Questions: Readers-Writers Problem and Lock Patterns

Question 1

A database uses reader-preference locks. Traffic is 95% reads and 5% writes. Which problem is most likely to occur?

• A. Deadlock between multiple readers waiting for each other
• B. Writer starvation — writers may wait indefinitely while readers continuously arrive
• C. Reader starvation — writers will lock out readers for long periods
• D. Race conditions — concurrent readers can corrupt the shared data

Answer: (B)

Reader-preference allows new readers to enter as long as no writer is active. If reads are frequent (95%), there will almost always be at least one active reader, meaning the reader count never drops to zero and writers can wait indefinitely — writer starvation. Readers don't block each other (no deadlock between readers). Race conditions don't occur because writers still get exclusive access; the problem is that writers can't get that access promptly.

Question 2

What is the minimum information a fair readers-writers solution must track that a simple reader-preference solution does not?

• A. The number of active readers
• B. Whether any writer is currently active
• C. The arrival order or waiting queues for both readers and writers
• D. The total number of reads and writes that have occurred

Answer: (C)

Reader-preference tracks only the number of active readers and whether a writer is active — a purely count-based policy. This allows arbitrarily many new readers to jump ahead of a waiting writer. A fair solution must track arrival order so that when a writer is waiting, subsequent readers queue behind it rather than being admitted immediately. Fairness requires preventing queue-jumping, which requires knowing when each thread arrived relative to others.

Question 3

In a reader-preference solution, allowing multiple readers to access data simultaneously is safe because reads do not modify shared state.

• T. True
• F. False

Answer: True

Correct. The fundamental insight of the readers-writers problem is that read operations are non-destructive — two threads can read the same memory simultaneously without corrupting data or producing incorrect results. The exclusion requirement is only between writers and other accessors. Reader-preference exploits this safely by allowing concurrent reads, restricting only writer access to periods when no readers are active.

Question 4

Writer-preference solutions prevent writer starvation and should therefore be preferred over reader-preference solutions in most real-world systems.

• T. True
• F. False

Answer: False

False. Writer-preference prevents writer starvation but introduces reader starvation: if writes are frequent, readers may wait indefinitely. Neither preference-based solution is universally better — the right choice depends on the workload. Read-heavy systems may accept reader-preference; write-critical systems may prefer writer-preference. For balanced workloads, a fair solution (FIFO ordering, turnstile) prevents starvation on both sides but adds implementation complexity.

Question 5

Why does a fair readers-writers solution batch waiting readers together rather than serving them one at a time in strict FIFO order?

Model answer: If waiting readers were admitted one at a time in strict per-thread FIFO, the performance advantage of concurrent reading would be lost — readers would effectively be serialized. By releasing all readers queued behind a writer simultaneously once the writer completes, the system allows the concurrent reads that motivated the readers-writers design. The batching preserves the key insight: reads don't conflict with each other, so all readers waiting behind a writer can proceed together.

Strict per-reader FIFO would be correct (no starvation) but wasteful: R1 would acquire the lock, R2 would wait, R1 releases, R2 acquires — all sequential. Batching means: when writer W releases, all readers queued behind W are released together and read concurrently. The lock count climbs to n, they all read, the count drops to 0, allowing the next writer to proceed. This is both fair (W served before readers who arrived after W) and concurrent.