.get_first_parent().unwrap().into_iter()),
+            ),
+        }
+    }
+}
+
+impl IntoIterator for &RDD {
+    type Item = RDDElem;
+    type IntoIter = std::vec::IntoIter<Self::Item>;
+
+    fn into_iter(self) -> Self::IntoIter {
+        todo!()
     }
 }
 
@@ -34,17 +52,8 @@ enum RDDVariant {
             dyn Fn(
                 TaskContext,
                 PartitionIndex,
-                dyn Iterator<Item = RDDElem>,
-            ) -> dyn Iterator<Item = RDDElem>,
+                Box<dyn Iterator<Item = RDDElem>>,
+            ) -> Box<dyn Iterator<Item = RDDElem>>,
         >,
     },
 }
-
-/// [Dependency] cannnot be generic because there is no way for RDD to know the type of its each [Dependency]
-/// [Dependency] can be converted to [RDD]
-enum Dependency {
-    Narrow,
-    Shuffle,
-    OneToOne,
-    Range,
-}
diff --git a/rucat_job/src/stage.rs b/rucat_job/src/stage.rs
index 200d585..45eab18 100644
--- a/rucat_job/src/stage.rs
+++ b/rucat_job/src/stage.rs
@@ -1,6 +1,6 @@
 use serde_traitobject as st;
 
-use crate::task::{SubDataTrait, SubExecute, SubResultTrait, Task};
+use crate::task::{SubDataTrait, SubExecute, SubResultTrait};
 
 /// Driver will call `get_data` and `split`, and send `Task` to workers.
 /// Workers simplify `SubTask`s to `SubResult`s and send them back to the driver.