Differences

This shows you the differences between two versions of the page.

--- pub:members:xingda_wei [2026/10/30 17:59] – [Students] Add verified homepages for Dingyan Zhang and Yingyi Hao profile
+++ pub:members:xingda_wei [2026/10/30 18:02] (current) – [Research] profile
@@ Line 29: / Line 29: @@
     * **LLM request scheduling:** LMetric explores simple yet effective methods for scheduling LLM requests at scale. {{:publications:lmetric-osdi26.pdf|OSDI'26}} | [[https://github.com/blitz-serving/blitz-router|Code]]
     * **Elastic LLM serving:** BlitzScale enables fast, live model autoscaling; KunServe addresses memory overloading through parameter-centric memory management. {{:publications:blitzscale-osdi25.pdf|OSDI'25}} | {{:publications:kunserve-eurosys26.pdf|EuroSys'26}}
-    * **GPU checkpoint and restore:** PhoenixOS is the first system to support concurrent OS-level GPU checkpointing and restoration using validated speculation. {{:publications:phos-sosp25.pdf|SOSP'25}} | [[https://github.com/SJTU-IPADS/PhoenixOS|Code]]
+    * **GPU checkpoint and restore:** PhoenixOS is the first system to support concurrent OS-level GPU checkpointing and restoration using validated speculation for fault tolerance and fast startup. {{:publications:phos-sosp25.pdf|SOSP'25}} | [[https://github.com/SJTU-IPADS/PhoenixOS|Code]]
     * **Data access for AI:** AITurbo improves cloud storage I/O for AI jobs by coordinating storage and compute resources; our KVCache study characterizes access patterns to improve cache management. {{:publications:aiturbo-fast26.pdf|FAST'26}} | {{:publications:kvcache-atc25.pdf|ATC'25}}