Microsoft HPC 2012 R2 Update 3 is something new to us and we recently discovered that having 2700 computes nodes on one head node seems problematic at a point where schedule task on the server can't complete in the appropriate amount of time.
The computes nodes are CAD Workstations and the Head Node is a virtual machine with Intel Xeon E5-2680 Quad Core of 2.7 GHz with 16 GB of RAM.
A test was done by adding 2 more CPUs but even if the CPU usage is around 50 to 70% most of the time the Schedule Task never complete successfully. The cluster size was reduce to 500 and now the system is working but a lot of CPU is required and also
a lot of RAM.
What would be required to be able to use our 2700 compute nodes on the same head node ? How can the system be sized properly to work with all available compute nodes ?
