[Users] Warning: Synchronising group in local mode

[Erik Schnetter]

Vassilios

You cannot synchronize in local mode. Different processes may execute
local-mode functions different number of times, which would lead to a
deadlock. For example, if process 0 holds 1 large component, while process
1 holds 2 smaller components, then process 0 would synchronize once, while
process 1 would synchronize twice. Since synchronization is a global
operation for which all processes need to participate, this would fail.

Also -- what should it mean for a process to synchronize in this case?
Synchronization fills ghost points with valid data from other components.
If a process holds 2 components, and you synchronize after calculating the
first component, then it is not possible to obtain valid data for points
from the second component, since this component has not been calculated yet.

Synchronization in local mode works in certain very restricted cases to
ensure backward compatibility with PUGH. I believe it works if each process
holds exactly 1 component.

-erik


On Tue, Feb 20, 2018 at 5:22 PM, Vassilios Mewes <vxmsma at rit.edu> wrote:

> Hello all,
>
> I am getting the following warnings on BlueWaters when running the ET in a
> unigrid run and calling
> "CCTK_SyncGroup" from within a function that is scheduled as local:
>
> ^[[1mWARNING[L2,P0] (Carpet):^[[0m Synchronising group "ADMBASE::METRIC"
> in local mode
>
> and so on for every group that is synced in these local functions.
>
> I haven't seen this warning on bluesky (local RIT cluster) or my MacBook.
>
> Is it disallowed to sync in local mode?
>
> Best wishes,
>
> Vassili
>
>
>
>
> _______________________________________________
> Users mailing list
> Users at einsteintoolkit.org
> http://lists.einsteintoolkit.org/mailman/listinfo/users
>
>


-- 
Erik Schnetter <schnetter at cct.lsu.edu>
http://www.perimeterinstitute.ca/personal/eschnetter/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.einsteintoolkit.org/pipermail/users/attachments/20180221/21a5a1cd/attachment.html