We Live in Interesting Times
summary
Argues memory management is more relevant than ever and gives concrete findings: a fleet study shows allocator changes cut TLB stalls by about 20% stalled cycles and saved money; an ARM Pointer Authentication defense is undermined by the PAC-Man attack; experiments with Intel HTM reveal orders-of-magnitude disagreement in cache capacity measurements that break algorithms; a new distillation methodology puts a lower bound on GC overhead and shows cycles can make serial GC look cheapest while Shenandoah/ZGC need large headroom; MMTk ports to many runtimes are underway.
AI-written summary · about
chapters (13)
- 0:00Start
- 1:01Steve Blackburn
- 1:42The V8 Javascript Virtual Machine
- 7:51Barbara Liskov
- 12:33Optimizing the Allocation Sequence
- 22:01How Eviction Works in a Cache
- 25:24Is gc Really a Solved Problem
- 27:25Ideal Application Cost
- 30:34Serial Gc
- 34:15Failure To Translate
- 36:12Failure To Measure
- 46:28Optane
- 55:34What Can Architects like Myself Do To Help the Memory Management Community
related talks
56mWhy Data Access Patterns Have Always Mattered
45mThe Joy of Building Large Scale Systems
42mAdventures in Efficiency and Performance
57mInvited Talk: Bridging Safety and Performance
01h 50mUnlocking Modern CPU Power - Next-Gen C++ Optimization Techniques
14mFreedom from memory interference
01h 01mWhat Programmers Should Know About Memory Allocation
50mSmashing the Stack: Let’s Make It Less Fun And Unprofitable!