Google New Grad VO: Concurrent Fault-Tolerant File Broadcast Across a Data Center

Role: Software Engineer, New Grad
Result: Additional interview requested
Report time: February 2025

Design/coding prompt

A data center contains clusters of machines. Every cluster connects in a star topology to the same switch, and every wire has equal bandwidth. Design the fastest way to distribute a file to every machine.

The candidate first proposed BFS. The interviewer said that was not fast enough and asked for more concurrency and fault tolerance. The candidate then described and partially implemented a flooding-style algorithm that keeps machines active rather than leaving nodes idle.

A commenter suggested doubling the number of file holders each round, but the interviewer had explicitly objected to plans with idle nodes.

Missing contract details

The source does not define:

• Which machine initially owns the file
• File size or chunkability
• Whether a machine may forward while receiving
• The failure model
• Whether the objective is total completion time, rounds, or bandwidth usage

Those choices determine the correct design, so no algorithm or system-design link is asserted.

The candidate found the interview unusual, raised the issue with recruiting, and later reported that Google requested an additional round.