Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Explore a conference talk addressing the challenge of mmap_sem contention in large-scale systems with thousands of CPUs. Delve into various attempts to mitigate this issue, including Speculative Page Fault, RangeLock, and Scalable Address Spaces Using RCU Balanced Trees. Examine the emerging solution discussed at the Linux Plumbers Conference 2019, which proposes a split lock mechanism similar to PTE locking. Learn about the potential use of Maple Trees to replace VMA RB trees and double-linked lists. Gain insights into VMA rules, locking strategies, and the complexities of memory management in multi-threaded environments. Understand the challenges and proposed solutions for improving scalability in Linux memory management for high-performance computing systems.
Syllabus
Introduction
VMA Rules
Locking the VMA
No existing VMA
Mapping is painful
Extending the VMA
VMA lock structure
VMA lock for readers
VMA lock for proc
Switching VMA
Finding a gap
Locking
One lock
Hazards
Plans
VMA
Taught by
Linux Plumbers Conference