[Hadstruct] Next meeting...

[David Richards]

Dear All,

Raza reminded me that I hadn’t circulated the minutes, just posted them to the GitHub page, so I’ve appended them below.  Most importantly, we were to have our next meeting at 11am Friday.  

One thing we discussed was ensuring that we are aware, and are using, our allocations.  PLANNING FOR THIS SHOULD BE THE MAJOR FRIDAY AGENDA ITEM.

I have created a google sheet, which you will find on the Hadstruc wiki at https://github.com/JeffersonLab/HadStruc/wiki/Computer-Allocations-and-Usage.  I’ve entered the LRAC (frontera), Xsede and USQCD allocations, but there are several others out there that I need to fill in, notably those in France, and the NERSC/INCITE allocations.  Please would those on the front-iine of those allocations either send me the information, or fill in the google sheet yourself.

Note that the google sheet is unprotected, and can be edited by anyone who has the link  - I didn’t want to force everyone to have a google account to gain access, but I need to find a better solution to this.  

Regards,
David





******************************************************************************************************

Smearing

ER has tracked down the issue that was affecting the smearing - on the 48^3 lattices, unsmeared links had gone into the smearing algorithm for the unphased eigenvectors; for the phased runs, the eigenvectors now include smearing. On the 32^3 lattices, where there is a need to continue the ensemble, there was a realization (or lack of realization by DR) that the correct smearing parameter was 0.1 rather than 0.08.

Computational strategy

There was discussion regarding the strategy for using GPU nodes, and whether the GPU and CPU could both be used, and how to make use of multi-node efficiently. Frank will build the new Genprop4/hybrid code for Frontera.

Allocations

We have a new Xsede allocation announced for Bridges and Stampede. Keeping track of all this is getting a little complicated, so DR will add an allocation page to the Wiki.

Factorization of Quasi-PDF/pseudo-PDF

JQ has circulated a summary, and these will be posted to the Wiki.

Spatial anisotropic

Anisotropy in two directions was considerably more complicated, and code development in the case of the HMC. Work by FNAL, Dublin Group? However, the work with only one anisotropic z direction would still give useful information, namely a broader range of Ioffe time and smaller values of the separation z.

Next Meeting

Friday 26th Sept, 11am EST.