Hacker News Reader

The usual phrase is "scan resistance", which is especially important for databases. A pure LRU policy does poorly on scans; LFU does better but performs worse than LRU on other workloads. The original ARC paper[0] has a good discussion of the tradeoffs.

[0] https://www.usenix.org/legacy/events/fast03/tech/full_papers...

View on HN

Scan resistance and loop resistance are completely different. They are also distinguished in the papers. Note that ARC has scan resistance but no loop resistance.

None of the papers linked above or linked in the GitHub repo mention “loop resistance.”

Not completely different; they're both sequential access patterns and equivalent whenever the loop size exceeds cache size.

Not equivalent completely. Scan is a one-time access, so it never hits in scanning. Loop is multi-time accesses, so it can get many hits in looping if it has loop resistance. No hits on scan resistance in looping. Over.

I am disgusted by people who are repulsed by even the correction of a complete error. They must have bad grades in college. From now on, we have no choice but to leave it alone even if a false rumor is spread.

Reply on news.ycombinator.com