Storage & ML systems · Harvard SEAS

I measure how systems behave, then make the common path simpler, faster, and more efficient.

GPU memory changing shape between quiet load, a traffic spike, and recovery

Latest post07.23.26

Make model precision elastic.

The archive · 2026

Blog

Experiments, arguments, and practical advice from systems research.

  1. Your LRU cache is working too hard Across 6,357 production traces, most LRU promotions did little useful work. Delaying the decision made caches faster without making them less effective.
  2. Authorship inflation at OSDI: counting papers two ways OSDI author lists have more than doubled since 1994. Whether the field's most prolific authors are actually publishing more depends on whether a paper counts as one, or as one divided by its author count.
  3. Make model precision elastic MorphServe temporarily converts selected model layers into KV-cache headroom when traffic spikes, then restores full precision when the pressure passes.
  4. Why FIFO is (almost) all you need for cache eviction A simple first-in-first-out queue, given one quick second chance, beats far more complex eviction policies on real workloads. Here is the intuition behind S3-FIFO and SIEVE.
  5. Measurement-first systems research The most useful systems I have built started not with a design, but with a long, uncomfortable look at how the existing system actually behaved in production.
  6. Advice for prospective systems PhD students I am recruiting students who want to build real systems for storage, databases, and machine learning. Here is how I think about the work and what I look for.

Currently at Harvard

Start a conversation