New and Updated Documentation for HLRE
2007/05/31
A new guide from NEC with index in convenient PDF format. Login required
2006/09/27
A new guide from NEC with index and in convenient PDF format. Login required
2006/07/17
The first result of many discussions after our
presentation at this year's
NUG meeting is: if you
got problems scaling your code from one to eight CPUs you
might want to run your program with F_PROGINF=DETAIL2. If the
output then shows something like
Memory Network (sec) : 10.874843
while user time is
User Time (sec) : 15.487728
you got a big problem because you spend 70% of your time waiting
for memory which kills scaling very efficiently. The difference
to bank conflict time (which is low)
Bank (sec) : 0.004435
is that one task waits for banks currently used by
other
tasks. In the talk we tried to explain - among other things - under
which circumstances memory congestion occurs and what can be done to
avoid it.
2006/02/06
Concept for a new project-orientated data management at DKRZ (as of January 2006)
2005/06/24
A new guide from NEC in convenient PDF format. Login required
2005/01/14
qflops now also shows efficiency information for multi-node jobs.
2004/08/19
All Fortran/C compiler switches and much more in one handy PDF. Download
directly fom NEC.
2004/07/06
You know qstat, but there are more tools which help you keeping an eye on your jobs.
Read more...
2004/06/08
Our frontend server cross.dkrz.de can be used for more things than just cross compiling
your models. There are some points to keep in mind, though.
Read more ... (German)
2004/04/08
Tired of waiting for your job to finish in the SERI queue? Our new
step by step guide to Parallelizing Sequential Code with MPI shows you how to use up to
128 CPUs of HLRE for your job
and decrease turn around times dramatically. English version by
Jana Meyer.
Read more ...