# Huge performance fluctuations in parallel benchmark: insights?

**URL:** <https://discourse.julialang.org/t/huge-performance-fluctuations-in-parallel-benchmark-insights/67953>\
**Category:** Performance\
**Tags:** parallel, multithreading, benchmarktools\
**Created:** [September 9, 2021, 6:58pm UTC](https://discourse.julialang.org/t/huge-performance-fluctuations-in-parallel-benchmark-insights/67953 "2021-09-09T18:58:02Z")\
**Posts on this page:** 1\
**Showing post:** 14

<div class="post-metadata">

**Author:** ![carstenbauer](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/carstenbauer/32/4981_2.png) [@carstenbauer](https://discourse.julialang.org/u/carstenbauer)\
**Post date:** [September 10, 2021, 2:12pm UTC](https://discourse.julialang.org/t/huge-performance-fluctuations-in-parallel-benchmark-insights/67953/14 "2021-09-10T14:12:13Z")

</div>

> [@johnh](#):
>
> Even better use hwloc / lstopo
> 
> [Portable Hardware Locality (hwloc)](https://www.open-mpi.org/projects/hwloc/lstopo/)  
> For linux yum install hwloc or apt-get install hwloc

Note that we have [Hwloc.jl](https://github.com/JuliaParallel/Hwloc.jl), which also downloads the library via [Hwloc\_jll.jl](https://github.com/JuliaBinaryWrappers/Hwloc_jll.jl).

> [@johnh](#):
>
> If this is a physical machine which you have access to have you tried disabling hyperthreading in the BIOS?

Depending on the details of the machine, just setting `JULIA_EXCLUSIVE=1` might do the job (i.e. pin julia threads to physical cpus). Otherwise, `numactl` and `likwid-pin` could be options, see [this discussion](https://discourse.julialang.org/t/thread-affinitization-pinning-julia-threads-to-cores/58069/5) for more information.

---

_[View the full topic](https://discourse.julialang.org/t/huge-performance-fluctuations-in-parallel-benchmark-insights/67953)._
