The first benchmark results of Extreme LUD

[Yuki Shiino]

Hi team,

Let me share the very first benchmark results of Extreme Lightweight UaF Detector.

I have implemented the very initial version of Extreme Lightweight UaF Detector (yet unmatured), and have run some benchmarks. Here is a report about it.

We're planning to run a finch experiment to see its effectiveness.

Cheers,

Yuki Shiino

[Yuki Shiino]

Hi team,

Let me share the second benchmark results of the Extreme LUD after applying some optimizations. The results look better than before (the overhead is smaller than before).

https://docs.google.com/document/d/1t1DfFcXA5nC_nwVH8m1kZNq6cNapaOaBCQAOBE-JtKE/edit?usp=sharing

Cheers,

Yuki Shiino

2024年3月19日(火) 16:26 Yuki Shiino <yukis...@chromium.org>:

[Kentaro Hara]

Thanks Yuki for all the optimizations! The performance regression on Speedometer3 when you enable E-LUD on all processes (including the renderer) is only 0.2 - 0.4% ish. This is a great number. We are planning to enable E-LUD only on non-renderer processes -- let's start a launch process to start an in-wild experiment :)

--
You received this message because you are subscribed to the Google Groups "memory-safety-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to memory-safety-...@chromium.org.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/memory-safety-dev/CAN0uC_RQYds%3Dt_1%3DRdUSo0_Yppbb4SMLAuf84dYkFiqjjGi4%3DA%40mail.gmail.com.

--

Kentaro Hara, Tokyo