QuantumATK Forum
QuantumATK => Installation and License Questions => Topic started by: dhurba on June 5, 2012, 07:41
-
I am running ATK in a 8 core server ..now three other server has come to my lab all are 8 cores ie total of 32 cores...
my license are
[2011-12-19 16:43:10] atksemaster (v11.Cool (1 license(s)) license type: exclusive
[2011-12-19 16:43:10] atkseslave (v11.Cool (3 license(s)) license type: exclusive
[2011-12-19 16:43:10] vnl (v11.Cool (1 license(s)) license type: exclusive
Is it possible to use all the four servers to full potential. what is core based license and workstation based license
doesn't 3 slave license can accommodate 3 servers ..or other extra license is required..
-
3 slaves is a perfect license to run in parallel on this setup.
-
3 slaves is a perfect license to run in parallel on this setup.
someone told that 1 slave is needed for 1 core.
ie 4 servers*8 core=32 core =32 slaves :(
-
Slave licenses have nothing to do with cores, in principle.
Slave licenses are used by ATK to submit multiple MPI processes. Having N slave licenses allows you to run an calculation parallelized over N+1 (the master is also used) processes, using a command like "mpiexec -n N+1". If you choose to run all these processes on the same computer (using the cores as processors), or on different computers, is up to you, and is not dependent on the licenses you have.
If you submit the N+1 processors on different machines, ATK can additionally thread over the cores, but you don't need any extra licenses for that. This offers the best performance possible with ATK; running more MPI processes than sockets on the same machine is not advisable.
In the future ATK will become memory-parallel, and that will change the picture a bit, and make running multiple MPI processes on the same computer more efficient.
-
In general, how do we decide how many slave licenses are needed per master? We have quantumwise 13.8.1 in our lab and are upgrading to the latest version.
-
When you discuss the upgrade with your account manager at Synopsys, you will find that our packaging has changed since ver 13.8, such that this question is not so relevant anymore from the price perspective.
But anyway, the calculations typically scale to 100 per master or more, with an optimum around 40-60 (although it's extremely dependent on the type of system you run).