[cadynce] Additional GT4 simulations completed, results summarized..

[Gautam Thaker]

The link below shows the summary of my results for GT4.

https://repo.isis.vanderbilt.edu/twiki/bin/view/CADYNCE/OneFiftyFiftyCase1#Summary_of_All_the_Runs

It is curious to note that forcing all "critical path" processing to be highest priority 
does not give us much different results than "equal priorities". I have verified that 
preemptive, priority based scheduling is properly modeled. I have to look more into this 
issue, but I suspect results may hold and are prob. due to fact that "critical path" 
processing of all 10 strings simply creates enough interference to make these larger end 
to end times so.

I wonder if we can do an allocation where all 60 processes of 10 critical strings (each 
having 6 processes) can be spread in such a manner on 50 blades that no more than one 
critical process is on one blade (which it shares w/ other non-critical process) for 40 of 
them and 10 of them do share. The shared ones can be the ones that somehow would have 
least impact.

Gaurav, can we try to find such a 50 blade allocation? I would like to run this case as 
"special, critical path sensitive allocation".

Gautam