On Wed, 25 May 2022 at 15:09, David Rowley <dgrowle...@gmail.com> wrote: > I didn't test the performance of an aset.c context. I imagine it's > likely to be less overhead due to aset.c being generally slower from > having to jump through a few more hoops during palloc/pfree.
I've attached the results from doing the same test with a standard allocset context. With the exception of the 8 byte chunk size test, there just seems to be a 3-4% slowdown on my machine. David
alloc size aset MemoryContextLink (tps) aset master (tps) compare 8 1.138802 1.268889 89.75% 16 1.925864 1.994912 96.54% 32 3.838095 3.978061 96.48% 64 7.651818 7.897249 96.89% 128 15.013001 15.538472 96.62% 256 30.41917 31.562403 96.38% 512 60.941237 61.674914 98.81% 1024 120.757656 124.668702 96.86% 2048 236.199197 242.766508 97.29%
