Cache Fundamentals & Design Goals Quiz

Q1. EASY: Which of the following best describes temporal locality in the context of caching? Recently accessed items are likely to be accessed again soon.
Accesses tend to occur in memory addresses that are close to one another.
Data that has not been used recently is quickly evicted from the cache.
Cache entries are stored based on the time at which they were added.

Q2. EASY: Which of the following is NOT a recognized category of cache miss? Capacity miss
Concurrent miss
Conflict miss
Compulsory miss

Q3. EASY: In caching, what does the acronym LRU stand for? Longest Retained Unit
Least Regularly Used
Least Recently Used
Last Recorded Usage

Q4. EASY: Memcached and Redis are primarily used as: Distributed file systems for persistent storage
Web server software for serving HTTP requests
Relational database engines for transaction processing
In-memory caches to store key-value data and reduce database load

Q5. EASY: A cache hit ratio of 90% means that: 90% of all requests are being served from the cache rather than the origin.
90% of the cache's capacity is filled with data.
90% of requests must go to the backend data source.
Each cached item is accessed 90 times before expiration.

Q6. MEDIUM: Which cache write strategy ensures that data is written to main memory at the same time it is written to the cache? Write-back
Write-around
Write-through
No-write allocate

Q7. MEDIUM: What is the primary benefit of using a CDN (Content Delivery Network) for caching web content? It intentionally increases traffic to the origin server.
It reduces content delivery latency by serving users from edge caches closer to them.
It guarantees that cached content will never expire or be evicted.
It doubles the amount of data a user must download.

Q8. MEDIUM: A certain cache has a 95% hit rate, a 1 ns hit latency, and a 50 ns miss penalty. Approximately what is the average memory access time (AMAT)? About 3.5 ns
About 1 ns
About 2.5 ns
About 50 ns

Q9. MEDIUM: Which cache eviction algorithm uses multiple lists to track both recency and frequency of access, adapting its behavior to the workload? LRU (Least Recently Used)
LFU (Least Frequently Used)
Random replacement
ARC (Adaptive Replacement Cache)

Q10. MEDIUM: What is an effective way to mitigate a cache stampede (thundering herd) when a highly requested cached item expires? Use a very short TTL (expiration) for that item to reduce load.
Disable caching for that item entirely.
Have only one request populate the cache (using locks or stale-while-revalidate) while others wait or serve stale data.
Switch the cache to a write-through policy for that item.

Q11. MEDIUM: What is a potential drawback of storing cache entries in a compressed form? It significantly increases the memory required to store data in the cache.
It adds CPU overhead and latency to compress/decompress data on each cache access.
It prevents the cache from evicting items when needed.
It causes a higher cache miss rate for frequently accessed items.

Q12. HARD: If a single cache key becomes extremely hot (receiving a huge fraction of traffic) in a distributed cache, what is one way to avoid overloading the cache node responsible for that key? Reduce the key's time-to-live (TTL) so it gets evicted more frequently.
Use a write-around caching policy for that key.
Turn off consistent hashing in the cache cluster.
Replicate that key's data to multiple cache nodes and distribute the read requests among them.

Q13. HARD: In a multi-CPU system with each core having its own cache, what problem can arise with a write-back cache if one core updates a data item in its cache? The updated cache line will be immediately evicted from all other caches.
Other cores may have stale data in their caches, requiring a cache coherence mechanism.
The memory will leak because write-back caches do not free data.
The hashing of cache keys becomes inconsistent across cores.

Q14. HARD: A consistently high number of evictions per second in a cache is likely a sign of what? The working set of data is too large for the cache, causing constant replacement (thrashing).
The cache's hit ratio is nearly 100%.
The cache is using a write-through policy.
Clients are not reading from the cache at all.

Q15. HARD: Why do cache performance metrics often include latency percentiles (e.g., p99) instead of just average latency? Because the 99th percentile latency is always lower than the average latency.
Because percentile metrics show the cache's hit ratio over time.
Because percentiles highlight tail latencies (slowest requests) that averages can mask.
Because an average cannot be computed for cache access times.

system-design