Reproduce “When Cache Pressure Becomes a Trap: Metastability in Agent Serving” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chengzhi Lu, Boxi Zhou, Yanying Lin, Dmitrii Ustiugov
The authors show how a short session surge can leave service degraded after the surge ends. In an SGLang experiment, terminating residual burst sessions restored service within two minutes, while simulation showed that later arrivals affect recovery. In a Terminal-Bench test, post-surge goodput was 17 versus 138 sessions per hour without a surge; these findings are specific to tested setups.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.