RFR: 8312132: Add tracking of multiple address spaces in NMT [v105]

[Thomas Stuefe]

On Thu, 23 May 2024 12:41:46 GMT, Johan Sjölen <jsjolen at openjdk.org> wrote:

> > We claim that:
> > > Our Treap-based approach in this patch gives a performance boost such that we see 25x better performance in a benchmark.
> > 
> > 
> > May I ask how you ran it? I would like to be able to reproduce our claim.
> 
> Sure, it was a while since I ran the benchmark. You're going to have to do a bit of work here, to get it working.
> 
> You take this file: https://github.com/tstuefe/jdk/blob/6be830cd2e90a009effb016fbda2e92e1fca8247/test/hotspot/gtest/nmt/test_nmtvmadict.cpp#L1
> 
> And you port it to the VMATree instead of VMADict (or whatever it's called). Then you run it and look at output. You could also take one of the stress tests that I made, remove the verification calls, and run the same stress test for VirtualMemoryTracker.

The claim makes also sense if you think about it. A binary tree will always grossly outperform a linked list for sorted insert/delete.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/18289#issuecomment-2128629567