Overview
Explore a conference talk on innovative page-table and TLB management techniques for NUMA systems. Delve into the challenges of memory management operations in highly threaded applications, particularly in NUMA settings. Learn about Hydra, a novel page-table management mechanism that enables transparent, on-demand, and partial page-table replication across NUMA nodes. Discover how Hydra's approach not only avoids replication-related slowdowns but also significantly improves performance on memory allocation/deallocation and access control operations. Examine the implementation of Hydra in Linux on x86_64 and its evaluation on 4- and 8-socket systems, showcasing impressive runtime improvements for applications like Webserver and Memcached.
Syllabus
USENIX ATC '24 - Scalable and Effective Page-table and TLB management on NUMA Systems
Taught by
USENIX