[Developers] schedule group MoL_AllRHS at CCTK_ANALYSIS

Erik Schnetter schnetter at cct.lsu.edu
Thu Sep 7 09:21:09 CDT 2006

On Sep 7, 2006, at 08:56:43, Christian David Ott wrote:

> Hi,
> On Thu, 7 Sep 2006, Luca Baiotti wrote:
>> So, could we have a parameter to exclude the following from the  
>> schedule?
>>    schedule group MoL_AllRHS at CCTK_ANALYSIS
>>    {
>>      TRIGGERS: ADM_BSSN_sources
>>      STORAGE: ADM_BSSN_sources
>>      SYNC: ADM_BSSN_sources
>>    } "Calculate the update terms"
>>    schedule group MoL_PreStep in MoL_AllRHS
>>    {
>>    } "Prepare the update terms"
>>    schedule group MoL_CalcRHS in MoL_AllRHS after MoL_PreStep
>>    {
>>    } "Calculate the update terms"
>>    schedule group MoL_PostRHS in MoL_AllRHS after MoL_CalcRHS
>>    {
>>    } "Modify the update terms"
> why would you like to exclude it? MoL_AllRHS is never
> executed at CCTK_ANALYSIS unless you request output for the group
> ADM_BSSN_sources.


you may have looked at some Cactus timer output, and you may have  
seen numbers indicating that a lot of time is spent in the analysis  
bin calculating these update terms.  The Cactus timer output is wrong.

The problem is the following.  Assume you schedule routine A in two  
bins B1 and B2, then Cactus will keep a running total of how much  
time is spent in routine A, and will not split this into "time spent  
in A in B1" and "time spent in A in B2".  In the output, this time is  
shown twice, both for B1 and B2, giving the impression that twice as  
much time is spent in routine A as is actually the case.  If you  
compare e.g. the sum of the times of all the routines scheduled in  
the analysis bin with the time reported for the analysis bin itself,  
then you will see that they do not match up.

I have looked into correcting this problem, and it would have been  
non-trivial, so I gave up.  Cactus associates a single timer with  
every routine, but since routines may be scheduled multiple times,  
this is not sufficient.  It is not immediately clear how to  
distinguish the calls "A in B1" and "A in B2".


Erik Schnetter <schnetter at cct.lsu.edu>

My email is as private as my paper mail.  I therefore support encrypting
and signing email messages.  Get my PGP key from www.keyserver.net.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 186 bytes
Desc: This is a digitally signed message part
Url : http://www.cactuscode.org/pipermail/developers/attachments/20060907/584cbe14/attachment.bin 

More information about the Developers mailing list