Systems on Harvard Systems Grouphttps://systems.seas.harvard.edu/tags/systems/Recent content in Systems on Harvard Systems GroupHugoenCopyright © 2026 The President and Fellows of Harvard CollegeTue, 28 Jul 2026 11:00:00 -0400Why FIFO is (almost) all you need for cache evictionhttps://systems.seas.harvard.edu/blog/fifo-is-all-you-need/Tue, 28 Jul 2026 11:00:00 -0400https://systems.seas.harvard.edu/blog/fifo-is-all-you-need/<p>Cache eviction is usually presented as a ranking problem. LRU ranks objects by recency. LFU ranks them by frequency. More advanced algorithms combine several signals, adapt their parameters, and maintain increasingly sophisticated data structures to decide which object is least valuable.</p>MorphServe: Making Model Precision Elastic for Bursty LLM Servinghttps://systems.seas.harvard.edu/blog/make-model-precision-elastic/Mon, 27 Jul 2026 11:30:00 -0400https://systems.seas.harvard.edu/blog/make-model-precision-elastic/<p>Most LLM servers treat a deployed model as fixed: its weights have one precision, its KV cache has one memory budget, and both remain unchanged until the process restarts.</p>