A tenant that fills a shared LLM prefix cache with new, never-reused data is made to pay for evictions first, protecting other tenants' cached prompts while preserving idle-capacity borrowing.
ThunderServe: High-performance and cost-efficient LLM serving in cloud environments,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.DC 1years
2026 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Preserving Admission Responsibility in Multi-Tenant Large Language Model Prefix Caches
A tenant that fills a shared LLM prefix cache with new, never-reused data is made to pay for evictions first, protecting other tenants' cached prompts while preserving idle-capacity borrowing.