Capacity Management: Guided Practitioner Satnav - When will I get there? Categorization (13 of 17)

Previously we looked at the status of capacity management within the Wintel virtual infrastructure (VI) farm at a large enterprise and will now take a look at the categorization process.

The initial phase of the migration has been designed and managed using the Capacity Planner service provided by VMware.  Whilst this does an admirable job of calculating the physical requirements for virtualizing this list of potential candidates it doesn’t provide any categorisation for the future allocation of these servers.

Being able to categorise the servers prior to virtualisation will provide:

•          Additional verification on the sizing of the infrastructure

•          Categorized existing servers to provide a valuable feed into Capacity Planning by attaching a resource requirement to an existing workload to create a frame of reference for future server requirements.

The resulting categorization should be used in the power classification used for specifying the VM or resource pool, failover requirements etc.  The process provides a basic method for categorizing migration candidates and how much capacity they will use.

The first stage is determine the physical aspects of the categorization process and to calculate the normalised power rating for the physical server to provide a rating that reflects its usage comparable to the rest of the estate. The second stage is to determine how this will relate to the virtual infrastructure.  This is done by calculating the power rating of the ESX server, determining the P2V ratio i.e. how many virtual machines you plan to run per host and then using this information, calculate the various categories.  In this case the server has a SpecInt of 40 with a peak utilization of 15% and hence a rating of 6.

Categorization example

Physical:         the original server has a peak CPU utilisation of 15% and a SpecInt rating of 40, so the normalised power rating of the original physical server is derived by

N = S x (U/100) = 40 x (10/100) = 6

Virtual:             the new server to support the total VI has a SpecInt value of say 200 and the plan is to run with a P2V ratio of say 1:20 and the tiers to be defined are bronze, silver, gold and platinum:

Bronze             = 200 / 20                    = 10

Silver               = Bronze x 2   = 20

Gold                = Silver x 2                  = 40

Platinum          = uncapped

Thus the initial physical server above would be rated as a bronze in P2V with associated priorities, but could be promoted in the light of any business priority or user reported problems to a higher grade. Or to put it a different way, with these definitions, it is possible to virtualise 20 bronze servers, 10 silver servers or 5 gold servers per ESX host.  Moving forward we have an approximate way of determining the virtual requirements of a future migration candidate or a new server by comparing it with an existing installation. Using this process would mean that instead of having to say to expect 100 projects of varying sizes, you may be in the position to actually put a more specific number to these predictions.

I’ll be looking at medium term improvements that can be expected on Wednesday…

Adam Grummitt
Distinguished Engineer