vSphere Cluster calculator


Welcome to the latest version of my cluster calculator!

You can type in the field of the online table and the online table will calculate how many hosts you need. You can make the calculation for 2 different setups side by side.

Take note that all values in black are VARIABLES (About 20 variables can be adjusted). Changing any of them will change the recommendations in the top and throughout the whole table. This is a work in progress though, potentially storage recommendations will be made as well. Check regularly to see new updates.

All the values, percentages, consolidation ratios currently in the table are examples. These are not the values I would tell you to use.

TPS

NEW: I have added the possibility to add estimated TPS savings to the table. I know, TPS has been taken out of the equation since the latest updates, but some people still enable it because it won’t affect security in their case, or and the memory savings are more important to them.

Here is how the TPS calculation works. For every year I calculate the total amount of RAM needed based on the % of growth for each year. After I have calculated the growth (in the below example, for 3 years) only then I calculate the TPS savings. Take note that what ever memory is used for NFV, has no TPS savings in my calculations. In this case field H116.

=((((((H96+H101+H106+H111)*H117)*H121)*H123)*(100%-H124)))+H116
=((((((Memory 1 +Memory 2 +Memory 3 +Memory 4)* % growth year 0 )* % growth year 1)*% growth year 2)*(100%-TPS savings in %)))+ Memory 100% NFV

DIFFERENT TYPE OF VMs

In this updated version I have added the possibility to add different types of VM’s. So why did I provide this option?

Well for one, if you have a moderate load of General Purpose servers, and a load of Database Servers, the Database Servers’ impact on the CPUs will be higher. Also the vCPU to pCPU ratio should be lower for the database servers than for the General Purpose servers. You do not want a higher amount of ReadyTime for database servers. In fact you should keep this as low as possible..

I have added the possibility to see how many hosts you need per cluster while maintaining N+x redundancy for all clusters ( equally sized Clusters!!!!). The more clusters, the more hosts you will need to be able to guarantee the N+ x redundancy for all clusters. N is defined in Line 77.

Secondly, you might already have some servers and want to add additional servers and see what the effect on your hosts and clusters will be.

Thirdly, Telco virtualization really has a different CPU and RAM usage. So I have given them their own slot. It is important to understand that TELCO requires you to pin the vCPU to physical CPUs. Read: you should not share the physical CPUs which will be used for Telco servers. In the spreadsheet you will see that some values are in red. I urge you not to change these.

On the assumption that  Telco servers for most companies will be ‘one time deployments’ only, they are left out of the equations to calculate growth for all years.

However if you are a service provider and adding Telco  appliances to your clusters is something you do have to do on a ‘regular’ basis, all you need to do is to bring the value of the Consolidated Total Peak for CPU within the (()) and multiply it all by the growth % for each Year’s line. You will have to do the same for RAM as well.

When I think about explaining how this table deals with calculating CPU usage per server type, the first word which comes to mind here is ‘relative‘. All my calculations, or rather the calculations done in this xls table are performed in relation to a servers’ need to have quicker access to MHz or slots on the cores. The more servers have to share CPU cycles on the same cores, the longer they potentially have to wait until free CPU cycles are available again.

If you do not need all the different types of servers, set the black values to 0 and bear in mind that where a % is listed, the % symbol should remain in the field. so a 0% should be typed as 0%

As with the first calculator, this calculator is mostly build on speculation and assumptions. Please keep this in mind. We try to mimic the real workload so we can make a more or less calculated guess on how many hosts and clusters we will need to service our future workloads.

Sincere thanks to Ariel Sanchez, Ahmed Ragab and Richard Diaz for their willingness to review the first version of this spreadsheet.

Recommended reads:

Sean Crookston, Harley Stagner : Managing and Optimizing VMware vSphere Deployments

Best practices and advanced features for VMware High Availability

Also have a look at Josh Odgers wonderfull online cluster calculator

Kim