Skip to content

use tcmalloc in HLT2 throughput test

Christoph Hasse requested to merge chasse_alloc into master

I did a quick comparison running hlt2_pp_thor and hlt1_pp_default with:

  1. gaudirun.py -T which uses tcmalloc
  2. jemalloc from cvmfs
  3. mimalloc

For hlt2, we see a good but pretty much identical speedup for all three. For hlt1, jemalloc is terrible (2x slower), tcmalloc makes it slightly worse mimalloc is pretty much same speed as usual.

So for hlt1 none of them make sense and for hl2 there isn't really any difference which is great because tcmalloc is already available in gaudirun.py so we only need to activate it.

cc @rmatev @sstahl

Merge request reports