Colocation to Virtualization

The replacements have arrived.  Just got in three 3850 M2 servers specs are as follows per server:

- four quad core 2.9ghz Xeons
- 64 gigs of RAM
- two 73 gig internal drives.  Configured raid 1

Don’t let the two 73 gig drives fool you.  We only use em for the ESX install.  All storage is done either over 16 gig fiber connects to an EMC (dual 4gig fiber cards load balancing) or via 2 gigabit ISCSI connections.  These servers are more modular than the lower class of IBM servers.  The plan is to upgrade each server from 64 to 128 gigs of RAM next year.  The servers also have two hot-swappable pci-x slots and embedded DRAM on the memory controller.  This DRAM holds a “snoop filter lookup table” that routes cache requests to the correct microprocessor bus instead of sending the requests to all processors.  After all configuration and setup was done, I loaded 50 Windows and Linux mix VMs to one server.  I was able to get the RAM utilization up to 53 gigs and the CPUs were resting at around 23% utilization!  Just goes to show that the real bottle neck of virtualization is RAM.