Quantcast
Channel: Clusters and HPC Technology
Viewing all articles
Browse latest Browse all 927

MPI 2019's mpitune

$
0
0

Hi all

Has anybody has much luck/experience with mpitune under 2019?  It feels like a **lot** more work than the equivalent sort of activities under previous versions, so I'm wondering if I'm making it more difficult than need be?

Specific 'challenges':

1. 'msg_size' is not considered in the example/supplied tuning configurations.  I feel it should be, as that in particular affects the algorithm choice for optimal performance (e.g. ALLTOALLV switches optimal algorithm at message size of 1KB).  I guess I can manually fiddle with the mpitune generated JSON configuration, but that doesn't feel quite right, unless I'm just being lazy.

2.  I'm not quite sure I understand the supplied and downloadable (https://software.intel.com/en-us/articles/replacing-tuning-configuration...) tuning files, perhaps due to a lack of words surrounding them. Are they used by default in any shape or only when specified?

3.  'Autotuning' - linked to the above - is this intended to be used 'live' and repeatedly or should I be using it once and then capturing something from it?  Lots of config variables surrounding this.

Perhaps I'm missing some key document or reading or understanding, but any comments or thoughts would be appreciated.

~~
A


Viewing all articles
Browse latest Browse all 927

Trending Articles



<script src="https://jsc.adskeeper.com/r/s/rssing.com.1596347.js" async> </script>