Hadoop employs the strategy of data replication to balance load and ensure data availability across the cluster. Data is replicated across multiple nodes, providing fault tolerance and enabling parallel processing by allowing tasks to be executed on the closest available data copy.

The primary challenge in unit testing Hadoop applications involving HDFS is handling large datasets. Unit testing typically involves smaller datasets, and dealing with the volume of data in HDFS during testing poses challenges. Strategies like using smaller datasets or mocking HDFS interactions are often employed to address this challenge.

A local job runner in Hadoop unit testing simulates the Hadoop environment locally. It allows developers to test their MapReduce jobs on a single machine before deploying them on a Hadoop cluster, facilitating faster development cycles and easier debugging.

When a Hadoop job is running slower than expected, the initial check should focus on Hadoop configuration. This includes parameters related to memory, task allocation, and parallelism. Suboptimal configuration settings can significantly impact job performance.

The hdfs-site.xml file in Hadoop configuration specifies the number of replicas for each block in HDFS. This configuration is essential for ensuring fault tolerance and data reliability by controlling the replication factor of data blocks across the cluster.

One strategy to optimize a Reduce task that is slower than the Map tasks is Speculative Execution. In this approach, multiple instances of the same Reduce task are launched on different nodes, and the one that finishes first is accepted, reducing the overall job completion time.

HBase is a distributed NoSQL database that integrates with the Hadoop ecosystem for efficient data storage and retrieval. It is designed to handle large volumes of sparse data and is well-suited for random, real-time read/write access to Hadoop data.

Resource management is a key aspect of managing and optimizing cluster performance in Hadoop. Tools like YARN (Yet Another Resource Negotiator) play a crucial role in efficiently allocating and managing resources for running applications in the Hadoop cluster.

Apache Spark's ClusterManager feature allows for dynamic allocation of resources based on workload. The ClusterManager dynamically adjusts the resources allocated to Spark applications based on their needs, optimizing resource utilization.

The component of Apache Pig that translates scripts into MapReduce jobs is the Pig Compiler. It takes Pig Latin scripts as input and converts them into a series of MapReduce jobs that can be executed on a Hadoop cluster for data processing.

MapReduce Combiner is an optimization technique that allows for efficient data aggregation before sending data to the reducers. It helps reduce the amount of data shuffled across the network, improving overall performance in MapReduce jobs.

Oozie ensures efficient workflow management in complex data pipelines through its Workflow feature. Workflows in Oozie allow you to define a sequence of actions, manage dependencies, and handle the flow of data between Hadoop jobs. This is essential for orchestrating interdependent tasks and ensuring the overall efficiency of the data processing pipeline.