In a typical MapReduce framework, which guarantees are provided regarding parallelism and fault tolerance?

• A. Tasks are executed serially to preserve deterministic ordering.
• B. The framework guarantees no retries, as failures are considered fatal.
• C. Tasks run in parallel; framework retries failed tasks; data is replicated; reducers aggregate results.
• D. Reducing is performed before mapping.

Answer: (C)

MapReduce guarantees parallel execution and fault tolerance through a few key ideas. Map tasks run concurrently across multiple nodes, each handling different chunks of input data. If a task fails, the framework automatically retries it (often on a different node), and the system relies on replicated data in the storage layer so work can be recovered or reissued. After the map phase, reducers aggregate the intermediate results to produce the final output. This combination—parallel map tasks, automatic retries for failures, data replication for resilience, and reduction to combine results—embodies how MapReduce handles both parallelism and fault tolerance.

Serial execution and a lack of retries don’t fit, since MapReduce is designed to run in parallel and to recover from failures by retrying tasks. Reducing before mapping would disrupt the required flow, as mapping must produce key-value pairs that the reduction phase can then combine.