Lenovo Storage S3200 with Fibre Channel
11
utilization, it is a simple matter of adding up memory usage across the servers and then
adding some overhead. For CPU utilization, it is most likely simplest to follow some
steps such as the below to determine what should be purchased:
Current CPU core GHz and quantity
Current utilization percent averaged across servers to be virtualized
Multiply current average utilization by quantity and GHz or current cores
o For ease of calculation, let’s assume we measured a 20% peak utilization
o Separate calculations for different server types or workloads may be done
and then added together
o 112 cores * 2.4GHz * .20 for utilization = 53.76GHz required
Multiply proposed CPU core GHz by core quantity in server for server total
o 3.0GHz * 20 cores (across 2 CPUs) = 60GHz
Multiply previous value on current setup by 1.20 for overhead and then divide by
servers
o 53.76GHz * 1.20 = 64.512GHz
o 64.512GHz / 60GHz per server = 2 servers required to safely run required
workload
To prepare for the event of a failure, an additional server should be available in an HA
cluster. It could be assumed that the second server in most scenarios with the above
calculation would be fine, so it is up to the discretion of the administrator familiar with the
workloads as to what his requirements are.
Network Sizing
Network sizing generally is very simple when utilizing a 10 Gbps network as it is unlikely
that that amount of bandwidth will be exceeded and even more unlikely when using the
commonly suggested method of bonding two adapter ports together to a form a 20 Gbps
trunk. If network usage is a concern, PerfMon on Windows and tools such as ntop or
cacti may be deployed on Linux.
Storage Sizing
There are two primary considerations when sizing storage in both a bare metal and a
virtualized environment. The most obvious consideration is storage capacity that has
several features related to it that can save a lot of money when purchasing drives. The
second consideration is the performance of the storage that can be the latency to reach