hesela.devReuse time
Reuse time is the elapsed logical or physical time between consecutive accesses to the same data identity.
State the clock and endpoint convention: in a request-index clock, subtracting the previous index from the current index gives one for immediate repetition. This is not the count of distinct intervening identities. A cold-start first appearance has no observed preceding access. [1]
An unordered reuse-time histogram does not uniquely determine the reuse-distance histogram. This is an established non-equivalence, not a failure of every statistical estimator. [1]
Exact example: AACBAB and ABAABC each access A three times, B twice and C once, with one interval each of lengths 1, 2 and 3. Empty two-slot LRU misses 4/6 and 3/6 respectively. The ordered interval sequences differ. These are constructed equal-size traces with immediate demand admission, not measured SSD performance. [2]
Relations
Citations
- Yuan, Ding, Smith, Denning and Zhang. A Relational Theory of Locality. ACM TACO 16(3), article 33 (2019), sections 2.2-2.3.2. DOI: 10.1145/3341109.
- Hesela. Exact reuse-time histogram counterexample, assumptions and reproducible JSON.
Sources & caveats
- Primary-paper review and exact constructed arithmetic. No new theorem, hardware experiment or performance replication.