นโยบายการนำข้อมูลออกจากแคช

A cache is fast because it lives in limited memory. When it fills up, it must evict something to make room for new data. The eviction policy decides what to drop.

A good policy keeps the data most likely to be reused.

The metric that matters is the hit rate: the fraction of requests served from cache. A better eviction policy raises the hit rate, which means fewer slow trips to the database or origin.

FIFO (First In, First Out) evicts the oldest inserted item regardless of usage. It is simple but ignores access patterns, so a frequently used old item can be wrongly evicted.

LRU evicts the item that has not been accessed for the longest time. It assumes recently used data will be used again soon — true for most workloads, which is why LRU is the default in many caches.

A classic LRU uses an ordered map. On access, move the key to the most-recent end; when full, evict from the least-recent end.

LFU evicts the item accessed the fewest times. It favors long-term popular items over recent bursts. The downside: a once-popular item can linger long after it stops being useful.

A TTL (time to live) expires items after a fixed duration regardless of memory pressure. It bounds staleness and is often combined with LRU: TTL controls freshness, LRU controls memory.

Some caches offer random eviction (cheap, surprisingly decent) and scope variants: evict only keys with a TTL set, or evict from all keys. Redis exposes policies like allkeys-lru and volatile-ttl.

When a hot key is evicted or expires, many clients may simultaneously rebuild it — a thundering herd that hammers the origin. Mitigate with request coalescing (single-flight) or slightly randomized TTLs to spread out expirations.

Match the policy to the workload:

• Recency-driven traffic: LRU
• Stable popularity: LFU
• Freshness-critical data: TTL
• Uniform access: random is fine and cheap

CDNs apply the same ideas at the edge. Each edge node has finite storage and evicts (often LRU) plus honors Cache-Control: max-age as a TTL. Understanding eviction explains why a cold edge node has a low initial hit rate.

You learned how caches manage limited memory:

• Eviction policies aim to maximize hit rate
• FIFO, LRU, LFU, TTL, and random each suit different patterns
• Thundering herds need coalescing or jittered TTLs
• CDNs apply the same eviction logic at the edge